Improve loading ckpt

* Correct controlnet out of list error

* Apply suggestions from code review

* correct tests

* correct tests

* fix

* test all

* Apply suggestions from code review

* test all

* test all

* Apply suggestions from code review

* Apply suggestions from code review

* fix more tests

* Fix more

* Apply suggestions from code review

* finish

* Apply suggestions from code review

* Update src/diffusers/schedulers/scheduling_k_dpm_2_ancestral_discrete.py

* finish

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -49,3 +49,32 @@ body:
       placeholder: diffusers version, platform, python version, ...
     validations:
       required: true
+  - type: textarea
+    id: who-can-help
+    attributes:
+      label: Who can help?
+      description: |
+        Your issue will be replied to more quickly if you can figure out the right person to tag with @
+        If you know how to use git blame, that is the easiest way, otherwise, here is a rough guide of **who to tag**.
+        
+        All issues are read by one of the core maintainers, so if you don't know who to tag, just leave this blank and
+        a core maintainer will ping the right person.
+        
+        Please tag fewer than 3 people.
+        
+        General library related questions: @patrickvonplaten and @sayakpaul
+
+        Questions on the training examples: @williamberman, @sayakpaul, @yiyixuxu
+
+        Questions on memory optimizations, LoRA, float16, etc.: @williamberman, @patrickvonplaten, and @sayakpaul
+
+        Questions on schedulers: @patrickvonplaten and @williamberman
+
+        Questions on models and pipelines: @patrickvonplaten, @sayakpaul, and @williamberman
+
+        Questions on JAX- and MPS-related things: @pcuenca
+
+        Questions on audio pipelines: @patrickvonplaten, @kashif, and @sanchit-gandhi 
+        
+        Documentation: @stevhliu and @yiyixuxu
+      placeholder: "@Username ..."

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,60 @@
+# What does this PR do?
+
+<!--
+Congratulations! You've made it this far! You're not quite done yet though.
+
+Once merged, your PR is going to appear in the release notes with the title you set, so make sure it's a great title that fully reflects the extent of your awesome contribution.
+
+Then, please replace this with a description of the change and which issue is fixed (if applicable). Please also include relevant motivation and context. List any dependencies (if any) that are required for this change.
+
+Once you're done, someone will review your PR shortly (see the section "Who can review?" below to tag some potential reviewers). They may suggest changes to make the code even better. If no one reviewed your PR after a week has passed, don't hesitate to post a new comment @-mentioning the same persons---sometimes notifications get lost.
+-->
+
+<!-- Remove if not applicable -->
+
+Fixes # (issue)
+
+
+## Before submitting
+- [ ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
+- [ ] Did you read the [contributor guideline](https://github.com/huggingface/diffusers/blob/main/CONTRIBUTING.md)?
+- [ ] Did you read our [philosophy doc](https://github.com/huggingface/diffusers/blob/main/PHILOSOPHY.md) (important for complex PRs)?
+- [ ] Was this discussed/approved via a Github issue or the [forum](https://discuss.huggingface.co/)? Please add a link to it if that's the case.
+- [ ] Did you make sure to update the documentation with your changes? Here are the
+      [documentation guidelines](https://github.com/huggingface/diffusers/tree/main/docs), and
+      [here are tips on formatting docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation).
+- [ ] Did you write any new necessary tests?
+
+
+## Who can review?
+
+Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
+members/contributors who may be interested in your PR.
+
+<!-- Your PR will be replied to more quickly if you can figure out the right person to tag with @
+
+ If you know how to use git blame, that is the easiest way, otherwise, here is a rough guide of **who to tag**.
+ Please tag fewer than 3 people.
+
+Core library:
+
+- Schedulers: @williamberman and @patrickvonplaten
+- Pipelines:  @patrickvonplaten and @sayakpaul
+- Training examples: @sayakpaul and @patrickvonplaten
+- Docs: @stevenliu and @yiyixu
+- JAX and MPS: @pcuenca
+- Audio: @sanchit-gandhi
+- General functionalities: @patrickvonplaten and @sayakpaul
+
+Integrations:
+
+- deepspeed: HF Trainer/Accelerate: @pacman100
+
+HF projects:
+
+- accelerate: [different repo](https://github.com/huggingface/accelerate)
+- datasets: [different repo](https://github.com/huggingface/datasets)
+- transformers: [different repo](https://github.com/huggingface/transformers)
+- safetensors: [different repo](https://github.com/huggingface/safetensors)
+
+-->

.github/workflows/build_documentation.yml

@@ -5,6 +5,7 @@ on:
     branches:
       - main
       - doc-builder*
+      - v*-release
       - v*-patch
 
 jobs:
@@ -14,6 +15,7 @@ jobs:
       commit_sha: ${{ github.sha }}
       package: diffusers
       notebook_folder: diffusers_doc
-      languages: en ko
+      languages: en ko zh
     secrets:
       token: ${{ secrets.HUGGINGFACE_PUSH }}
+      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}

.github/workflows/delete_doc_comment.yml

@@ -1,13 +1,14 @@
-name: Delete dev documentation
+name: Delete doc comment
 
 on:
-  pull_request:
-    types: [ closed ]
+  workflow_run:
+    workflows: ["Delete doc comment trigger"]
+    types:
+      - completed
 
 
 jobs:
   delete:
     uses: huggingface/doc-builder/.github/workflows/delete_doc_comment.yml@main
-    with:
-      pr_number: ${{ github.event.number }}
-      package: diffusers
+    secrets:
+      comment_bot_token: ${{ secrets.COMMENT_BOT_TOKEN }}

.github/workflows/delete_doc_comment_trigger.yml

@@ -0,0 +1,12 @@
+name: Delete doc comment trigger
+
+on:
+  pull_request:
+    types: [ closed ]
+
+
+jobs:
+  delete:
+    uses: huggingface/doc-builder/.github/workflows/delete_doc_comment_trigger.yml@main
+    with:
+      pr_number: ${{ github.event.number }}

.github/workflows/pr_dependency_test.yml

@@ -0,0 +1,32 @@
+name: Run dependency tests
+
+on:
+  pull_request:
+    branches:
+      - main
+  push:
+    branches:
+      - main
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  check_dependencies:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: "3.7"
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e .
+          pip install pytest
+      - name: Check for soft dependencies
+        run: |
+          pytest tests/others/test_dependencies.py
+      

.github/workflows/pr_tests.yml

@@ -81,7 +81,7 @@ jobs:
       if: ${{ matrix.config.framework == 'pytorch_models' }}
       run: |
         python -m pytest -n 2 --max-worker-restart=0 --dist=loadfile \
-          -s -v -k "not Flax and not Onnx" \
+          -s -v -k "not Flax and not Onnx and not Dependency" \
           --make-reports=tests_${{ matrix.config.report }} \
           tests/models tests/schedulers tests/others
 

.github/workflows/push_tests.yml

@@ -17,6 +17,7 @@ jobs:
   run_slow_tests:
     strategy:
       fail-fast: false
+      max-parallel: 1
       matrix:
         config:
           - name: Slow PyTorch CUDA tests on Ubuntu

.github/workflows/upload_pr_documentation.yml

@@ -0,0 +1,16 @@
+name: Upload PR Documentation
+
+on:
+  workflow_run:
+    workflows: ["Build PR Documentation"]
+    types:
+      - completed
+
+jobs:
+  build:
+    uses: huggingface/doc-builder/.github/workflows/upload_pr_documentation.yml@main
+    with:
+      package_name: diffusers
+    secrets:
+      hf_token: ${{ secrets.HF_DOC_BUILD_PUSH }}
+      comment_bot_token: ${{ secrets.COMMENT_BOT_TOKEN }}

CONTRIBUTING.md

@@ -125,14 +125,14 @@ Awesome! Tell us what problem it solved for you.
 
 You can open a feature request [here](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feature_request.md&title=).
 
-#### 2.3 Feedback. 
+#### 2.3 Feedback.
 
 Feedback about the library design and why it is good or not good helps the core maintainers immensely to build a user-friendly library. To understand the philosophy behind the current design philosophy, please have a look [here](https://huggingface.co/docs/diffusers/conceptual/philosophy). If you feel like a certain design choice does not fit with the current design philosophy, please explain why and how it should be changed. If a certain design choice follows the design philosophy too much, hence restricting use cases, explain why and how it should be changed.
 If a certain design choice is very useful for you, please also leave a note as this is great feedback for future design decisions.
 
 You can open an issue about feedback [here](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=).
 
-#### 2.4 Technical questions. 
+#### 2.4 Technical questions.
 
 Technical questions are mainly about why certain code of the library was written in a certain way, or what a certain part of the code does. Please make sure to link to the code in question and please provide detail on
 why this part of the code is difficult to understand.
@@ -394,8 +394,8 @@ passes. You should run the tests impacted by your changes like this:
  ```bash
  $ pytest tests/<TEST_TO_RUN>.py
  ```
- 
-Before you run the tests, please make sure you install the dependencies required for testing. You can do so 
+
+Before you run the tests, please make sure you install the dependencies required for testing. You can do so
 with this command:
 
  ```bash

PHILOSOPHY.md

@@ -27,18 +27,18 @@ In a nutshell, Diffusers is built to be a natural extension of PyTorch. Therefor
 
 ## Simple over easy
 
-As PyTorch states, **explicit is better than implicit** and **simple is better than complex**. This design philosophy is reflected in multiple parts of the library: 
+As PyTorch states, **explicit is better than implicit** and **simple is better than complex**. This design philosophy is reflected in multiple parts of the library:
 - We follow PyTorch's API with methods like [`DiffusionPipeline.to`](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.to) to let the user handle device management.
 - Raising concise error messages is preferred to silently correct erroneous input. Diffusers aims at teaching the user, rather than making the library as easy to use as possible.
 - Complex model vs. scheduler logic is exposed instead of magically handled inside. Schedulers/Samplers are separated from diffusion models with minimal dependencies on each other. This forces the user to write the unrolled denoising loop. However, the separation allows for easier debugging and gives the user more control over adapting the denoising process or switching out diffusion models or schedulers.
-- Separately trained components of the diffusion pipeline, *e.g.* the text encoder, the unet, and the variational autoencoder, each have their own model class. This forces the user to handle the interaction between the different model components, and the serialization format separates the model components into different files. However, this allows for easier debugging and customization. Dreambooth or textual inversion training 
+- Separately trained components of the diffusion pipeline, *e.g.* the text encoder, the unet, and the variational autoencoder, each have their own model class. This forces the user to handle the interaction between the different model components, and the serialization format separates the model components into different files. However, this allows for easier debugging and customization. Dreambooth or textual inversion training
 is very simple thanks to diffusers' ability to separate single components of the diffusion pipeline.
 
 ## Tweakable, contributor-friendly over abstraction
 
-For large parts of the library, Diffusers adopts an important design principle of the [Transformers library](https://github.com/huggingface/transformers), which is to prefer copy-pasted code over hasty abstractions. This design principle is very opinionated and stands in stark contrast to popular design principles such as [Don't repeat yourself (DRY)](https://en.wikipedia.org/wiki/Don%27t_repeat_yourself). 
+For large parts of the library, Diffusers adopts an important design principle of the [Transformers library](https://github.com/huggingface/transformers), which is to prefer copy-pasted code over hasty abstractions. This design principle is very opinionated and stands in stark contrast to popular design principles such as [Don't repeat yourself (DRY)](https://en.wikipedia.org/wiki/Don%27t_repeat_yourself).
 In short, just like Transformers does for modeling files, diffusers prefers to keep an extremely low level of abstraction and very self-contained code for pipelines and schedulers.
-Functions, long code blocks, and even classes can be copied across multiple files which at first can look like a bad, sloppy design choice that makes the library unmaintainable. 
+Functions, long code blocks, and even classes can be copied across multiple files which at first can look like a bad, sloppy design choice that makes the library unmaintainable.
 **However**, this design has proven to be extremely successful for Transformers and makes a lot of sense for community-driven, open-source machine learning libraries because:
 - Machine Learning is an extremely fast-moving field in which paradigms, model architectures, and algorithms are changing rapidly, which therefore makes it very difficult to define long-lasting code abstractions.
 - Machine Learning practitioners like to be able to quickly tweak existing code for ideation and research and therefore prefer self-contained code over one that contains many abstractions.
@@ -47,10 +47,10 @@ Functions, long code blocks, and even classes can be copied across multiple file
 At Hugging Face, we call this design the **single-file policy** which means that almost all of the code of a certain class should be written in a single, self-contained file. To read more about the philosophy, you can have a look
 at [this blog post](https://huggingface.co/blog/transformers-design-philosophy).
 
-In diffusers, we follow this philosophy for both pipelines and schedulers, but only partly for diffusion models. The reason we don't follow this design fully for diffusion models is because almost all diffusion pipelines, such 
+In diffusers, we follow this philosophy for both pipelines and schedulers, but only partly for diffusion models. The reason we don't follow this design fully for diffusion models is because almost all diffusion pipelines, such
 as [DDPM](https://huggingface.co/docs/diffusers/v0.12.0/en/api/pipelines/ddpm), [Stable Diffusion](https://huggingface.co/docs/diffusers/v0.12.0/en/api/pipelines/stable_diffusion/overview#stable-diffusion-pipelines), [UnCLIP (Dalle-2)](https://huggingface.co/docs/diffusers/v0.12.0/en/api/pipelines/unclip#overview) and [Imagen](https://imagen.research.google/) all rely on the same diffusion model, the [UNet](https://huggingface.co/docs/diffusers/api/models#diffusers.UNet2DConditionModel).
 
-Great, now you should have generally understood why 🧨 Diffusers is designed the way it is 🤗. 
+Great, now you should have generally understood why 🧨 Diffusers is designed the way it is 🤗.
 We try to apply these design principles consistently across the library. Nevertheless, there are some minor exceptions to the philosophy or some unlucky design choices. If you have feedback regarding the design, we would ❤️  to hear it [directly on GitHub](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=).
 
 ## Design Philosophy in Details
@@ -89,7 +89,7 @@ The following design principles are followed:
 - Models should by default have the highest precision and lowest performance setting.
 - To integrate new model checkpoints whose general architecture can be classified as an architecture that already exists in Diffusers, the existing model architecture shall be adapted to make it work with the new checkpoint. One should only create a new file if the model architecture is fundamentally different.
 - Models should be designed to be easily extendable to future changes. This can be achieved by limiting public function arguments, configuration arguments, and "foreseeing" future changes, *e.g.* it is usually better to add `string` "...type" arguments that can easily be extended to new future types instead of boolean `is_..._type` arguments. Only the minimum amount of changes shall be made to existing architectures to make a new model checkpoint work.
-- The model design is a difficult trade-off between keeping code readable and concise and supporting many model checkpoints. For most parts of the modeling code, classes shall be adapted for new model checkpoints, while there are some exceptions where it is preferred to add new classes to make sure the code is kept concise and 
+- The model design is a difficult trade-off between keeping code readable and concise and supporting many model checkpoints. For most parts of the modeling code, classes shall be adapted for new model checkpoints, while there are some exceptions where it is preferred to add new classes to make sure the code is kept concise and
 readable longterm, such as [UNet blocks](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py) and [Attention processors](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
 
 ### Schedulers
@@ -97,9 +97,9 @@ readable longterm, such as [UNet blocks](https://github.com/huggingface/diffuser
 Schedulers are responsible to guide the denoising process for inference as well as to define a noise schedule for training. They are designed as individual classes with loadable configuration files and strongly follow the **single-file policy**.
 
 The following design principles are followed:
-- All schedulers are found in [`src/diffusers/schedulers`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers). 
-- Schedulers are **not** allowed to import from large utils files and shall be kept very self-contained. 
-- One scheduler python file corresponds to one scheduler algorithm (as might be defined in a paper). 
+- All schedulers are found in [`src/diffusers/schedulers`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers).
+- Schedulers are **not** allowed to import from large utils files and shall be kept very self-contained.
+- One scheduler python file corresponds to one scheduler algorithm (as might be defined in a paper).
 - If schedulers share similar functionalities, we can make use of the `#Copied from` mechanism.
 - Schedulers all inherit from `SchedulerMixin` and `ConfigMixin`.
 - Schedulers can be easily swapped out with the [`ConfigMixin.from_config`](https://huggingface.co/docs/diffusers/main/en/api/configuration#diffusers.ConfigMixin.from_config) method as explained in detail [here](./using-diffusers/schedulers.mdx).

README.md

@@ -1,6 +1,6 @@
 <p align="center">
     <br>
-    <img src="./docs/source/en/imgs/diffusers_library.jpg" width="400"/>
+    <img src="https://github.com/huggingface/diffusers/blob/main/docs/source/en/imgs/diffusers_library.jpg" width="400"/>
     <br>
 <p>
 <p align="center">
@@ -25,12 +25,12 @@
 
 ## Installation
 
-We recommend installing 🤗 Diffusers in a virtual environment from PyPi or Conda. For more details about installing [PyTorch](https://pytorch.org/get-started/locally/) and [Flax](https://flax.readthedocs.io/en/latest/installation.html), please refer to their official documentation.
+We recommend installing 🤗 Diffusers in a virtual environment from PyPi or Conda. For more details about installing [PyTorch](https://pytorch.org/get-started/locally/) and [Flax](https://flax.readthedocs.io/en/latest/#installation), please refer to their official documentation.
 
 ### PyTorch
 
 With `pip` (official package):
-    
+
 ```bash
 pip install --upgrade diffusers[torch]
 ```
@@ -107,7 +107,7 @@ Check out the [Quickstart](https://huggingface.co/docs/diffusers/quicktour) to l
 | [Training](https://huggingface.co/docs/diffusers/training/overview) | Guides for how to train a diffusion model for different tasks with different training techniques.                                                                                               |
 ## Contribution
 
-We ❤️  contributions from the open-source community! 
+We ❤️  contributions from the open-source community!
 If you want to contribute to this library, please check out our [Contribution guide](https://github.com/huggingface/diffusers/blob/main/CONTRIBUTING.md).
 You can look out for [issues](https://github.com/huggingface/diffusers/issues) you'd like to tackle to contribute to the library.
 - See [Good first issues](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22) for general opportunities to contribute
@@ -128,70 +128,70 @@ just hang out ☕.
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Unconditional Image Generation</td>
-    <td><a href="./api/pipelines/ddpm"> DDPM </a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/ddpm"> DDPM </a></td>
     <td><a href="https://huggingface.co/google/ddpm-ema-church-256"> google/ddpm-ema-church-256 </a></td>
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Text-to-Image</td>
-    <td><a href="./api/pipelines/stable_diffusion/text2img">Stable Diffusion Text-to-Image</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/text2img">Stable Diffusion Text-to-Image</a></td>
       <td><a href="https://huggingface.co/runwayml/stable-diffusion-v1-5"> runwayml/stable-diffusion-v1-5 </a></td>
   </tr>
   <tr>
     <td>Text-to-Image</td>
-    <td><a href="./api/pipelines/unclip">unclip</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/unclip">unclip</a></td>
       <td><a href="https://huggingface.co/kakaobrain/karlo-v1-alpha"> kakaobrain/karlo-v1-alpha </a></td>
   </tr>
   <tr>
     <td>Text-to-Image</td>
-    <td><a href="./api/pipelines/if">if</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/if">if</a></td>
       <td><a href="https://huggingface.co/DeepFloyd/IF-I-XL-v1.0"> DeepFloyd/IF-I-XL-v1.0 </a></td>
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Text-guided Image-to-Image</td>
-    <td><a href="./api/pipelines/stable_diffusion/controlnet">Controlnet</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/controlnet">Controlnet</a></td>
       <td><a href="https://huggingface.co/lllyasviel/sd-controlnet-canny"> lllyasviel/sd-controlnet-canny </a></td>
   </tr>
   <tr>
     <td>Text-guided Image-to-Image</td>
-    <td><a href="./api/pipelines/stable_diffusion/pix2pix">Instruct Pix2Pix</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/pix2pix">Instruct Pix2Pix</a></td>
       <td><a href="https://huggingface.co/timbrooks/instruct-pix2pix"> timbrooks/instruct-pix2pix </a></td>
   </tr>
   <tr>
     <td>Text-guided Image-to-Image</td>
-    <td><a href="./api/pipelines/stable_diffusion/img2img">Stable Diffusion Image-to-Image</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/img2img">Stable Diffusion Image-to-Image</a></td>
       <td><a href="https://huggingface.co/runwayml/stable-diffusion-v1-5"> runwayml/stable-diffusion-v1-5 </a></td>
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Text-guided Image Inpainting</td>
-    <td><a href="./api/pipelines/stable_diffusion/inpaint">Stable Diffusion Inpaint</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/inpaint">Stable Diffusion Inpaint</a></td>
       <td><a href="https://huggingface.co/runwayml/stable-diffusion-inpainting"> runwayml/stable-diffusion-inpainting </a></td>
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Image Variation</td>
-    <td><a href="./stable_diffusion/image_variation">Stable Diffusion Image Variation</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/image_variation">Stable Diffusion Image Variation</a></td>
       <td><a href="https://huggingface.co/lambdalabs/sd-image-variations-diffusers"> lambdalabs/sd-image-variations-diffusers </a></td>
   </tr>
   <tr style="border-top: 2px solid black">
     <td>Super Resolution</td>
-    <td><a href="./stable_diffusion/stable_diffusion/upscale">Stable Diffusion Upscale</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/upscale">Stable Diffusion Upscale</a></td>
       <td><a href="https://huggingface.co/stabilityai/stable-diffusion-x4-upscaler"> stabilityai/stable-diffusion-x4-upscaler </a></td>
   </tr>
   <tr>
     <td>Super Resolution</td>
-    <td><a href="./stable_diffusion/latent_upscale">Stable Diffusion Latent Upscale</a></td>
+    <td><a href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/latent_upscale">Stable Diffusion Latent Upscale</a></td>
       <td><a href="https://huggingface.co/stabilityai/sd-x2-latent-upscaler"> stabilityai/sd-x2-latent-upscaler </a></td>
   </tr>
 </table>
 
 ## Popular libraries using 🧨 Diffusers
 
-- https://github.com/microsoft/TaskMatrix 
-- https://github.com/invoke-ai/InvokeAI 
-- https://github.com/apple/ml-stable-diffusion   
-- https://github.com/Sanster/lama-cleaner 
+- https://github.com/microsoft/TaskMatrix
+- https://github.com/invoke-ai/InvokeAI
+- https://github.com/apple/ml-stable-diffusion
+- https://github.com/Sanster/lama-cleaner
 - https://github.com/IDEA-Research/Grounded-Segment-Anything
-- https://github.com/ashawkey/stable-dreamfusion 
-- https://github.com/deep-floyd/IF  
+- https://github.com/ashawkey/stable-dreamfusion
+- https://github.com/deep-floyd/IF
 - https://github.com/bentoml/BentoML
 - https://github.com/bmaltais/kohya_ss
 - +3000 other amazing GitHub repositories 💪

docker/diffusers-pytorch-cuda/Dockerfile

@@ -38,6 +38,8 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip && \
         scipy \
         tensorboard \
         transformers \
-        omegaconf
+        omegaconf \
+        pytorch-lightning \
+        xformers
 
 CMD ["/bin/bash"]

docs/source/_config.py

@@ -6,4 +6,4 @@ INSTALL_CONTENT = """
 # ! pip install git+https://github.com/huggingface/diffusers.git
 """
 
-notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]
+notebook_first_cells = [{"type": "code", "content": INSTALL_CONTENT}]

docs/source/en/_toctree.yml

@@ -28,8 +28,8 @@
       title: Load community pipelines
     - local: using-diffusers/using_safetensors
       title: Load safetensors
-    - local: using-diffusers/kerascv
-      title: Load KerasCV Stable Diffusion checkpoints
+    - local: using-diffusers/other-formats
+      title: Load different Stable Diffusion formats
     title: Loading & Hub
   - sections:
     - local: using-diffusers/pipeline_overview
@@ -50,6 +50,8 @@
       title: Distributed inference with multiple GPUs
     - local: using-diffusers/reusing_seeds
       title: Improve image quality with deterministic generation
+    - local: using-diffusers/control_brightness
+      title: Control image brightness
     - local: using-diffusers/reproducibility
       title: Create reproducible pipelines
     - local: using-diffusers/custom_pipeline_examples
@@ -130,8 +132,8 @@
   title: Conceptual Guides
 - sections:
   - sections:
-    - local: api/models
-      title: Models
+    - local: api/attnprocessor
+      title: Attention Processor
     - local: api/diffusion_pipeline
       title: Diffusion Pipeline
     - local: api/logging
@@ -142,12 +144,42 @@
       title: Outputs
     - local: api/loaders
       title: Loaders
+    - local: api/utilities
+      title: Utilities
+    - local: api/image_processor
+      title: VAE Image Processor
     title: Main Classes
+  - sections:
+    - local: api/models/overview
+      title: Overview
+    - local: api/models/unet
+      title: UNet1DModel
+    - local: api/models/unet2d
+      title: UNet2DModel
+    - local: api/models/unet2d-cond
+      title: UNet2DConditionModel
+    - local: api/models/unet3d-cond
+      title: UNet3DConditionModel
+    - local: api/models/vq
+      title: VQModel
+    - local: api/models/autoencoderkl
+      title: AutoencoderKL
+    - local: api/models/transformer2d
+      title: Transformer2D
+    - local: api/models/transformer_temporal
+      title: Transformer Temporal
+    - local: api/models/prior_transformer
+      title: Prior Transformer
+    - local: api/models/controlnet
+      title: ControlNet
+    title: Models
   - sections:
     - local: api/pipelines/overview
       title: Overview
     - local: api/pipelines/alt_diffusion
       title: AltDiffusion
+    - local: api/pipelines/attend_and_excite
+      title: Attend and Excite
     - local: api/pipelines/audio_diffusion
       title: Audio Diffusion
     - local: api/pipelines/audioldm
@@ -162,22 +194,34 @@
       title: DDIM
     - local: api/pipelines/ddpm
       title: DDPM
+    - local: api/pipelines/diffedit
+      title: DiffEdit
     - local: api/pipelines/dit
       title: DiT
     - local: api/pipelines/if
       title: IF
+    - local: api/pipelines/pix2pix
+      title: InstructPix2Pix
+    - local: api/pipelines/kandinsky
+      title: Kandinsky
     - local: api/pipelines/latent_diffusion
       title: Latent Diffusion
+    - local: api/pipelines/panorama
+      title: MultiDiffusion Panorama
     - local: api/pipelines/paint_by_example
       title: PaintByExample
+    - local: api/pipelines/paradigms
+      title: Parallel Sampling of Diffusion Models
+    - local: api/pipelines/pix2pix_zero
+      title: Pix2Pix Zero
     - local: api/pipelines/pndm
       title: PNDM
     - local: api/pipelines/repaint
       title: RePaint
-    - local: api/pipelines/stable_diffusion_safe
-      title: Safe Stable Diffusion
     - local: api/pipelines/score_sde_ve
       title: Score SDE VE
+    - local: api/pipelines/self_attention_guidance
+      title: Self-Attention Guidance
     - local: api/pipelines/semantic_stable_diffusion
       title: Semantic Guidance
     - local: api/pipelines/spectrogram_diffusion
@@ -195,31 +239,23 @@
         title: Depth-to-Image
       - local: api/pipelines/stable_diffusion/image_variation
         title: Image-Variation
-      - local: api/pipelines/stable_diffusion/upscale
-        title: Super-Resolution
+      - local: api/pipelines/stable_diffusion/stable_diffusion_safe
+        title: Safe Stable Diffusion
+      - local: api/pipelines/stable_diffusion/stable_diffusion_2
+        title: Stable Diffusion 2
       - local: api/pipelines/stable_diffusion/latent_upscale
         title: Stable-Diffusion-Latent-Upscaler
-      - local: api/pipelines/stable_diffusion/pix2pix
-        title: InstructPix2Pix
-      - local: api/pipelines/stable_diffusion/attend_and_excite
-        title: Attend and Excite
-      - local: api/pipelines/stable_diffusion/pix2pix_zero
-        title: Pix2Pix Zero
-      - local: api/pipelines/stable_diffusion/self_attention_guidance
-        title: Self-Attention Guidance
-      - local: api/pipelines/stable_diffusion/panorama
-        title: MultiDiffusion Panorama
-      - local: api/pipelines/stable_diffusion/model_editing
-        title: Text-to-Image Model Editing
-      - local: api/pipelines/stable_diffusion/diffedit
-        title: DiffEdit
+      - local: api/pipelines/stable_diffusion/upscale
+        title: Super-Resolution
+      - local: api/pipelines/stable_diffusion/ldm3d_diffusion
+        title: LDM3D Text-to-(RGB, Depth)
       title: Stable Diffusion
-    - local: api/pipelines/stable_diffusion_2
-      title: Stable Diffusion 2
     - local: api/pipelines/stable_unclip
       title: Stable unCLIP
     - local: api/pipelines/stochastic_karras_ve
       title: Stochastic Karras VE
+    - local: api/pipelines/model_editing
+      title: Text-to-Image Model Editing
     - local: api/pipelines/text_to_video
       title: Text-to-Video
     - local: api/pipelines/text_to_video_zero
@@ -228,6 +264,8 @@
       title: UnCLIP
     - local: api/pipelines/latent_diffusion_uncond
       title: Unconditional Latent Diffusion
+    - local: api/pipelines/unidiffuser
+      title: UniDiffuser
     - local: api/pipelines/versatile_diffusion
       title: Versatile Diffusion
     - local: api/pipelines/vq_diffusion

docs/source/en/api/attnprocessor.mdx

@@ -0,0 +1,42 @@
+# Attention Processor
+
+An attention processor is a class for applying different types of attention mechanisms.
+
+## AttnProcessor
+[[autodoc]] models.attention_processor.AttnProcessor
+
+## AttnProcessor2_0
+[[autodoc]] models.attention_processor.AttnProcessor2_0
+
+## LoRAAttnProcessor
+[[autodoc]] models.attention_processor.LoRAAttnProcessor
+
+## LoRAAttnProcessor2_0
+[[autodoc]] models.attention_processor.LoRAAttnProcessor2_0
+
+## CustomDiffusionAttnProcessor
+[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor
+
+## AttnAddedKVProcessor
+[[autodoc]] models.attention_processor.AttnAddedKVProcessor
+
+## AttnAddedKVProcessor2_0
+[[autodoc]] models.attention_processor.AttnAddedKVProcessor2_0
+
+## LoRAAttnAddedKVProcessor
+[[autodoc]] models.attention_processor.LoRAAttnAddedKVProcessor
+
+## XFormersAttnProcessor
+[[autodoc]] models.attention_processor.XFormersAttnProcessor
+
+## LoRAXFormersAttnProcessor
+[[autodoc]] models.attention_processor.LoRAXFormersAttnProcessor
+
+## CustomDiffusionXFormersAttnProcessor
+[[autodoc]] models.attention_processor.CustomDiffusionXFormersAttnProcessor
+
+## SlicedAttnProcessor
+[[autodoc]] models.attention_processor.SlicedAttnProcessor
+
+## SlicedAttnAddedKVProcessor
+[[autodoc]] models.attention_processor.SlicedAttnAddedKVProcessor

docs/source/en/api/configuration.mdx

@@ -12,8 +12,13 @@ specific language governing permissions and limitations under the License.
 
 # Configuration
 
-Schedulers from [`~schedulers.scheduling_utils.SchedulerMixin`] and models from [`ModelMixin`] inherit from [`ConfigMixin`] which conveniently takes care of storing all the parameters that are 
-passed to their respective `__init__` methods in a JSON-configuration file.
+Schedulers from [`~schedulers.scheduling_utils.SchedulerMixin`] and models from [`ModelMixin`] inherit from [`ConfigMixin`] which stores all the parameters that are passed to their respective `__init__` methods in a JSON-configuration file.
+
+<Tip>
+
+To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with `huggingface-cli login`.
+
+</Tip>
 
 ## ConfigMixin
 

docs/source/en/api/diffusion_pipeline.mdx

@@ -12,36 +12,25 @@ specific language governing permissions and limitations under the License.
 
 # Pipelines
 
-The [`DiffusionPipeline`] is the easiest way to load any pretrained diffusion pipeline from the [Hub](https://huggingface.co/models?library=diffusers) and to use it in inference.
+The [`DiffusionPipeline`] is the quickest way to load any pretrained diffusion pipeline from the [Hub](https://huggingface.co/models?library=diffusers) for inference.
 
 <Tip>
-	
-	One should not use the Diffusion Pipeline class for training or fine-tuning a diffusion model. Individual 
-	components of diffusion pipelines are usually trained individually, so we suggest to directly work 
-	with [`UNetModel`] and [`UNetConditionModel`].
+
+You shouldn't use the [`DiffusionPipeline`] class for training or finetuning a diffusion model. Individual 
+components (for example, [`UNet2DModel`] and [`UNet2DConditionModel`]) of diffusion pipelines are usually trained individually, so we suggest directly working with them instead.
 
 </Tip>
 
-Any diffusion pipeline that is loaded with [`~DiffusionPipeline.from_pretrained`] will automatically 
-detect the pipeline type, *e.g.* [`StableDiffusionPipeline`] and consequently load each component of the 
-pipeline and pass them into the `__init__` function of the pipeline, *e.g.* [`~StableDiffusionPipeline.__init__`].
+The pipeline type (for example [`StableDiffusionPipeline`]) of any diffusion pipeline loaded with [`~DiffusionPipeline.from_pretrained`] is automatically 
+detected and pipeline components are loaded and passed to the `__init__` function of the pipeline.
 
 Any pipeline object can be saved locally with [`~DiffusionPipeline.save_pretrained`].
 
 ## DiffusionPipeline
+
 [[autodoc]] DiffusionPipeline
 	- all
 	- __call__
 	- device
 	- to
 	- components
-
-## ImagePipelineOutput
-By default diffusion pipelines return an object of class
-
-[[autodoc]] pipelines.ImagePipelineOutput
-
-## AudioPipelineOutput
-By default diffusion pipelines return an object of class
-
-[[autodoc]] pipelines.AudioPipelineOutput

docs/source/en/api/image_processor.mdx

@@ -0,0 +1,27 @@
+<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# VAE Image Processor
+
+The [`VaeImageProcessor`] provides a unified API for [`StableDiffusionPipeline`]'s to prepare image inputs for VAE encoding and post-processing outputs once they're decoded. This includes transformations such as resizing, normalization, and conversion between PIL Image, PyTorch, and NumPy arrays. 
+
+All pipelines with [`VaeImageProcessor`] accepts PIL Image, PyTorch tensor, or NumPy arrays as image inputs and returns outputs based on the `output_type` argument by the user. You can pass encoded image latents directly to the pipeline and return latents from the pipeline as a specific output with the `output_type` argument (for example `output_type="pt"`). This allows you to take the generated latents from one pipeline and pass it to another pipeline as input without leaving the latent space. It also makes it much easier to use multiple pipelines together by passing PyTorch tensors directly between different pipelines. 
+
+## VaeImageProcessor
+
+[[autodoc]] image_processor.VaeImageProcessor
+
+## VaeImageProcessorLDM3D
+
+The [`VaeImageProcessorLDM3D`] accepts RGB and depth inputs and returns RGB and depth outputs.
+
+[[autodoc]] image_processor.VaeImageProcessorLDM3D

docs/source/en/api/loaders.mdx

@@ -12,31 +12,26 @@ specific language governing permissions and limitations under the License.
 
 # Loaders
 
-There are many ways to train adapter neural networks for diffusion models, such as 
-- [Textual Inversion](./training/text_inversion.mdx)
-- [LoRA](https://github.com/cloneofsimo/lora)
-- [Hypernetworks](https://arxiv.org/abs/1609.09106)
+Adapters (textual inversion, LoRA, hypernetworks) allow you to modify a diffusion model to generate images in a specific style without training or finetuning the entire model. The adapter weights are typically only a tiny fraction of the pretrained model's which making them very portable. 🤗 Diffusers provides an easy-to-use `LoaderMixin` API to load adapter weights.
 
-Such adapter neural networks often only consist of a fraction of the number of weights compared 
-to the pretrained model and as such are very portable. The Diffusers library offers an easy-to-use
-API to load such adapter neural networks via the [`loaders.py` module](https://github.com/huggingface/diffusers/blob/main/src/diffusers/loaders.py). 
+<Tip warning={true}>
 
-**Note**: This module is still highly experimental and prone to future changes.
+🧪 The `LoaderMixins` are highly experimental and prone to future changes. To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with `huggingface-cli login`.
 
-## LoaderMixins
+</Tip>
 
-### UNet2DConditionLoadersMixin
+## UNet2DConditionLoadersMixin
 
 [[autodoc]] loaders.UNet2DConditionLoadersMixin
 
-### TextualInversionLoaderMixin
+## TextualInversionLoaderMixin
 
 [[autodoc]] loaders.TextualInversionLoaderMixin
 
-### LoraLoaderMixin
+## LoraLoaderMixin
 
 [[autodoc]] loaders.LoraLoaderMixin
 
-### FromCkptMixin
+## FromCkptMixin
 
 [[autodoc]] loaders.FromCkptMixin

docs/source/en/api/logging.mdx

@@ -12,12 +12,9 @@ specific language governing permissions and limitations under the License.
 
 # Logging
 
-🧨 Diffusers has a centralized logging system, so that you can setup the verbosity of the library easily.
+🤗 Diffusers has a centralized logging system to easily manage the verbosity of the library. The default verbosity is set to `WARNING`.
 
-Currently the default verbosity of the library is `WARNING`.
-
-To change the level of verbosity, just use one of the direct setters. For instance, here is how to change the verbosity
-to the INFO level.
+To change the verbosity level, use one of the direct setters. For instance, to change the verbosity to the `INFO` level.
 
 ```python
 import diffusers
@@ -33,7 +30,7 @@ DIFFUSERS_VERBOSITY=error ./myprogram.py
 ```
 
 Additionally, some `warnings` can be disabled by setting the environment variable
-`DIFFUSERS_NO_ADVISORY_WARNINGS` to a true value, like *1*. This will disable any warning that is logged using
+`DIFFUSERS_NO_ADVISORY_WARNINGS` to a true value, like `1`. This disables any warning logged by
 [`logger.warning_advice`]. For example:
 
 ```bash
@@ -52,20 +49,21 @@ logger.warning("WARN")
 ```
 
 
-All the methods of this logging module are documented below, the main ones are
+All methods of the logging module are documented below. The main methods are
 [`logging.get_verbosity`] to get the current level of verbosity in the logger and
-[`logging.set_verbosity`] to set the verbosity to the level of your choice. In order (from the least
-verbose to the most verbose), those levels (with their corresponding int values in parenthesis) are:
+[`logging.set_verbosity`] to set the verbosity to the level of your choice. 
 
-- `diffusers.logging.CRITICAL` or `diffusers.logging.FATAL` (int value, 50): only report the most
-  critical errors.
-- `diffusers.logging.ERROR` (int value, 40): only report errors.
-- `diffusers.logging.WARNING` or `diffusers.logging.WARN` (int value, 30): only reports error and
-  warnings. This is the default level used by the library.
-- `diffusers.logging.INFO` (int value, 20): reports error, warnings and basic information.
-- `diffusers.logging.DEBUG` (int value, 10): report all information.
+In order from the least verbose to the most verbose:
 
-By default, `tqdm` progress bars will be displayed during model download. [`logging.disable_progress_bar`] and [`logging.enable_progress_bar`] can be used to suppress or unsuppress this behavior.
+|                                                    Method | Integer value |                                         Description |
+|----------------------------------------------------------:|--------------:|----------------------------------------------------:|
+| `diffusers.logging.CRITICAL` or `diffusers.logging.FATAL` |            50 |                only report the most critical errors |
+|                                 `diffusers.logging.ERROR` |            40 |                                  only report errors |
+|   `diffusers.logging.WARNING` or `diffusers.logging.WARN` |            30 |           only report errors and warnings (default) |
+|                                  `diffusers.logging.INFO` |            20 | only report errors, warnings, and basic information |
+|                                 `diffusers.logging.DEBUG` |            10 |                              report all information |
+
+By default, `tqdm` progress bars are displayed during model download. [`logging.disable_progress_bar`] and [`logging.enable_progress_bar`] are used to enable or disable this behavior.
 
 ## Base setters
 

docs/source/en/api/models.mdx

@@ -1,107 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Models
-
-Diffusers contains pretrained models for popular algorithms and modules for creating the next set of diffusion models.
-The primary function of these models is to denoise an input sample, by modeling the distribution $p_\theta(\mathbf{x}_{t-1}|\mathbf{x}_t)$.
-The models are built on the base class ['ModelMixin'] that is a `torch.nn.module` with basic functionality for saving and loading models both locally and from the HuggingFace hub.
-
-## ModelMixin
-[[autodoc]] ModelMixin
-
-## UNet2DOutput
-[[autodoc]] models.unet_2d.UNet2DOutput
-
-## UNet2DModel
-[[autodoc]] UNet2DModel
-
-## UNet1DOutput
-[[autodoc]] models.unet_1d.UNet1DOutput
-
-## UNet1DModel
-[[autodoc]] UNet1DModel
-
-## UNet2DConditionOutput
-[[autodoc]] models.unet_2d_condition.UNet2DConditionOutput
-
-## UNet2DConditionModel
-[[autodoc]] UNet2DConditionModel
-
-## UNet3DConditionOutput
-[[autodoc]] models.unet_3d_condition.UNet3DConditionOutput
-
-## UNet3DConditionModel
-[[autodoc]] UNet3DConditionModel
-
-## DecoderOutput
-[[autodoc]] models.vae.DecoderOutput
-
-## VQEncoderOutput
-[[autodoc]] models.vq_model.VQEncoderOutput
-
-## VQModel
-[[autodoc]] VQModel
-
-## AutoencoderKLOutput
-[[autodoc]] models.autoencoder_kl.AutoencoderKLOutput
-
-## AutoencoderKL
-[[autodoc]] AutoencoderKL
-
-## Transformer2DModel
-[[autodoc]] Transformer2DModel
-
-## Transformer2DModelOutput
-[[autodoc]] models.transformer_2d.Transformer2DModelOutput
-
-## TransformerTemporalModel
-[[autodoc]] models.transformer_temporal.TransformerTemporalModel
-
-## Transformer2DModelOutput
-[[autodoc]] models.transformer_temporal.TransformerTemporalModelOutput
-
-## PriorTransformer
-[[autodoc]] models.prior_transformer.PriorTransformer
-
-## PriorTransformerOutput
-[[autodoc]] models.prior_transformer.PriorTransformerOutput
-
-## ControlNetOutput
-[[autodoc]] models.controlnet.ControlNetOutput
-
-## ControlNetModel
-[[autodoc]] ControlNetModel
-
-## FlaxModelMixin
-[[autodoc]] FlaxModelMixin
-
-## FlaxUNet2DConditionOutput
-[[autodoc]] models.unet_2d_condition_flax.FlaxUNet2DConditionOutput
-
-## FlaxUNet2DConditionModel
-[[autodoc]] FlaxUNet2DConditionModel
-
-## FlaxDecoderOutput
-[[autodoc]] models.vae_flax.FlaxDecoderOutput
-
-## FlaxAutoencoderKLOutput
-[[autodoc]] models.vae_flax.FlaxAutoencoderKLOutput
-
-## FlaxAutoencoderKL
-[[autodoc]] FlaxAutoencoderKL
-
-## FlaxControlNetOutput
-[[autodoc]] models.controlnet_flax.FlaxControlNetOutput
-
-## FlaxControlNetModel
-[[autodoc]] FlaxControlNetModel

docs/source/en/api/models/autoencoderkl.mdx

@@ -0,0 +1,31 @@
+# AutoencoderKL
+
+The variational autoencoder (VAE) model with KL loss was introduced in [Auto-Encoding Variational Bayes](https://arxiv.org/abs/1312.6114v11) by Diederik P. Kingma and Max Welling. The model is used in 🤗 Diffusers to encode images into latents and to decode latent representations into images.
+
+The abstract from the paper is:
+
+*How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions, and large datasets? We introduce a stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case. Our contributions are two-fold. First, we show that a reparameterization of the variational lower bound yields a lower bound estimator that can be straightforwardly optimized using standard stochastic gradient methods. Second, we show that for i.i.d. datasets with continuous latent variables per datapoint, posterior inference can be made especially efficient by fitting an approximate inference model (also called a recognition model) to the intractable posterior using the proposed lower bound estimator. Theoretical advantages are reflected in experimental results.*
+
+## AutoencoderKL
+
+[[autodoc]] AutoencoderKL
+
+## AutoencoderKLOutput
+
+[[autodoc]] models.autoencoder_kl.AutoencoderKLOutput
+
+## DecoderOutput
+
+[[autodoc]] models.vae.DecoderOutput
+
+## FlaxAutoencoderKL
+
+[[autodoc]] FlaxAutoencoderKL
+
+## FlaxAutoencoderKLOutput
+
+[[autodoc]] models.vae_flax.FlaxAutoencoderKLOutput
+
+## FlaxDecoderOutput
+
+[[autodoc]] models.vae_flax.FlaxDecoderOutput

docs/source/en/api/models/controlnet.mdx

@@ -0,0 +1,23 @@
+# ControlNet
+
+The ControlNet model was introduced in [Adding Conditional Control to Text-to-Image Diffusion Models](https://huggingface.co/papers/2302.05543) by Lvmin Zhang and Maneesh Agrawala. It provides a greater degree of control over text-to-image generation by conditioning the model on additional inputs such as edge maps, depth maps, segmentation maps, and keypoints for pose detection.
+
+The abstract from the paper is:
+
+*We present a neural network structure, ControlNet, to control pretrained large diffusion models to support additional input conditions. The ControlNet learns task-specific conditions in an end-to-end way, and the learning is robust even when the training dataset is small (< 50k). Moreover, training a ControlNet is as fast as fine-tuning a diffusion model, and the model can be trained on a personal devices. Alternatively, if powerful computation clusters are available, the model can scale to large amounts (millions to billions) of data. We report that large diffusion models like Stable Diffusion can be augmented with ControlNets to enable conditional inputs like edge maps, segmentation maps, keypoints, etc. This may enrich the methods to control large diffusion models and further facilitate related applications.*
+
+## ControlNetModel
+
+[[autodoc]] ControlNetModel
+
+## ControlNetOutput
+
+[[autodoc]] models.controlnet.ControlNetOutput
+
+## FlaxControlNetModel
+
+[[autodoc]] FlaxControlNetModel
+
+## FlaxControlNetOutput
+
+[[autodoc]] models.controlnet_flax.FlaxControlNetOutput

docs/source/en/api/models/overview.mdx

@@ -0,0 +1,12 @@
+# Models
+
+🤗 Diffusers provides pretrained models for popular algorithms and modules to create custom diffusion systems. The primary function of models is to denoise an input sample as modeled by the distribution \\(p_{\theta}(x_{t-1}|x_{t})\\).
+
+All models are built from the base [`ModelMixin`] class which is a [`torch.nn.module`](https://pytorch.org/docs/stable/generated/torch.nn.Module.html) providing basic functionality for saving and loading models, locally and from the Hugging Face Hub.
+
+## ModelMixin
+[[autodoc]] ModelMixin
+
+## FlaxModelMixin
+
+[[autodoc]] FlaxModelMixin

docs/source/en/api/models/prior_transformer.mdx

@@ -0,0 +1,16 @@
+# Prior Transformer
+
+The Prior Transformer was originally introduced in [Hierarchical Text-Conditional Image Generation with CLIP Latents
+](https://huggingface.co/papers/2204.06125) by Ramesh et al. It is used to predict CLIP image embeddings from CLIP text embeddings; image embeddings are predicted through a denoising diffusion process.
+
+The abstract from the paper is:
+
+*Contrastive models like CLIP have been shown to learn robust representations of images that capture both semantics and style. To leverage these representations for image generation, we propose a two-stage model: a prior that generates a CLIP image embedding given a text caption, and a decoder that generates an image conditioned on the image embedding. We show that explicitly generating image representations improves image diversity with minimal loss in photorealism and caption similarity. Our decoders conditioned on image representations can also produce variations of an image that preserve both its semantics and style, while varying the non-essential details absent from the image representation. Moreover, the joint embedding space of CLIP enables language-guided image manipulations in a zero-shot fashion. We use diffusion models for the decoder and experiment with both autoregressive and diffusion models for the prior, finding that the latter are computationally more efficient and produce higher-quality samples.*
+
+## PriorTransformer
+
+[[autodoc]] PriorTransformer
+
+## PriorTransformerOutput
+
+[[autodoc]] models.prior_transformer.PriorTransformerOutput

docs/source/en/api/models/transformer2d.mdx

@@ -0,0 +1,29 @@
+# Transformer2D
+
+A Transformer model for image-like data from [CompVis](https://huggingface.co/CompVis) that is based on the [Vision Transformer](https://huggingface.co/papers/2010.11929) introduced by Dosovitskiy et al. The [`Transformer2DModel`] accepts discrete (classes of vector embeddings) or continuous (actual embeddings) inputs.
+
+When the input is **continuous**:
+
+1. Project the input and reshape it to `(batch_size, sequence_length, feature_dimension)`.
+2. Apply the Transformer blocks in the standard way.
+3. Reshape to image.
+
+When the input is **discrete**:
+
+<Tip>
+
+It is assumed one of the input classes is the masked latent pixel. The predicted classes of the unnoised image don't contain a prediction for the masked pixel because the unnoised image cannot be masked.
+
+</Tip>
+
+1. Convert input (classes of latent pixels) to embeddings and apply positional embeddings.
+2. Apply the Transformer blocks in the standard way.
+3. Predict classes of unnoised image.
+
+## Transformer2DModel
+
+[[autodoc]] Transformer2DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.transformer_2d.Transformer2DModelOutput

docs/source/en/api/models/transformer_temporal.mdx

@@ -0,0 +1,11 @@
+# Transformer Temporal
+
+A Transformer model for video-like data.
+
+## TransformerTemporalModel
+
+[[autodoc]] models.transformer_temporal.TransformerTemporalModel
+
+## TransformerTemporalModelOutput
+
+[[autodoc]] models.transformer_temporal.TransformerTemporalModelOutput

docs/source/en/api/models/unet.mdx

@@ -0,0 +1,13 @@
+# UNet1DModel
+
+The [UNet](https://huggingface.co/papers/1505.04597) model was originally introduced by Ronneberger et al for biomedical image segmentation, but it is also commonly used in 🤗 Diffusers because it outputs images that are the same size as the input. It is one of the most important components of a diffusion system because it facilitates the actual diffusion process. There are several variants of the UNet model in 🤗 Diffusers, depending on it's number of dimensions and whether it is a conditional model or not. This is a 1D UNet model.
+
+The abstract from the paper is:
+
+*There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net.*
+
+## UNet1DModel
+[[autodoc]] UNet1DModel
+
+## UNet1DOutput
+[[autodoc]] models.unet_1d.UNet1DOutput

docs/source/en/api/models/unet2d-cond.mdx

@@ -0,0 +1,19 @@
+# UNet2DConditionModel
+
+The [UNet](https://huggingface.co/papers/1505.04597) model was originally introduced by Ronneberger et al for biomedical image segmentation, but it is also commonly used in 🤗 Diffusers because it outputs images that are the same size as the input. It is one of the most important components of a diffusion system because it facilitates the actual diffusion process. There are several variants of the UNet model in 🤗 Diffusers, depending on it's number of dimensions and whether it is a conditional model or not. This is a 2D UNet conditional model.
+
+The abstract from the paper is:
+
+*There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net.*
+
+## UNet2DConditionModel
+[[autodoc]] UNet2DConditionModel
+
+## UNet2DConditionOutput
+[[autodoc]] models.unet_2d_condition.UNet2DConditionOutput
+
+## FlaxUNet2DConditionModel
+[[autodoc]] models.unet_2d_condition_flax.FlaxUNet2DConditionModel
+
+## FlaxUNet2DConditionOutput
+[[autodoc]] models.unet_2d_condition_flax.FlaxUNet2DConditionOutput

docs/source/en/api/models/unet2d.mdx

@@ -0,0 +1,13 @@
+# UNet2DModel
+
+The [UNet](https://huggingface.co/papers/1505.04597) model was originally introduced by Ronneberger et al for biomedical image segmentation, but it is also commonly used in 🤗 Diffusers because it outputs images that are the same size as the input. It is one of the most important components of a diffusion system because it facilitates the actual diffusion process. There are several variants of the UNet model in 🤗 Diffusers, depending on it's number of dimensions and whether it is a conditional model or not. This is a 2D UNet model.
+
+The abstract from the paper is:
+
+*There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net.*
+
+## UNet2DModel
+[[autodoc]] UNet2DModel
+
+## UNet2DOutput
+[[autodoc]] models.unet_2d.UNet2DOutput

docs/source/en/api/models/unet3d-cond.mdx

@@ -0,0 +1,13 @@
+# UNet3DConditionModel
+
+The [UNet](https://huggingface.co/papers/1505.04597) model was originally introduced by Ronneberger et al for biomedical image segmentation, but it is also commonly used in 🤗 Diffusers because it outputs images that are the same size as the input. It is one of the most important components of a diffusion system because it facilitates the actual diffusion process. There are several variants of the UNet model in 🤗 Diffusers, depending on it's number of dimensions and whether it is a conditional model or not. This is a 3D UNet conditional model.
+
+The abstract from the paper is:
+
+*There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at http://lmb.informatik.uni-freiburg.de/people/ronneber/u-net.*
+
+## UNet3DConditionModel
+[[autodoc]] UNet3DConditionModel
+
+## UNet3DConditionOutput
+[[autodoc]] models.unet_3d_condition.UNet3DConditionOutput

docs/source/en/api/models/vq.mdx

@@ -0,0 +1,15 @@
+# VQModel
+
+The VQ-VAE model was introduced in [Neural Discrete Representation Learning](https://huggingface.co/papers/1711.00937) by Aaron van den Oord, Oriol Vinyals and Koray Kavukcuoglu. The model is used in 🤗 Diffusers to decode latent representations into images. Unlike [`AutoencoderKL`], the [`VQModel`] works in a quantized latent space.
+
+The abstract from the paper is:
+
+*Learning useful representations without supervision remains a key challenge in machine learning. In this paper, we propose a simple yet powerful generative model that learns such discrete representations. Our model, the Vector Quantised-Variational AutoEncoder (VQ-VAE), differs from VAEs in two key ways: the encoder network outputs discrete, rather than continuous, codes; and the prior is learnt rather than static. In order to learn a discrete latent representation, we incorporate ideas from vector quantisation (VQ). Using the VQ method allows the model to circumvent issues of "posterior collapse" -- where the latents are ignored when they are paired with a powerful autoregressive decoder -- typically observed in the VAE framework. Pairing these representations with an autoregressive prior, the model can generate high quality images, videos, and speech as well as doing high quality speaker conversion and unsupervised learning of phonemes, providing further evidence of the utility of the learnt representations.*
+
+## VQModel
+
+[[autodoc]] VQModel
+
+## VQEncoderOutput
+
+[[autodoc]] models.vq_model.VQEncoderOutput

docs/source/en/api/outputs.mdx

@@ -10,13 +10,11 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# BaseOutputs
+# Outputs
 
-All models have outputs that are instances of subclasses of [`~utils.BaseOutput`]. Those are
-data structures containing all the information returned by the model, but that can also be used as tuples or
-dictionaries.
+All models outputs are subclasses of [`~utils.BaseOutput`], data structures containing all the information returned by the model. The outputs can also be used as tuples or dictionaries.
 
-Let's see how this looks in an example:
+For example:
 
 ```python
 from diffusers import DDIMPipeline
@@ -25,31 +23,45 @@ pipeline = DDIMPipeline.from_pretrained("google/ddpm-cifar10-32")
 outputs = pipeline()
 ```
 
-The `outputs` object is a [`~pipelines.ImagePipelineOutput`], as we can see in the
-documentation of that class below, it means it has an image attribute.
+The `outputs` object is a [`~pipelines.ImagePipelineOutput`] which means it has an image attribute.
 
-You can access each attribute as you would usually do, and if that attribute has not been returned by the model, you will get `None`:
+You can access each attribute as you normally would or with a keyword lookup, and if that attribute is not returned by the model, you will get `None`:
 
 ```python
 outputs.images
-```
-
-or via keyword lookup
-
-```python
 outputs["images"]
 ```
 
-When considering our `outputs` object as tuple, it only considers the attributes that don't have `None` values.
-Here for instance, we could retrieve images via indexing:
+When considering the `outputs` object as a tuple, it only considers the attributes that don't have `None` values.
+For instance, retrieving an image by indexing into it returns the tuple `(outputs.images)`:
 
 ```python
 outputs[:1]
 ```
 
-which will return the tuple `(outputs.images)` for instance.
+<Tip>
+
+To check a specific pipeline or model output, refer to its corresponding API documentation.
+
+</Tip>
 
 ## BaseOutput
 
 [[autodoc]] utils.BaseOutput
     - to_tuple
+
+## ImagePipelineOutput
+
+[[autodoc]] pipelines.ImagePipelineOutput
+
+## FlaxImagePipelineOutput
+
+[[autodoc]] pipelines.pipeline_flax_utils.FlaxImagePipelineOutput
+
+## AudioPipelineOutput
+
+[[autodoc]] pipelines.AudioPipelineOutput
+
+## ImageTextPipelineOutput
+
+[[autodoc]] ImageTextPipelineOutput

docs/source/en/api/pipelines/kandinsky.mdx

@@ -0,0 +1,365 @@
+<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+http://www.apache.org/licenses/LICENSE-2.0
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# Kandinsky
+
+## Overview
+
+Kandinsky 2.1 inherits best practices from [DALL-E 2](https://arxiv.org/abs/2204.06125) and [Latent Diffusion](https://huggingface.co/docs/diffusers/api/pipelines/latent_diffusion), while introducing some new ideas.
+
+It uses [CLIP](https://huggingface.co/docs/transformers/model_doc/clip) for encoding images and text, and a diffusion image prior (mapping) between latent spaces of CLIP modalities. This approach enhances the visual performance of the model and unveils new horizons in blending images and text-guided image manipulation.
+
+The Kandinsky model is created by [Arseniy Shakhmatov](https://github.com/cene555), [Anton Razzhigaev](https://github.com/razzant), [Aleksandr Nikolich](https://github.com/AlexWortega), [Igor Pavlov](https://github.com/boomb0om), [Andrey Kuznetsov](https://github.com/kuznetsoffandrey) and [Denis Dimitrov](https://github.com/denndimitrov) and the original codebase can be found [here](https://github.com/ai-forever/Kandinsky-2)
+
+## Available Pipelines:
+
+| Pipeline | Tasks |
+|---|---|
+| [pipeline_kandinsky.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py) | *Text-to-Image Generation* |
+| [pipeline_kandinsky_inpaint.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py) | *Image-Guided Image Generation* |
+| [pipeline_kandinsky_img2img.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py) | *Image-Guided Image Generation* |
+
+## Usage example
+
+In the following, we will walk you through some examples of how to use the Kandinsky pipelines to create some visually aesthetic artwork.
+
+### Text-to-Image Generation
+
+For text-to-image generation, we need to use both [`KandinskyPriorPipeline`] and [`KandinskyPipeline`].
+The first step is to encode text prompts with CLIP and then diffuse the CLIP text embeddings to CLIP image embeddings,
+as first proposed in [DALL-E 2](https://cdn.openai.com/papers/dall-e-2.pdf).
+Let's throw a fun prompt at Kandinsky to see what it comes up with.
+
+```py
+prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
+```
+
+First, let's instantiate the prior pipeline and the text-to-image pipeline. Both 
+pipelines are diffusion models.
+
+
+```py
+from diffusers import DiffusionPipeline
+import torch
+
+pipe_prior = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16)
+pipe_prior.to("cuda")
+
+t2i_pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+t2i_pipe.to("cuda")
+```
+
+<Tip warning={true}>
+
+By default, the text-to-image pipeline use [`DDIMScheduler`], you can change the scheduler to [`DDPMScheduler`]
+
+```py
+scheduler = DDPMScheduler.from_pretrained("kandinsky-community/kandinsky-2-1", subfolder="ddpm_scheduler")
+t2i_pipe = DiffusionPipeline.from_pretrained(
+    "kandinsky-community/kandinsky-2-1", scheduler=scheduler, torch_dtype=torch.float16
+)
+t2i_pipe.to("cuda")
+```
+
+</Tip>
+
+Now we pass the prompt through the prior to generate image embeddings. The prior
+returns both the image embeddings corresponding to the prompt and negative/unconditional image 
+embeddings corresponding to an empty string.
+
+```py
+image_embeds, negative_image_embeds = pipe_prior(prompt, guidance_scale=1.0).to_tuple()
+```
+
+<Tip warning={true}>
+
+The text-to-image pipeline expects both `image_embeds`, `negative_image_embeds` and the original 
+`prompt` as the text-to-image pipeline uses another text encoder to better guide the second diffusion 
+process of `t2i_pipe`.
+
+By default, the prior returns unconditioned negative image embeddings corresponding to the negative prompt of `""`.
+For better results, you can also pass a `negative_prompt` to the prior. This will increase the effective batch size
+of the prior by a factor of 2.
+
+```py
+prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
+negative_prompt = "low quality, bad quality"
+
+image_embeds, negative_image_embeds = pipe_prior(prompt, negative_prompt, guidance_scale=1.0).to_tuple()
+```
+
+</Tip>
+
+
+Next, we can pass the embeddings as well as the prompt to the text-to-image pipeline. Remember that 
+in case you are using a customized negative prompt, that you should pass this one also to the text-to-image pipelines
+with `negative_prompt=negative_prompt`:
+
+```py
+image = t2i_pipe(
+    prompt, image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, height=768, width=768
+).images[0]
+image.save("cheeseburger_monster.png")
+```
+
+One cheeseburger monster coming up! Enjoy! 
+
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/cheeseburger.png)
+
+The Kandinsky model works extremely well with creative prompts. Here is some of the amazing art that can be created using the exact same process but with different prompts.
+
+```python
+prompt = "bird eye view shot of a full body woman with cyan light orange magenta makeup, digital art, long braided hair her face separated by makeup in the style of yin Yang surrealism, symmetrical face, real image, contrasting tone, pastel gradient background"
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/hair.png)
+
+```python
+prompt = "A car exploding into colorful dust"
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/dusts.png)
+
+```python
+prompt = "editorial photography of an organic, almost liquid smoke style armchair"
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/smokechair.png)
+
+```python
+prompt = "birds eye view of a quilted paper style alien planet landscape, vibrant colours, Cinematic lighting"
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/alienplanet.png)
+
+
+### Text Guided Image-to-Image Generation
+
+The same Kandinsky model weights can be used for text-guided image-to-image translation. In this case, just make sure to load the weights using the [`KandinskyImg2ImgPipeline`] pipeline.
+
+**Note**: You can also directly move the weights of the text-to-image pipelines to the image-to-image pipelines
+without loading them twice by making use of the [`~DiffusionPipeline.components`] function as explained [here](#converting-between-different-pipelines).
+
+Let's download an image.
+
+```python
+from PIL import Image
+import requests
+from io import BytesIO
+
+# download image
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+response = requests.get(url)
+original_image = Image.open(BytesIO(response.content)).convert("RGB")
+original_image = original_image.resize((768, 512))
+```
+
+![img](https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg)
+
+```python
+import torch
+from diffusers import KandinskyImg2ImgPipeline, KandinskyPriorPipeline
+
+# create prior
+pipe_prior = KandinskyPriorPipeline.from_pretrained(
+    "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16
+)
+pipe_prior.to("cuda")
+
+# create img2img pipeline
+pipe = KandinskyImg2ImgPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+pipe.to("cuda")
+
+prompt = "A fantasy landscape, Cinematic lighting"
+negative_prompt = "low quality, bad quality"
+
+image_embeds, negative_image_embeds = pipe_prior(prompt, negative_prompt).to_tuple()
+
+out = pipe(
+    prompt,
+    image=original_image,
+    image_embeds=image_embeds,
+    negative_image_embeds=negative_image_embeds,
+    height=768,
+    width=768,
+    strength=0.3,
+)
+
+out.images[0].save("fantasy_land.png")
+```
+
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/img2img_fantasyland.png)
+
+
+### Text Guided Inpainting Generation
+
+You can use [`KandinskyInpaintPipeline`] to edit images. In this example, we will add a hat to the portrait of a cat.
+
+```py
+from diffusers import KandinskyInpaintPipeline, KandinskyPriorPipeline
+from diffusers.utils import load_image
+import torch
+import numpy as np
+
+pipe_prior = KandinskyPriorPipeline.from_pretrained(
+    "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16
+)
+pipe_prior.to("cuda")
+
+prompt = "a hat"
+prior_output = pipe_prior(prompt)
+
+pipe = KandinskyInpaintPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-inpaint", torch_dtype=torch.float16)
+pipe.to("cuda")
+
+init_image = load_image(
+    "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/kandinsky/cat.png"
+)
+
+mask = np.ones((768, 768), dtype=np.float32)
+# Let's mask out an area above the cat's head
+mask[:250, 250:-250] = 0
+
+out = pipe(
+    prompt,
+    image=init_image,
+    mask_image=mask,
+    **prior_output,
+    height=768,
+    width=768,
+    num_inference_steps=150,
+)
+
+image = out.images[0]
+image.save("cat_with_hat.png")
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/inpaint_cat_hat.png)
+
+### Interpolate 
+
+The [`KandinskyPriorPipeline`] also comes with a cool utility function that will allow you to interpolate the latent space of different images and texts super easily. Here is an example of how you can create an Impressionist-style portrait for your pet based on "The Starry Night". 
+
+Note that you can interpolate between texts and images - in the below example, we passed a text prompt "a cat" and two images to the `interplate` function, along with a `weights` variable containing the corresponding weights for each condition we interplate. 
+
+```python
+from diffusers import KandinskyPriorPipeline, KandinskyPipeline
+from diffusers.utils import load_image
+import PIL
+
+import torch
+
+pipe_prior = KandinskyPriorPipeline.from_pretrained(
+    "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16
+)
+pipe_prior.to("cuda")
+
+img1 = load_image(
+    "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/kandinsky/cat.png"
+)
+
+img2 = load_image(
+    "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main" "/kandinsky/starry_night.jpeg"
+)
+
+# add all the conditions we want to interpolate, can be either text or image
+images_texts = ["a cat", img1, img2]
+
+# specify the weights for each condition in images_texts
+weights = [0.3, 0.3, 0.4]
+
+# We can leave the prompt empty
+prompt = ""
+prior_out = pipe_prior.interpolate(images_texts, weights)
+
+pipe = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+pipe.to("cuda")
+
+image = pipe(prompt, **prior_out, height=768, width=768).images[0]
+
+image.save("starry_cat.png")
+```
+![img](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/starry_cat.png)
+
+
+## Optimization
+
+Running Kandinsky in inference requires running both a first prior pipeline: [`KandinskyPriorPipeline`]
+and a second image decoding pipeline which is one of [`KandinskyPipeline`], [`KandinskyImg2ImgPipeline`], or [`KandinskyInpaintPipeline`].
+
+The bulk of the computation time will always be the second image decoding pipeline, so when looking 
+into optimizing the model, one should look into the second image decoding pipeline.
+
+When running with PyTorch < 2.0, we strongly recommend making use of [`xformers`](https://github.com/facebookresearch/xformers)
+to speed-up the optimization. This can be done by simply running:
+
+```py
+from diffusers import DiffusionPipeline
+import torch
+
+t2i_pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+t2i_pipe.enable_xformers_memory_efficient_attention()
+```
+
+When running on PyTorch >= 2.0, PyTorch's SDPA attention will automatically be used. For more information on 
+PyTorch's SDPA, feel free to have a look at [this blog post](https://pytorch.org/blog/accelerated-diffusers-pt-20/).
+
+To have explicit control , you can also manually set the pipeline to use PyTorch's 2.0 efficient attention:
+
+```py
+from diffusers.models.attention_processor import AttnAddedKVProcessor2_0
+
+t2i_pipe.unet.set_attn_processor(AttnAddedKVProcessor2_0())
+```
+
+The slowest and most memory intense attention processor is the default `AttnAddedKVProcessor` processor.
+We do **not** recommend using it except for testing purposes or cases where very high determistic behaviour is desired. 
+You can set it with:
+
+```py
+from diffusers.models.attention_processor import AttnAddedKVProcessor
+
+t2i_pipe.unet.set_attn_processor(AttnAddedKVProcessor())
+```
+
+With PyTorch >= 2.0, you can also use Kandinsky with `torch.compile` which depending 
+on your hardware can signficantly speed-up your inference time once the model is compiled.
+To use Kandinsksy with `torch.compile`, you can do:
+
+```py
+t2i_pipe.unet.to(memory_format=torch.channels_last)
+t2i_pipe.unet = torch.compile(t2i_pipe.unet, mode="reduce-overhead", fullgraph=True)
+```
+
+After compilation you should see a very fast inference time. For more information,
+feel free to have a look at [Our PyTorch 2.0 benchmark](https://huggingface.co/docs/diffusers/main/en/optimization/torch2.0).
+
+
+
+
+
+## KandinskyPriorPipeline
+
+[[autodoc]] KandinskyPriorPipeline
+	- all
+	- __call__
+	- interpolate
+	
+## KandinskyPipeline
+
+[[autodoc]] KandinskyPipeline
+	- all
+	- __call__
+
+## KandinskyImg2ImgPipeline
+
+[[autodoc]] KandinskyImg2ImgPipeline
+	- all
+	- __call__
+
+## KandinskyInpaintPipeline
+
+[[autodoc]] KandinskyInpaintPipeline
+	- all
+	- __call__

docs/source/en/api/pipelines/overview.mdx

@@ -54,10 +54,14 @@ available a colab notebook to directly try them out.
 | [if](./if) | [**IF**](https://github.com/deep-floyd/IF) | Image Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/deepfloyd_if_free_tier_google_colab.ipynb)
 | [if_img2img](./if) | [**IF**](https://github.com/deep-floyd/IF) | Image-to-Image Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/deepfloyd_if_free_tier_google_colab.ipynb)
 | [if_inpainting](./if) | [**IF**](https://github.com/deep-floyd/IF) | Image-to-Image Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/deepfloyd_if_free_tier_google_colab.ipynb)
+| [kandinsky](./kandinsky) | **Kandinsky** | Text-to-Image Generation | 
+| [kandinsky_inpaint](./kandinsky) | **Kandinsky** | Image-to-Image Generation | 
+| [kandinsky_img2img](./kandinsky) | **Kandinsksy** | Image-to-Image Generation |
 | [latent_diffusion](./latent_diffusion) | [**High-Resolution Image Synthesis with Latent Diffusion Models**](https://arxiv.org/abs/2112.10752)| Text-to-Image Generation | 
 | [latent_diffusion](./latent_diffusion) | [**High-Resolution Image Synthesis with Latent Diffusion Models**](https://arxiv.org/abs/2112.10752)| Super Resolution Image-to-Image | 
 | [latent_diffusion_uncond](./latent_diffusion_uncond) | [**High-Resolution Image Synthesis with Latent Diffusion Models**](https://arxiv.org/abs/2112.10752) | Unconditional Image Generation | 
 | [paint_by_example](./paint_by_example) | [**Paint by Example: Exemplar-based Image Editing with Diffusion Models**](https://arxiv.org/abs/2211.13227) | Image-Guided Image Inpainting | 
+| [paradigms](./paradigms) | [**Parallel Sampling of Diffusion Models**](https://arxiv.org/abs/2305.16317) | Text-to-Image Generation |
 | [pndm](./pndm) | [**Pseudo Numerical Methods for Diffusion Models on Manifolds**](https://arxiv.org/abs/2202.09778) | Unconditional Image Generation | 
 | [score_sde_ve](./score_sde_ve) | [**Score-Based Generative Modeling through Stochastic Differential Equations**](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation | 
 | [score_sde_vp](./score_sde_vp) | [**Score-Based Generative Modeling through Stochastic Differential Equations**](https://openreview.net/forum?id=PxTIG12RRHS) | Unconditional Image Generation | 
@@ -72,21 +76,20 @@ available a colab notebook to directly try them out.
 | [stable_diffusion_self_attention_guidance](./stable_diffusion/self_attention_guidance) | [**Self-Attention Guidance**](https://arxiv.org/abs/2210.00939) | Text-to-Image Generation |
 | [stable_diffusion_image_variation](./stable_diffusion/image_variation) | [**Stable Diffusion Image Variations**](https://github.com/LambdaLabsML/lambda-diffusers#stable-diffusion-image-variations) | Image-to-Image Generation |
 | [stable_diffusion_latent_upscale](./stable_diffusion/latent_upscale) | [**Stable Diffusion Latent Upscaler**](https://twitter.com/StabilityAI/status/1590531958815064065) | Text-Guided Super Resolution Image-to-Image |
-| [stable_diffusion_2](./stable_diffusion_2/) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Text-to-Image Generation | 
-| [stable_diffusion_2](./stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Text-Guided Image Inpainting | 
-| [stable_diffusion_2](./stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Depth-to-Image Text-Guided Generation |
-| [stable_diffusion_2](./stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Text-Guided Super Resolution Image-to-Image |
+| [stable_diffusion_2](./stable_diffusion/stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Text-Guided Image Inpainting | 
+| [stable_diffusion_2](./stable_diffusion/stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Depth-to-Image Text-Guided Generation |
+| [stable_diffusion_2](./stable_diffusion/stable_diffusion_2) | [**Stable Diffusion 2**](https://stability.ai/blog/stable-diffusion-v2-release) | Text-Guided Super Resolution Image-to-Image |
 | [stable_diffusion_safe](./stable_diffusion_safe) | [**Safe Stable Diffusion**](https://arxiv.org/abs/2211.05105) | Text-Guided Generation | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ml-research/safe-latent-diffusion/blob/main/examples/Safe%20Latent%20Diffusion.ipynb)
 | [stable_unclip](./stable_unclip) | **Stable unCLIP** | Text-to-Image Generation |
 | [stable_unclip](./stable_unclip) | **Stable unCLIP** | Image-to-Image Text-Guided Generation |
 | [stochastic_karras_ve](./stochastic_karras_ve) | [**Elucidating the Design Space of Diffusion-Based Generative Models**](https://arxiv.org/abs/2206.00364) | Unconditional Image Generation |
-| [text_to_video_sd](./api/pipelines/text_to_video) | [Modelscope's Text-to-video-synthesis Model in Open Domain](https://modelscope.cn/models/damo/text-to-video-synthesis/summary) | Text-to-Video Generation |
-| [unclip](./unclip) | [Hierarchical Text-Conditional Image Generation with CLIP Latents](https://arxiv.org/abs/2204.06125) | Text-to-Image Generation |
-| [versatile_diffusion](./versatile_diffusion) | [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://arxiv.org/abs/2211.08332) | Text-to-Image Generation | 
-| [versatile_diffusion](./versatile_diffusion) | [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://arxiv.org/abs/2211.08332) | Image Variations Generation | 
-| [versatile_diffusion](./versatile_diffusion) | [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://arxiv.org/abs/2211.08332) | Dual Image and Text Guided Generation | 
-| [vq_diffusion](./vq_diffusion) | [Vector Quantized Diffusion Model for Text-to-Image Synthesis](https://arxiv.org/abs/2111.14822) | Text-to-Image Generation | 
-| [text_to_video_zero](./text_to_video_zero) | [Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators](https://arxiv.org/abs/2303.13439) | Text-to-Video Generation |
+| [text_to_video_sd](./api/pipelines/text_to_video) | [**Modelscope's Text-to-video-synthesis Model in Open Domain**](https://modelscope.cn/models/damo/text-to-video-synthesis/summary) | Text-to-Video Generation |
+| [unclip](./unclip) | [**Hierarchical Text-Conditional Image Generation with CLIP Latents](https://arxiv.org/abs/2204.06125) | Text-to-Image Generation |
+| [versatile_diffusion](./versatile_diffusion) | [**Versatile Diffusion: Text, Images and Variations All in One Diffusion Model**](https://arxiv.org/abs/2211.08332) | Text-to-Image Generation | 
+| [versatile_diffusion](./versatile_diffusion) | [**Versatile Diffusion: Text, Images and Variations All in One Diffusion Model**](https://arxiv.org/abs/2211.08332) | Image Variations Generation | 
+| [versatile_diffusion](./versatile_diffusion) | [**Versatile Diffusion: Text, Images and Variations All in One Diffusion Model**](https://arxiv.org/abs/2211.08332) | Dual Image and Text Guided Generation | 
+| [vq_diffusion](./vq_diffusion) | [**Vector Quantized Diffusion Model for Text-to-Image Synthesis**](https://arxiv.org/abs/2111.14822) | Text-to-Image Generation | 
+| [text_to_video_zero](./text_to_video_zero) | [**Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators**](https://arxiv.org/abs/2303.13439) | Text-to-Video Generation |
 
 
 **Note**: Pipelines are simple examples of how to play around with the diffusion systems as described in the corresponding papers. 
@@ -113,105 +116,3 @@ each pipeline, one should look directly into the respective pipeline.
 
 **Note**: All pipelines have PyTorch's autograd disabled by decorating the `__call__` method with a [`torch.no_grad`](https://pytorch.org/docs/stable/generated/torch.no_grad.html) decorator because pipelines should
 not be used for training. If you want to store the gradients during the forward pass, we recommend writing your own pipeline, see also our [community-examples](https://github.com/huggingface/diffusers/tree/main/examples/community).
-
-## Contribution
-
-We are more than happy about any contribution to the officially supported pipelines 🤗. We aspire
-all of our pipelines to be  **self-contained**, **easy-to-tweak**, **beginner-friendly** and for **one-purpose-only**.
-
-- **Self-contained**: A pipeline shall be as self-contained as possible. More specifically, this means that all functionality should be either directly defined in the pipeline file itself, should be inherited from (and only from) the [`DiffusionPipeline` class](.../diffusion_pipeline) or be directly attached to the model and scheduler components of the pipeline. 
-- **Easy-to-use**: Pipelines should be extremely easy to use - one should be able to load the pipeline and 
-use it for its designated task, *e.g.* text-to-image generation, in just a couple of lines of code. Most 
-logic including pre-processing, an unrolled diffusion loop, and post-processing should all happen inside the `__call__` method.
-- **Easy-to-tweak**: Certain pipelines will not be able to handle all use cases and tasks that you might like them to. If you want to use a certain pipeline for a specific use case that is not yet supported, you might have to copy the pipeline file and tweak the code to your needs. We try to make the pipeline code as readable as possible so that each part –from pre-processing to diffusing to post-processing– can easily be adapted. If you would like the community to benefit from your customized pipeline, we would love to see a contribution to our [community-examples](https://github.com/huggingface/diffusers/tree/main/examples/community). If you feel that an important pipeline should be part of the official pipelines but isn't, a contribution to the [official pipelines](./overview) would be even better.
-- **One-purpose-only**: Pipelines should be used for one task and one task only. Even if two tasks are very similar from a modeling point of view, *e.g.* image2image translation and in-painting, pipelines shall be used for one task only to keep them *easy-to-tweak* and *readable*.
-
-## Examples
-
-### Text-to-Image generation with Stable Diffusion
-
-```python
-# make sure you're logged in with `huggingface-cli login`
-from diffusers import StableDiffusionPipeline, LMSDiscreteScheduler
-
-pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
-pipe = pipe.to("cuda")
-
-prompt = "a photo of an astronaut riding a horse on mars"
-image = pipe(prompt).images[0]
-
-image.save("astronaut_rides_horse.png")
-```
-
-### Image-to-Image text-guided generation with Stable Diffusion
-
-The `StableDiffusionImg2ImgPipeline` lets you pass a text prompt and an initial image to condition the generation of new images.
-
-```python
-import requests
-from PIL import Image
-from io import BytesIO
-
-from diffusers import StableDiffusionImg2ImgPipeline
-
-# load the pipeline
-device = "cuda"
-pipe = StableDiffusionImg2ImgPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16).to(
-    device
-)
-
-# let's download an initial image
-url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
-
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-init_image = init_image.resize((768, 512))
-
-prompt = "A fantasy landscape, trending on artstation"
-
-images = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5).images
-
-images[0].save("fantasy_landscape.png")
-```
-You can also run this example on colab [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/image_2_image_using_diffusers.ipynb)
-
-### Tweak prompts reusing seeds and latents
-
-You can generate your own latents to reproduce results, or tweak your prompt on a specific result you liked. [This notebook](https://github.com/pcuenca/diffusers-examples/blob/main/notebooks/stable-diffusion-seeds.ipynb) shows how to do it step by step. You can also run it in Google Colab [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/pcuenca/diffusers-examples/blob/main/notebooks/stable-diffusion-seeds.ipynb)
-
-
-### In-painting using Stable Diffusion
-
-The `StableDiffusionInpaintPipeline` lets you edit specific parts of an image by providing a mask and text prompt.
-
-```python
-import PIL
-import requests
-import torch
-from io import BytesIO
-
-from diffusers import StableDiffusionInpaintPipeline
-
-
-def download_image(url):
-    response = requests.get(url)
-    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
-
-
-img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
-mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
-
-init_image = download_image(img_url).resize((512, 512))
-mask_image = download_image(mask_url).resize((512, 512))
-
-pipe = StableDiffusionInpaintPipeline.from_pretrained(
-    "runwayml/stable-diffusion-inpainting",
-    torch_dtype=torch.float16,
-)
-pipe = pipe.to("cuda")
-
-prompt = "Face of a yellow cat, high resolution, sitting on a park bench"
-image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
-```
-
-You can also run this example on colab [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/in_painting_with_stable_diffusion_using_diffusers.ipynb)

docs/source/en/api/pipelines/panorama.mdx

@@ -52,6 +52,14 @@ image = pipe(prompt).images[0]
 image.save("dolomites.png")
 ```
 
+<Tip>
+
+While calling this pipeline, it's possible to specify the `view_batch_size` to have a >1 value. 
+For some GPUs with high performance, higher a `view_batch_size`, can speedup the generation
+and increase the VRAM usage.
+
+</Tip>
+
 ## StableDiffusionPanoramaPipeline
 [[autodoc]] StableDiffusionPanoramaPipeline
 	- __call__

docs/source/en/api/pipelines/paradigms.mdx

@@ -0,0 +1,83 @@
+<!--Copyright 2023 ParaDiGMS authors and The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# Parallel Sampling of Diffusion Models (ParaDiGMS)
+
+## Overview
+
+[Parallel Sampling of Diffusion Models](https://arxiv.org/abs/2305.16317) by Andy Shih, Suneel Belkhale, Stefano Ermon, Dorsa Sadigh, Nima Anari.
+
+The abstract of the paper is the following:
+
+*Diffusion models are powerful generative models but suffer from slow sampling, often taking 1000 sequential denoising steps for one sample. As a result, considerable efforts have been directed toward reducing the number of denoising steps, but these methods hurt sample quality. Instead of reducing the number of denoising steps (trading quality for speed), in this paper we explore an orthogonal approach: can we run the denoising steps in parallel (trading compute for speed)? In spite of the sequential nature of the denoising steps, we show that surprisingly it is possible to parallelize sampling via Picard iterations, by guessing the solution of future denoising steps and iteratively refining until convergence. With this insight, we present ParaDiGMS, a novel method to accelerate the sampling of pretrained diffusion models by denoising multiple steps in parallel. ParaDiGMS is the first diffusion sampling method that enables trading compute for speed and is even compatible with existing fast sampling techniques such as DDIM and DPMSolver. Using ParaDiGMS, we improve sampling speed by 2-4x across a range of robotics and image generation models, giving state-of-the-art sampling speeds of 0.2s on 100-step DiffusionPolicy and 16s on 1000-step StableDiffusion-v2 with no measurable degradation of task reward, FID score, or CLIP score.*
+
+Resources:
+
+* [Paper](https://arxiv.org/abs/2305.16317).
+* [Original Code](https://github.com/AndyShih12/paradigms).
+
+## Available Pipelines:
+
+| Pipeline | Tasks | Demo
+|---|---|:---:|
+| [StableDiffusionParadigmsPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_paradigms.py) | *Faster Text-to-Image Generation* | |
+
+This pipeline was contributed by [`AndyShih12`](https://github.com/AndyShih12) in this [PR](https://github.com/huggingface/diffusers/pull/3716/).
+
+## Usage example
+
+```python
+import torch
+from diffusers import DDPMParallelScheduler
+from diffusers import StableDiffusionParadigmsPipeline
+
+scheduler = DDPMParallelScheduler.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="scheduler")
+
+pipe = StableDiffusionParadigmsPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5", scheduler=scheduler, torch_dtype=torch.float16
+)
+pipe = pipe.to("cuda")
+
+ngpu, batch_per_device = torch.cuda.device_count(), 5
+pipe.wrapped_unet = torch.nn.DataParallel(pipe.unet, device_ids=[d for d in range(ngpu)])
+
+prompt = "a photo of an astronaut riding a horse on mars"
+image = pipe(prompt, parallel=ngpu * batch_per_device, num_inference_steps=1000).images[0]
+```
+
+<Tip>
+This pipeline improves sampling speed by running denoising steps in parallel, at the cost of increased total FLOPs.
+Therefore, it is better to call this pipeline when running on multiple GPUs. Otherwise, without enough GPU bandwidth
+sampling may be even slower than sequential sampling.
+
+The two parameters to play with are `parallel` (batch size) and `tolerance`. 
+- If it fits in memory, for 1000-step DDPM you can aim for a batch size of around 100 
+(e.g. 8 GPUs and batch_per_device=12 to get parallel=96). Higher batch size
+may not fit in memory, and lower batch size gives less parallelism. 
+- For tolerance, using a higher tolerance may get better speedups but can risk sample quality degradation. 
+If there is quality degradation with the default tolerance, then use a lower tolerance (e.g. 0.001).
+
+For 1000-step DDPM on 8 A100 GPUs, you can expect around a 3x speedup by StableDiffusionParadigmsPipeline instead of StableDiffusionPipeline
+by setting parallel=80 and tolerance=0.1.
+</Tip>
+
+<Tip>
+Diffusers also offers distributed inference support for generating multiple prompts
+in parallel on multiple GPUs. Check out the docs [here](https://huggingface.co/docs/diffusers/main/en/training/distributed_inference).
+
+In contrast, this pipeline is designed for speeding up sampling of a single prompt (by using multiple GPUs).
+</Tip>
+
+## StableDiffusionParadigmsPipeline
+[[autodoc]] StableDiffusionParadigmsPipeline
+	- __call__
+	- all

docs/source/en/api/pipelines/stable_diffusion/ldm3d_diffusion.mdx

@@ -0,0 +1,55 @@
+<!--Copyright 2023 The Intel Labs Team Authors and HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# LDM3D
+
+LDM3D was proposed in [LDM3D: Latent Diffusion Model for 3D](https://arxiv.org/abs/2305.10853) by Gabriela Ben Melech Stan, Diana Wofk, Scottie Fox, Alex Redden, Will Saxton, Jean Yu, Estelle Aflalo, Shao-Yen Tseng, Fabio Nonato, Matthias Muller, Vasudev Lal
+The abstract of the paper is the following:
+
+*This research paper proposes a Latent Diffusion Model for 3D (LDM3D) that generates both image and depth map data from a given text prompt, allowing users to generate RGBD images from text prompts. The LDM3D model is fine-tuned on a dataset of tuples containing an RGB image, depth map and caption, and validated through extensive experiments. We also develop an application called DepthFusion, which uses the generated RGB images and depth maps to create immersive and interactive 360-degree-view experiences using TouchDesigner. This technology has the potential to transform a wide range of industries, from entertainment and gaming to architecture and design. Overall, this paper presents a significant contribution to the field of generative AI and computer vision, and showcases the potential of LDM3D and DepthFusion to revolutionize content creation and digital experiences. A short video summarizing the approach can be found at [this url](https://t.ly/tdi2).*
+
+
+*Overview*:
+
+| Pipeline | Tasks | Colab | Demo
+|---|---|:---:|:---:|
+| [pipeline_stable_diffusion_ldm3d.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_ldm3d.py) | *Text-to-Image Generation* | - | -
+
+## Tips
+
+- LDM3D generates both an image and a depth map from a given text prompt, compared to the existing txt-to-img diffusion models such as [Stable Diffusion](./stable_diffusion/overview) that generates only an image.
+- With almost the same number of parameters, LDM3D achieves to create a latent space that can compress both the RGB images and the depth maps. 
+
+
+Running LDM3D is straighforward with the [`StableDiffusionLDM3DPipeline`]:
+
+```python
+>>> from diffusers import StableDiffusionLDM3DPipeline
+
+>>> pipe = StableDiffusionLDM3DPipeline.from_pretrained("Intel/ldm3d")
+prompt ="A picture of some lemons on a table"
+output = pipe(prompt)
+rgb_image, depth_image = output.rgb, output.depth
+rgb_image[0].save("lemons_ldm3d_rgb.jpg")
+depth_image[0].save("lemons_ldm3d_depth.png")
+```
+
+
+## StableDiffusionPipelineOutput
+[[autodoc]] pipelines.stable_diffusion.StableDiffusionPipelineOutput
+	- all
+	- __call__
+
+## StableDiffusionLDM3DPipeline
+[[autodoc]] StableDiffusionLDM3DPipeline
+	- all
+	- __call__

docs/source/en/api/pipelines/stable_diffusion/overview.mdx

@@ -26,19 +26,17 @@ For more details about how Stable Diffusion works and how it differs from the ba
 | Pipeline | Tasks | Colab | Demo
 |---|---|:---:|:---:|
 | [StableDiffusionPipeline](./text2img) | *Text-to-Image Generation* | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_diffusion.ipynb) | [🤗 Stable Diffusion](https://huggingface.co/spaces/stabilityai/stable-diffusion)
+| [StableDiffusionPipelineSafe](./stable_diffusion_safe) | *Text-to-Image Generation* | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/ml-research/safe-latent-diffusion/blob/main/examples/Safe%20Latent%20Diffusion.ipynb) | [![Huggingface Spaces](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Spaces-blue)](https://huggingface.co/spaces/AIML-TUDA/unsafe-vs-safe-stable-diffusion)
 | [StableDiffusionImg2ImgPipeline](./img2img) | *Image-to-Image Text-Guided Generation* | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/image_2_image_using_diffusers.ipynb) | [🤗 Diffuse the Rest](https://huggingface.co/spaces/huggingface/diffuse-the-rest)
-| [StableDiffusionInpaintPipeline](./inpaint) | **Experimental** – *Text-Guided Image Inpainting* | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/in_painting_with_stable_diffusion_using_diffusers.ipynb) | Coming soon
-| [StableDiffusionDepth2ImgPipeline](./depth2img) | **Experimental** – *Depth-to-Image Text-Guided Generation * | | Coming soon
-| [StableDiffusionImageVariationPipeline](./image_variation) | **Experimental** – *Image Variation Generation * | | [🤗 Stable Diffusion Image Variations](https://huggingface.co/spaces/lambdalabs/stable-diffusion-image-variations)
-| [StableDiffusionUpscalePipeline](./upscale) | **Experimental** – *Text-Guided Image Super-Resolution * | | Coming soon
-| [StableDiffusionLatentUpscalePipeline](./latent_upscale) | **Experimental** – *Text-Guided Image Super-Resolution * | | Coming soon
-| [StableDiffusionInstructPix2PixPipeline](./pix2pix) | **Experimental** – *Text-Based Image Editing * | | [InstructPix2Pix: Learning to Follow Image Editing Instructions](https://huggingface.co/spaces/timbrooks/instruct-pix2pix)
-| [StableDiffusionAttendAndExcitePipeline](./attend_and_excite) | **Experimental** – *Text-to-Image Generation * | | [Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models](https://huggingface.co/spaces/AttendAndExcite/Attend-and-Excite)
-| [StableDiffusionPix2PixZeroPipeline](./pix2pix_zero) | **Experimental** – *Text-Based Image Editing * | | [Zero-shot Image-to-Image Translation](https://arxiv.org/abs/2302.03027)
-| [StableDiffusionModelEditingPipeline](./model_editing) | **Experimental** – *Text-to-Image Model Editing * | | [Editing Implicit Assumptions in Text-to-Image Diffusion Models](https://arxiv.org/abs/2303.08084)
-| [StableDiffusionDiffEditPipeline](./diffedit) | **Experimental** – *Text-Based Image Editing * | | [DiffEdit: Diffusion-based semantic image editing with mask guidance](https://arxiv.org/abs/2210.11427)
-
-
+| [StableDiffusionInpaintPipeline](./inpaint) | **Experimental** – *Text-Guided Image Inpainting* | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/in_painting_with_stable_diffusion_using_diffusers.ipynb) | 
+| [StableDiffusionDepth2ImgPipeline](./depth2img) | **Experimental** – *Depth-to-Image Text-Guided Generation* | |
+| [StableDiffusionImageVariationPipeline](./image_variation) | **Experimental** – *Image Variation Generation* | | [🤗 Stable Diffusion Image Variations](https://huggingface.co/spaces/lambdalabs/stable-diffusion-image-variations)
+| [StableDiffusionUpscalePipeline](./upscale) | **Experimental** – *Text-Guided Image Super-Resolution* | |
+| [StableDiffusionLatentUpscalePipeline](./latent_upscale) | **Experimental** – *Text-Guided Image Super-Resolution* | |
+| [Stable Diffusion 2](./stable_diffusion_2)  | *Text-Guided Image Inpainting* | 
+| [Stable Diffusion 2](./stable_diffusion_2)  |  *Depth-to-Image Text-Guided Generation* |
+| [Stable Diffusion 2](./stable_diffusion_2)  | *Text-Guided Super Resolution Image-to-Image* |
+| [StableDiffusionLDM3DPipeline](./ldm3d)  | *Text-to-(RGB, Depth)* |
 
 ## Tips
 

docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_2.mdx

@@ -71,6 +71,64 @@ image = pipe(prompt, guidance_scale=9, num_inference_steps=25).images[0]
 image.save("astronaut.png")
 ```
 
+#### Experimental: "Common Diffusion Noise Schedules and Sample Steps are Flawed": 
+
+The paper **[Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/abs/2305.08891)** 
+claims that a mismatch between the training and inference settings leads to suboptimal inference generation results for Stable Diffusion.
+
+The abstract reads as follows:
+
+*We discover that common diffusion noise schedules do not enforce the last timestep to have zero signal-to-noise ratio (SNR),
+and some implementations of diffusion samplers do not start from the last timestep.
+Such designs are flawed and do not reflect the fact that the model is given pure Gaussian noise at inference, creating a discrepancy between training and inference.
+We show that the flawed design causes real problems in existing implementations. 
+In Stable Diffusion, it severely limits the model to only generate images with medium brightness and 
+prevents it from generating very bright and dark samples. We propose a few simple fixes: 
+- (1) rescale the noise schedule to enforce zero terminal SNR; 
+- (2) train the model with v prediction; 
+- (3) change the sampler to always start from the last timestep; 
+- (4) rescale classifier-free guidance to prevent over-exposure. 
+These simple changes ensure the diffusion process is congruent between training and inference and 
+allow the model to generate samples more faithful to the original data distribution.*
+
+You can apply all of these changes in `diffusers` when using [`DDIMScheduler`]:
+- (1) rescale the noise schedule to enforce zero terminal SNR; 
+```py
+pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, rescale_betas_zero_snr=True)
+```
+- (2) train the model with v prediction; 
+Continue fine-tuning a checkpoint with [`train_text_to_image.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image.py) or [`train_text_to_image_lora.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_lora.py)
+and `--prediction_type="v_prediction"`.
+- (3) change the sampler to always start from the last timestep; 
+```py
+pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
+```
+- (4) rescale classifier-free guidance to prevent over-exposure. 
+```py
+pipe(..., guidance_rescale=0.7)
+```
+
+An example is to use [this checkpoint](https://huggingface.co/ptx0/pseudo-journey-v2) 
+which has been fine-tuned using the `"v_prediction"`.
+
+The checkpoint can then be run in inference as follows:
+
+```py
+from diffusers import DiffusionPipeline, DDIMScheduler
+
+pipe = DiffusionPipeline.from_pretrained("ptx0/pseudo-journey-v2", torch_dtype=torch.float16)
+pipe.scheduler = DDIMScheduler.from_config(
+    pipe.scheduler.config, rescale_betas_zero_snr=True, timestep_spacing="trailing"
+)
+pipe.to("cuda")
+
+prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intricate detail, octane render, 8k"
+image = pipeline(prompt, guidance_rescale=0.7).images[0]
+```
+
+## DDIMScheduler
+[[autodoc]] DDIMScheduler
+
 ### Image Inpainting
 
 - *Image Inpainting (512x512 resolution)*: [stabilityai/stable-diffusion-2-inpainting](https://huggingface.co/stabilityai/stable-diffusion-2-inpainting) with [`StableDiffusionInpaintPipeline`]

docs/source/en/api/pipelines/text_to_video.mdx

@@ -37,9 +37,12 @@ Resources:
 | Pipeline | Tasks | Demo
 |---|---|:---:|
 | [TextToVideoSDPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py) | *Text-to-Video Generation* | [🤗 Spaces](https://huggingface.co/spaces/damo-vilab/modelscope-text-to-video-synthesis)
+| [VideoToVideoSDPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py) | *Text-Guided Video-to-Video Generation* | [(TODO)🤗 Spaces]()
 
 ## Usage example 
 
+### `text-to-video-ms-1.7b`
+
 Let's start by generating a short video with the default length of 16 frames (2s at 8 fps):
 
 ```python 
@@ -119,12 +122,72 @@ Here are some sample outputs:
     </tr>
 </table>
 
+### `cerspense/zeroscope_v2_576w` & `cerspense/zeroscope_v2_XL`
+
+Zeroscope are watermark-free model and have been trained on specific sizes such as `576x320` and `1024x576`.
+One should first generate a video using the lower resolution checkpoint [`cerspense/zeroscope_v2_576w`](https://huggingface.co/cerspense/zeroscope_v2_576w) with [`TextToVideoSDPipeline`],
+which can then be upscaled using [`VideoToVideoSDPipeline`] and [`cerspense/zeroscope_v2_XL`](https://huggingface.co/cerspense/zeroscope_v2_XL).
+
+
+```py
+import torch
+from diffusers import DiffusionPipeline
+from diffusers.utils import export_to_video
+
+pipe = DiffusionPipeline.from_pretrained("cerspense/zeroscope_v2_576w", torch_dtype=torch.float16)
+pipe.enable_model_cpu_offload()
+
+# memory optimization
+pipe.enable_vae_slicing()
+
+prompt = "Darth Vader surfing a wave"
+video_frames = pipe(prompt, num_frames=24).frames
+video_path = export_to_video(video_frames)
+video_path
+```
+
+Now the video can be upscaled:
+
+```py
+pipe = DiffusionPipeline.from_pretrained("cerspense/zeroscope_v2_XL", torch_dtype=torch.float16)
+pipe.vae.enable_slicing()
+pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+pipe.enable_model_cpu_offload()
+
+video = [Image.fromarray(frame).resize((1024, 576)) for frame in video_frames]
+
+video_frames = pipe(prompt, video=video, strength=0.6).frames
+video_path = export_to_video(video_frames)
+video_path
+```
+
+Here are some sample outputs: 
+
+<table>
+    <tr>
+        <td ><center>
+        Darth vader surfing in waves.
+        <br>
+        <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/darthvader_cerpense.gif"
+            alt="Darth vader surfing in waves."
+            style="width: 576px;" />
+        </center></td>
+    </tr>
+</table>
+
 ## Available checkpoints 
 
 * [damo-vilab/text-to-video-ms-1.7b](https://huggingface.co/damo-vilab/text-to-video-ms-1.7b/)
 * [damo-vilab/text-to-video-ms-1.7b-legacy](https://huggingface.co/damo-vilab/text-to-video-ms-1.7b-legacy)
+* [cerspense/zeroscope_v2_576w](https://huggingface.co/cerspense/zeroscope_v2_576w)
+* [cerspense/zeroscope_v2_XL](https://huggingface.co/cerspense/zeroscope_v2_XL)
 
 ## TextToVideoSDPipeline
 [[autodoc]] TextToVideoSDPipeline
 	- all
 	- __call__
+
+## VideoToVideoSDPipeline
+[[autodoc]] VideoToVideoSDPipeline
+	- all
+	- __call__

docs/source/en/api/pipelines/text_to_video_zero.mdx

@@ -80,6 +80,41 @@ You can change these parameters in the pipeline call:
 * Video length:
     * `video_length`, the number of frames video_length to be generated. Default: `video_length=8`
 
+We an also generate longer videos by doing the processing in a chunk-by-chunk manner:
+```python
+import torch
+import imageio
+from diffusers import TextToVideoZeroPipeline
+import numpy as np
+
+model_id = "runwayml/stable-diffusion-v1-5"
+pipe = TextToVideoZeroPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
+seed = 0
+video_length = 8
+chunk_size = 4
+prompt = "A panda is playing guitar on times square"
+
+# Generate the video chunk-by-chunk
+result = []
+chunk_ids = np.arange(0, video_length, chunk_size - 1)
+generator = torch.Generator(device="cuda")
+for i in range(len(chunk_ids)):
+    print(f"Processing chunk {i + 1} / {len(chunk_ids)}")
+    ch_start = chunk_ids[i]
+    ch_end = video_length if i == len(chunk_ids) - 1 else chunk_ids[i + 1]
+    # Attach the first frame for Cross Frame Attention
+    frame_ids = [0] + list(range(ch_start, ch_end))
+    # Fix the seed for the temporal consistency
+    generator.manual_seed(seed)
+    output = pipe(prompt=prompt, video_length=len(frame_ids), generator=generator, frame_ids=frame_ids)
+    result.append(output.images[1:])
+
+# Concatenate chunks and save
+result = np.concatenate(result)
+result = [(r * 255).astype("uint8") for r in result]
+imageio.mimsave("video.mp4", result, fps=4)
+```
+
 
 ### Text-To-Video with Pose Control
 To generate a video from prompt with additional pose control
@@ -202,7 +237,7 @@ can run with custom [DreamBooth](../training/dreambooth) models, as shown below
 
     reader = imageio.get_reader(video_path, "ffmpeg")
     frame_count = 8
-    video = [Image.fromarray(reader.get_data(i)) for i in range(frame_count)]
+    canny_edges = [Image.fromarray(reader.get_data(i)) for i in range(frame_count)]
     ```
 
 3. Run `StableDiffusionControlNetPipeline` with custom trained DreamBooth model
@@ -223,10 +258,10 @@ can run with custom [DreamBooth](../training/dreambooth) models, as shown below
     pipe.controlnet.set_attn_processor(CrossFrameAttnProcessor(batch_size=2))
 
     # fix latents for all frames
-    latents = torch.randn((1, 4, 64, 64), device="cuda", dtype=torch.float16).repeat(len(pose_images), 1, 1, 1)
+    latents = torch.randn((1, 4, 64, 64), device="cuda", dtype=torch.float16).repeat(len(canny_edges), 1, 1, 1)
 
     prompt = "oil painting of a beautiful girl avatar style"
-    result = pipe(prompt=[prompt] * len(pose_images), image=pose_images, latents=latents).images
+    result = pipe(prompt=[prompt] * len(canny_edges), image=canny_edges, latents=latents).images
     imageio.mimsave("video.mp4", result, fps=4)
     ```
 

docs/source/en/api/pipelines/unidiffuser.mdx

@@ -0,0 +1,204 @@
+<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# UniDiffuser
+
+The UniDiffuser model was proposed in [One Transformer Fits All Distributions in Multi-Modal Diffusion at Scale](https://arxiv.org/abs/2303.06555) by Fan Bao, Shen Nie, Kaiwen Xue, Chongxuan Li, Shi Pu, Yaole Wang, Gang Yue, Yue Cao, Hang Su, Jun Zhu.
+
+The abstract of the [paper](https://arxiv.org/abs/2303.06555) is the following:
+
+*This paper proposes a unified diffusion framework (dubbed UniDiffuser) to fit all distributions relevant to a set of multi-modal data in one model. Our key insight is -- learning diffusion models for marginal, conditional, and joint distributions can be unified as predicting the noise in the perturbed data, where the perturbation levels (i.e. timesteps) can be different for different modalities. Inspired by the unified view, UniDiffuser learns all distributions simultaneously with a minimal modification to the original diffusion model -- perturbs data in all modalities instead of a single modality, inputs individual timesteps in different modalities, and predicts the noise of all modalities instead of a single modality. UniDiffuser is parameterized by a transformer for diffusion models to handle input types of different modalities. Implemented on large-scale paired image-text data, UniDiffuser is able to perform image, text, text-to-image, image-to-text, and image-text pair generation by setting proper timesteps without additional overhead. In particular, UniDiffuser is able to produce perceptually realistic samples in all tasks and its quantitative results (e.g., the FID and CLIP score) are not only superior to existing general-purpose models but also comparable to the bespoken models (e.g., Stable Diffusion and DALL-E 2) in representative tasks (e.g., text-to-image generation).*
+
+Resources:
+
+* [Paper](https://arxiv.org/abs/2303.06555).
+* [Original Code](https://github.com/thu-ml/unidiffuser).
+
+Available Checkpoints are:
+- *UniDiffuser-v0 (512x512 resolution)* [thu-ml/unidiffuser-v0](https://huggingface.co/thu-ml/unidiffuser-v0)
+- *UniDiffuser-v1 (512x512 resolution)* [thu-ml/unidiffuser-v1](https://huggingface.co/thu-ml/unidiffuser-v1)
+
+This pipeline was contributed by our community member [dg845](https://github.com/dg845).
+
+## Available Pipelines:
+
+| Pipeline | Tasks | Demo | Colab |
+|:---:|:---:|:---:|:---:|
+| [UniDiffuserPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_unidiffuser.py) | *Joint Image-Text Gen*, *Text-to-Image*, *Image-to-Text*,<br> *Image Gen*, *Text Gen*, *Image Variation*, *Text Variation* | [🤗 Spaces](https://huggingface.co/spaces/thu-ml/unidiffuser) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/unidiffuser.ipynb) |
+
+## Usage Examples
+
+Because the UniDiffuser model is trained to model the joint distribution of (image, text) pairs, it is capable of performing a diverse range of generation tasks.
+
+### Unconditional Image and Text Generation
+
+Unconditional generation (where we start from only latents sampled from a standard Gaussian prior) from a [`UniDiffuserPipeline`] will produce a (image, text) pair:
+
+```python
+import torch
+
+from diffusers import UniDiffuserPipeline
+
+device = "cuda"
+model_id_or_path = "thu-ml/unidiffuser-v1"
+pipe = UniDiffuserPipeline.from_pretrained(model_id_or_path, torch_dtype=torch.float16)
+pipe.to(device)
+
+# Unconditional image and text generation. The generation task is automatically inferred.
+sample = pipe(num_inference_steps=20, guidance_scale=8.0)
+image = sample.images[0]
+text = sample.text[0]
+image.save("unidiffuser_joint_sample_image.png")
+print(text)
+```
+
+This is also called "joint" generation in the UniDiffusers paper, since we are sampling from the joint image-text distribution.
+
+Note that the generation task is inferred from the inputs used when calling the pipeline.
+It is also possible to manually specify the unconditional generation task ("mode") manually with [`UniDiffuserPipeline.set_joint_mode`]:
+
+```python
+# Equivalent to the above.
+pipe.set_joint_mode()
+sample = pipe(num_inference_steps=20, guidance_scale=8.0)
+```
+
+When the mode is set manually, subsequent calls to the pipeline will use the set mode without attempting the infer the mode.
+You can reset the mode with [`UniDiffuserPipeline.reset_mode`], after which the pipeline will once again infer the mode.
+
+You can also generate only an image or only text (which the UniDiffuser paper calls "marginal" generation since we sample from the marginal distribution of images and text, respectively):
+
+```python
+# Unlike other generation tasks, image-only and text-only generation don't use classifier-free guidance
+# Image-only generation
+pipe.set_image_mode()
+sample_image = pipe(num_inference_steps=20).images[0]
+# Text-only generation
+pipe.set_text_mode()
+sample_text = pipe(num_inference_steps=20).text[0]
+```
+
+### Text-to-Image Generation
+
+UniDiffuser is also capable of sampling from conditional distributions; that is, the distribution of images conditioned on a text prompt or the distribution of texts conditioned on an image.
+Here is an example of sampling from the conditional image distribution (text-to-image generation or text-conditioned image generation):
+
+```python
+import torch
+
+from diffusers import UniDiffuserPipeline
+
+device = "cuda"
+model_id_or_path = "thu-ml/unidiffuser-v1"
+pipe = UniDiffuserPipeline.from_pretrained(model_id_or_path, torch_dtype=torch.float16)
+pipe.to(device)
+
+# Text-to-image generation
+prompt = "an elephant under the sea"
+
+sample = pipe(prompt=prompt, num_inference_steps=20, guidance_scale=8.0)
+t2i_image = sample.images[0]
+t2i_image.save("unidiffuser_text2img_sample_image.png")
+```
+
+The `text2img` mode requires that either an input `prompt` or `prompt_embeds` be supplied. You can set the `text2img` mode manually with [`UniDiffuserPipeline.set_text_to_image_mode`].
+
+### Image-to-Text Generation
+
+Similarly, UniDiffuser can also produce text samples given an image (image-to-text or image-conditioned text generation):
+
+```python
+import torch
+
+from diffusers import UniDiffuserPipeline
+from diffusers.utils import load_image
+
+device = "cuda"
+model_id_or_path = "thu-ml/unidiffuser-v1"
+pipe = UniDiffuserPipeline.from_pretrained(model_id_or_path, torch_dtype=torch.float16)
+pipe.to(device)
+
+# Image-to-text generation
+image_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/unidiffuser/unidiffuser_example_image.jpg"
+init_image = load_image(image_url).resize((512, 512))
+
+sample = pipe(image=init_image, num_inference_steps=20, guidance_scale=8.0)
+i2t_text = sample.text[0]
+print(i2t_text)
+```
+
+The `img2text` mode requires that an input `image` be supplied. You can set the `img2text` mode manually with [`UniDiffuserPipeline.set_image_to_text_mode`].
+
+### Image Variation
+
+The UniDiffuser authors suggest performing image variation through a "round-trip" generation method, where given an input image, we first perform an image-to-text generation, and the perform a text-to-image generation on the outputs of the first generation.
+This produces a new image which is semantically similar to the input image:
+
+```python
+import torch
+
+from diffusers import UniDiffuserPipeline
+from diffusers.utils import load_image
+
+device = "cuda"
+model_id_or_path = "thu-ml/unidiffuser-v1"
+pipe = UniDiffuserPipeline.from_pretrained(model_id_or_path, torch_dtype=torch.float16)
+pipe.to(device)
+
+# Image variation can be performed with a image-to-text generation followed by a text-to-image generation:
+# 1. Image-to-text generation
+image_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/unidiffuser/unidiffuser_example_image.jpg"
+init_image = load_image(image_url).resize((512, 512))
+
+sample = pipe(image=init_image, num_inference_steps=20, guidance_scale=8.0)
+i2t_text = sample.text[0]
+print(i2t_text)
+
+# 2. Text-to-image generation
+sample = pipe(prompt=i2t_text, num_inference_steps=20, guidance_scale=8.0)
+final_image = sample.images[0]
+final_image.save("unidiffuser_image_variation_sample.png")
+```
+
+### Text Variation
+
+
+Similarly, text variation can be performed on an input prompt with a text-to-image generation followed by a image-to-text generation:
+
+```python
+import torch
+
+from diffusers import UniDiffuserPipeline
+
+device = "cuda"
+model_id_or_path = "thu-ml/unidiffuser-v1"
+pipe = UniDiffuserPipeline.from_pretrained(model_id_or_path, torch_dtype=torch.float16)
+pipe.to(device)
+
+# Text variation can be performed with a text-to-image generation followed by a image-to-text generation:
+# 1. Text-to-image generation
+prompt = "an elephant under the sea"
+
+sample = pipe(prompt=prompt, num_inference_steps=20, guidance_scale=8.0)
+t2i_image = sample.images[0]
+t2i_image.save("unidiffuser_text2img_sample_image.png")
+
+# 2. Image-to-text generation
+sample = pipe(image=t2i_image, num_inference_steps=20, guidance_scale=8.0)
+final_prompt = sample.text[0]
+print(final_prompt)
+```
+
+## UniDiffuserPipeline
+[[autodoc]] UniDiffuserPipeline
+	- all
+	- __call__

docs/source/en/api/schedulers/ddim.mdx

@@ -18,10 +18,71 @@ specific language governing permissions and limitations under the License.
 
 The abstract of the paper is the following:
 
-Denoising diffusion probabilistic models (DDPMs) have achieved high quality image generation without adversarial training, yet they require simulating a Markov chain for many steps to produce a sample. To accelerate sampling, we present denoising diffusion implicit models (DDIMs), a more efficient class of iterative implicit probabilistic models with the same training procedure as DDPMs. In DDPMs, the generative process is defined as the reverse of a Markovian diffusion process. We construct a class of non-Markovian diffusion processes that lead to the same training objective, but whose reverse process can be much faster to sample from. We empirically demonstrate that DDIMs can produce high quality samples 10× to 50× faster in terms of wall-clock time compared to DDPMs, allow us to trade off computation for sample quality, and can perform semantically meaningful image interpolation directly in the latent space.
+*Denoising diffusion probabilistic models (DDPMs) have achieved high quality image generation without adversarial training, 
+yet they require simulating a Markov chain for many steps to produce a sample. 
+To accelerate sampling, we present denoising diffusion implicit models (DDIMs), a more efficient class of iterative implicit probabilistic models
+with the same training procedure as DDPMs. In DDPMs, the generative process is defined as the reverse of a Markovian diffusion process. 
+We construct a class of non-Markovian diffusion processes that lead to the same training objective, but whose reverse process can be much faster to sample from.
+We empirically demonstrate that DDIMs can produce high quality samples 10× to 50× faster in terms of wall-clock time compared to DDPMs, allow us to trade off 
+computation for sample quality, and can perform semantically meaningful image interpolation directly in the latent space.*
 
 The original codebase of this paper can be found here: [ermongroup/ddim](https://github.com/ermongroup/ddim).
 For questions, feel free to contact the author on [tsong.me](https://tsong.me/).
 
+### Experimental: "Common Diffusion Noise Schedules and Sample Steps are Flawed": 
+
+The paper **[Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/abs/2305.08891)** 
+claims that a mismatch between the training and inference settings leads to suboptimal inference generation results for Stable Diffusion.
+
+The abstract reads as follows:
+
+*We discover that common diffusion noise schedules do not enforce the last timestep to have zero signal-to-noise ratio (SNR),
+and some implementations of diffusion samplers do not start from the last timestep.
+Such designs are flawed and do not reflect the fact that the model is given pure Gaussian noise at inference, creating a discrepancy between training and inference.
+We show that the flawed design causes real problems in existing implementations. 
+In Stable Diffusion, it severely limits the model to only generate images with medium brightness and 
+prevents it from generating very bright and dark samples. We propose a few simple fixes: 
+- (1) rescale the noise schedule to enforce zero terminal SNR; 
+- (2) train the model with v prediction; 
+- (3) change the sampler to always start from the last timestep; 
+- (4) rescale classifier-free guidance to prevent over-exposure. 
+These simple changes ensure the diffusion process is congruent between training and inference and 
+allow the model to generate samples more faithful to the original data distribution.*
+
+You can apply all of these changes in `diffusers` when using [`DDIMScheduler`]:
+- (1) rescale the noise schedule to enforce zero terminal SNR; 
+```py
+pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, rescale_betas_zero_snr=True)
+```
+- (2) train the model with v prediction; 
+Continue fine-tuning a checkpoint with [`train_text_to_image.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image.py) or [`train_text_to_image_lora.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_lora.py)
+and `--prediction_type="v_prediction"`.
+- (3) change the sampler to always start from the last timestep; 
+```py
+pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
+```
+- (4) rescale classifier-free guidance to prevent over-exposure. 
+```py
+pipe(..., guidance_rescale=0.7)
+```
+
+An example is to use [this checkpoint](https://huggingface.co/ptx0/pseudo-journey-v2) 
+which has been fine-tuned using the `"v_prediction"`.
+
+The checkpoint can then be run in inference as follows:
+
+```py
+from diffusers import DiffusionPipeline, DDIMScheduler
+
+pipe = DiffusionPipeline.from_pretrained("ptx0/pseudo-journey-v2", torch_dtype=torch.float16)
+pipe.scheduler = DDIMScheduler.from_config(
+    pipe.scheduler.config, rescale_betas_zero_snr=True, timestep_spacing="trailing"
+)
+pipe.to("cuda")
+
+prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intricate detail, octane render, 8k"
+image = pipeline(prompt, guidance_rescale=0.7).images[0]
+```
+
 ## DDIMScheduler
 [[autodoc]] DDIMScheduler

docs/source/en/api/utilities.mdx

@@ -0,0 +1,23 @@
+# Utilities
+
+Utility and helper functions for working with 🤗 Diffusers.
+
+## randn_tensor
+
+[[autodoc]] diffusers.utils.randn_tensor
+
+## numpy_to_pil
+
+[[autodoc]] utils.pil_utils.numpy_to_pil
+
+## pt_to_pil
+
+[[autodoc]] utils.pil_utils.pt_to_pil
+
+## load_image
+
+[[autodoc]] utils.testing_utils.load_image
+
+## export_to_video
+
+[[autodoc]] utils.testing_utils.export_to_video

docs/source/en/conceptual/contribution.mdx

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 We ❤️ contributions from the open-source community! Everyone is welcome, and all types of participation –not just code– are valued and appreciated. Answering questions, helping others, reaching out, and improving the documentation are all immensely valuable to the community, so don't be afraid and get involved if you're up for it!
 
-Everyone is encouraged to start by saying 👋 in our public Discord channel. We discuss the latest trends in diffusion models, ask questions, show off personal projects, help each other with contributions, or just hang out ☕. <a href="https://Discord.gg/G7tWnz98XR"><img alt="Join us on Discord" src="https://img.shields.io/Discord/823813159592001537?color=5865F2&logo=Discord&logoColor=white"></a>
+Everyone is encouraged to start by saying 👋 in our public Discord channel. We discuss the latest trends in diffusion models, ask questions, show off personal projects, help each other with contributions, or just hang out ☕. <a href="https://Discord.gg/G7tWnz98XR"><img alt="Join us on Discord" src="https://img.shields.io/discord/823813159592001537?color=5865F2&logo=discord&logoColor=white"></a>
 
 Whichever way you choose to contribute, we strive to be part of an open, welcoming, and kind community. Please, read our [code of conduct](https://github.com/huggingface/diffusers/blob/main/CODE_OF_CONDUCT.md) and be mindful to respect it during your interactions. We also recommend you become familiar with the [ethical guidelines](https://huggingface.co/docs/diffusers/conceptual/ethical_guidelines) that guide our project and ask you to adhere to the same principles of transparency and responsibility.
 

docs/source/en/conceptual/evaluation.mdx

@@ -37,7 +37,8 @@ We cover Diffusion models with the following pipelines:
 
 ## Qualitative Evaluation
 
-Qualitative evaluation typically involves human assessment of generated images. Quality is measured across aspects such as compositionality, image-text alignment, and spatial relations. Common prompts provide a degree of uniformity for subjective metrics. DrawBench and PartiPrompts are prompt datasets used for qualitative benchmarking. DrawBench and PartiPrompts were introduced by [Imagen](https://imagen.research.google/) and [Parti](https://parti.research.google/) respectively. 
+Qualitative evaluation typically involves human assessment of generated images. Quality is measured across aspects such as compositionality, image-text alignment, and spatial relations. Common prompts provide a degree of uniformity for subjective metrics.
+DrawBench and PartiPrompts are prompt datasets used for qualitative benchmarking. DrawBench and PartiPrompts were introduced by [Imagen](https://imagen.research.google/) and [Parti](https://parti.research.google/) respectively. 
 
 From the [official Parti website](https://parti.research.google/): 
 
@@ -51,7 +52,13 @@ PartiPrompts has the following columns:
 - Category of the prompt (such as “Abstract”, “World Knowledge”, etc.)
 - Challenge reflecting the difficulty (such as “Basic”, “Complex”, “Writing & Symbols”, etc.)
 
-These benchmarks allow for side-by-side human evaluation of different image generation models. Let’s see how we can use `diffusers` on a couple of PartiPrompts. 
+These benchmarks allow for side-by-side human evaluation of different image generation models. 
+
+For this, the 🧨 Diffusers team has built **Open Parti Prompts**, which is a community-driven qualitative benchmark based on Parti Prompts to compare state-of-the-art open-source diffusion models:
+- [Open Parti Prompts Game](https://huggingface.co/spaces/OpenGenAI/open-parti-prompts): For 10 parti prompts, 4 generated images are shown and the user selects the image that suits the prompt best.
+- [Open Parti Prompts Leaderboard](https://huggingface.co/spaces/OpenGenAI/parti-prompts-leaderboard): The leaderboard comparing the currently best open-sourced diffusion models to each other.
+
+To manually compare images, let’s see how we can use `diffusers` on a couple of PartiPrompts. 
 
 Below we show some prompts sampled across different challenges: Basic, Complex, Linguistic Structures, Imagination, and Writing & Symbols. Here we are using PartiPrompts as a [dataset](https://huggingface.co/datasets/nateraw/parti-prompts).
 

docs/source/en/index.mdx

@@ -94,3 +94,4 @@ The library has three main components:
 | [versatile_diffusion](./api/pipelines/versatile_diffusion) | [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://arxiv.org/abs/2211.08332) | Image Variations Generation |
 | [versatile_diffusion](./api/pipelines/versatile_diffusion) | [Versatile Diffusion: Text, Images and Variations All in One Diffusion Model](https://arxiv.org/abs/2211.08332) | Dual Image and Text Guided Generation |
 | [vq_diffusion](./api/pipelines/vq_diffusion) | [Vector Quantized Diffusion Model for Text-to-Image Synthesis](https://arxiv.org/abs/2111.14822) | Text-to-Image Generation |
+| [stable_diffusion_ldm3d](./api/pipelines/stable_diffusion/ldm3d_diffusion) | [LDM3D: Latent Diffusion Model for 3D](https://arxiv.org/abs/2305.10853) | Text to Image and Depth Generation |

docs/source/en/installation.mdx

@@ -23,7 +23,7 @@ Install 🤗 Diffusers for whichever deep learning library you're working with.
 
 You should install 🤗 Diffusers in a [virtual environment](https://docs.python.org/3/library/venv.html).
 If you're unfamiliar with Python virtual environments, take a look at this [guide](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
-A virtual environment makes it easier to manage different projects, and avoid compatibility issues between dependencies.
+A virtual environment makes it easier to manage different projects and avoid compatibility issues between dependencies.
 
 Start by creating a virtual environment in your project directory:
 
@@ -127,7 +127,7 @@ Your Python environment will find the `main` version of 🤗 Diffusers on the ne
 
 Our library gathers telemetry information during `from_pretrained()` requests.
 This data includes the version of Diffusers and PyTorch/Flax, the requested model or pipeline class,
-and the path to a pretrained checkpoint if it is hosted on the Hub.
+and the path to a pre-trained checkpoint if it is hosted on the Hub.
 This usage data helps us debug issues and prioritize new features.
 Telemetry is only sent when loading models and pipelines from the HuggingFace Hub,
 and is not collected during local usage.
@@ -143,4 +143,4 @@ export DISABLE_TELEMETRY=YES
 On Windows:
 ```bash
 set DISABLE_TELEMETRY=YES
-```
+```

docs/source/en/optimization/fp16.mdx

@@ -50,7 +50,6 @@ from diffusers import DiffusionPipeline
 
 pipe = DiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5",
-    
     torch_dtype=torch.float16,
 )
 pipe = pipe.to("cuda")
@@ -85,7 +84,6 @@ from diffusers import DiffusionPipeline
 
 pipe = DiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5",
-    
     torch_dtype=torch.float16,
 )
 pipe = pipe.to("cuda")
@@ -112,7 +110,6 @@ from diffusers import StableDiffusionPipeline
 
 pipe = StableDiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5",
-    
     torch_dtype=torch.float16,
 )
 pipe = pipe.to("cuda")
@@ -166,7 +163,6 @@ from diffusers import StableDiffusionPipeline
 
 pipe = StableDiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5",
-    
     torch_dtype=torch.float16,
 )
 
@@ -191,7 +187,6 @@ from diffusers import StableDiffusionPipeline
 
 pipe = StableDiffusionPipeline.from_pretrained(
     "runwayml/stable-diffusion-v1-5",
-    
     torch_dtype=torch.float16,
 )
 
@@ -409,7 +404,14 @@ Here are the speedups we obtain on a few Nvidia GPUs when running the inference
 | A100-SXM4-40GB    	| 18.6it/s            	| 29.it/s                        	|
 | A100-SXM-80GB    	| 18.7it/s            	| 29.5it/s                        	|
 
-To leverage it just make sure you have: 
+To leverage it just make sure you have:
+
+<Tip warning={true}>
+
+If you have PyTorch 2.0 installed, you shouldn't use xFormers!
+
+</Tip>
+
  - PyTorch > 1.12
  - Cuda available
  - [Installed the xformers library](xformers).

docs/source/en/optimization/habana.mdx

@@ -16,8 +16,8 @@ specific language governing permissions and limitations under the License.
 
 ## Requirements
 
-- Optimum Habana 1.5 or later, [here](https://huggingface.co/docs/optimum/habana/installation) is how to install it.
-- SynapseAI 1.9.
+- Optimum Habana 1.6 or later, [here](https://huggingface.co/docs/optimum/habana/installation) is how to install it.
+- SynapseAI 1.10.
 
 
 ## Inference Pipeline
@@ -41,7 +41,7 @@ pipeline = GaudiStableDiffusionPipeline.from_pretrained(
     scheduler=scheduler,
     use_habana=True,
     use_hpu_graphs=True,
-    gaudi_config="Habana/stable-diffusion",
+    gaudi_config="Habana/stable-diffusion-2",
 )
 ```
 
@@ -62,18 +62,18 @@ For more information, check out Optimum Habana's [documentation](https://hugging
 
 ## Benchmark
 
-Here are the latencies for Habana first-generation Gaudi and Gaudi2 with the [Habana/stable-diffusion](https://huggingface.co/Habana/stable-diffusion) Gaudi configuration (mixed precision bf16/fp32):
+Here are the latencies for Habana first-generation Gaudi and Gaudi2 with the [Habana/stable-diffusion](https://huggingface.co/Habana/stable-diffusion) and [Habana/stable-diffusion-2](https://huggingface.co/Habana/stable-diffusion-2) Gaudi configurations (mixed precision bf16/fp32):
 
 - [Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5) (512x512 resolution):
 
 |                        | Latency (batch size = 1) | Throughput (batch size = 8) |
 | ---------------------- |:------------------------:|:---------------------------:|
-| first-generation Gaudi | 4.22s                    | 0.29 images/s               |
-| Gaudi2                 | 1.70s                    | 0.925 images/s              |
+| first-generation Gaudi | 3.80s                    | 0.308 images/s              |
+| Gaudi2                 | 1.33s                    | 1.081 images/s              |
 
 - [Stable Diffusion v2.1](https://huggingface.co/stabilityai/stable-diffusion-2-1) (768x768 resolution):
 
 |                        | Latency (batch size = 1) | Throughput                      |
 | ---------------------- |:------------------------:|:-------------------------------:|
-| first-generation Gaudi | 23.3s                    | 0.045 images/s (batch size = 2) |
-| Gaudi2                 | 7.75s                    | 0.14 images/s (batch size = 5)  |
+| first-generation Gaudi | 10.2s                    | 0.108 images/s (batch size = 4) |
+| Gaudi2                 | 3.17s                    | 0.379 images/s (batch size = 8) |

docs/source/en/optimization/torch2.0.mdx

@@ -23,7 +23,7 @@ To benefit from the accelerated attention implementation and `torch.compile()`,
 when PyTorch 2.0 is available.
 
 ```bash
-pip install --upgrade torch torchvision diffusers
+pip install --upgrade torch diffusers
 ```
 
 ## Using accelerated transformers and `torch.compile`.

docs/source/en/quicktour.mdx

@@ -32,8 +32,9 @@ The quicktour is a simplified version of the introductory 🧨 Diffusers [notebo
 
 Before you begin, make sure you have all the necessary libraries installed:
 
-```bash
-!pip install --upgrade diffusers accelerate transformers
+```py
+# uncomment to install the necessary libraries in Colab
+#!pip install --upgrade diffusers accelerate transformers
 ```
 
 - [🤗 Accelerate](https://huggingface.co/docs/accelerate/index) speeds up model loading for inference and training.

docs/source/en/stable_diffusion.mdx

@@ -52,6 +52,8 @@ pipeline = pipeline.to("cuda")
 To make sure you can use the same image and improve on it, use a [`Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) and set a seed for [reproducibility](./using-diffusers/reproducibility):
 
 ```python
+import torch
+
 generator = torch.Generator("cuda").manual_seed(0)
 ```
 
@@ -266,6 +268,6 @@ image_grid(images)
 
 In this tutorial, you learned how to optimize a [`DiffusionPipeline`] for computational and memory efficiency as well as improving the quality of generated outputs. If you're interested in making your pipeline even faster, take a look at the following resources:
 
-- Learn how [PyTorch 2.0](./optimization/torch2.0) and [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html) can yield 5 - 300% faster inference speed.
+- Learn how [PyTorch 2.0](./optimization/torch2.0) and [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html) can yield 5 - 300% faster inference speed. On an A100 GPU, inference can be up to 50% faster!
 - If you can't use PyTorch 2, we recommend you install [xFormers](./optimization/xformers). Its memory-efficient attention mechanism works great with PyTorch 1.13.1 for faster speed and reduced memory consumption.
 - Other optimization techniques, such as model offloading, are covered in [this guide](./optimization/fp16).

docs/source/en/training/controlnet.mdx

@@ -97,7 +97,8 @@ accelerate launch train_controlnet.py \
  --learning_rate=1e-5 \
  --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
  --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
- --train_batch_size=4
+ --train_batch_size=4 \
+ --push_to_hub
 ```
 
 This default configuration requires ~38GB VRAM.
@@ -120,7 +121,8 @@ accelerate launch train_controlnet.py \
  --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
  --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
  --train_batch_size=1 \
- --gradient_accumulation_steps=4
+ --gradient_accumulation_steps=4 \
+  --push_to_hub
 ```
 
 ## Training with multiple GPUs
@@ -143,7 +145,8 @@ accelerate launch --mixed_precision="fp16" --multi_gpu train_controlnet.py \
  --train_batch_size=4 \
  --mixed_precision="fp16" \
  --tracker_project_name="controlnet-demo" \
- --report_to=wandb
+ --report_to=wandb \
+  --push_to_hub
 ```
 
 ## Example results
@@ -191,7 +194,8 @@ accelerate launch train_controlnet.py \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --gradient_checkpointing \
- --use_8bit_adam
+ --use_8bit_adam \
+  --push_to_hub
 ```
 
 ## Training on a 12 GB GPU
@@ -219,7 +223,8 @@ accelerate launch train_controlnet.py \
  --gradient_checkpointing \
  --use_8bit_adam \
  --enable_xformers_memory_efficient_attention \
- --set_grads_to_none
+ --set_grads_to_none \
+  --push_to_hub
 ```
 
 When using `enable_xformers_memory_efficient_attention`, please make sure to install `xformers` by `pip install xformers`. 
@@ -283,7 +288,8 @@ accelerate launch train_controlnet.py \
  --gradient_checkpointing \
  --enable_xformers_memory_efficient_attention \
  --set_grads_to_none \
- --mixed_precision fp16
+ --mixed_precision fp16 \
+ --push_to_hub
 ```
 
 ## Inference

docs/source/en/training/custom_diffusion.mdx

@@ -100,7 +100,8 @@ accelerate launch train_custom_diffusion.py \
   --lr_warmup_steps=0 \
   --max_train_steps=250 \
   --scale_lr --hflip  \
-  --modifier_token "<new1>" 
+  --modifier_token "<new1>" \
+  --push_to_hub
 ```
 
 **Use `--enable_xformers_memory_efficient_attention` for faster training with lower VRAM requirement (16GB per GPU). Follow [this guide](https://github.com/facebookresearch/xformers) for installation instructions.**
@@ -132,7 +133,8 @@ accelerate launch train_custom_diffusion.py \
   --scale_lr --hflip  \
   --modifier_token "<new1>" \
   --validation_prompt="<new1> cat sitting in a bucket" \
-  --report_to="wandb"
+  --report_to="wandb" \
+  --push_to_hub
 ```
 
 Here is an example [Weights and Biases page](https://wandb.ai/sayakpaul/custom-diffusion/runs/26ghrcau) where you can check out the intermediate results along with other training details.  
@@ -168,7 +170,8 @@ accelerate launch train_custom_diffusion.py \
   --max_train_steps=500 \
   --num_class_images=200 \
   --scale_lr --hflip  \
-  --modifier_token "<new1>+<new2>" 
+  --modifier_token "<new1>+<new2>" \
+  --push_to_hub
 ```
 
 Here is an example [Weights and Biases page](https://wandb.ai/sayakpaul/custom-diffusion/runs/3990tzkg) where you can check out the intermediate results along with other training details.  
@@ -207,7 +210,8 @@ accelerate launch train_custom_diffusion.py \
   --scale_lr --hflip --noaug \
   --freeze_model crossattn \
   --modifier_token "<new1>" \
-  --enable_xformers_memory_efficient_attention 
+  --enable_xformers_memory_efficient_attention \
+  --push_to_hub
 ```
 
 ## Inference

docs/source/en/training/dreambooth.mdx

@@ -12,8 +12,6 @@ specific language governing permissions and limitations under the License.
 
 # DreamBooth
 
-[[open-in-colab]]
-
 [DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text-to-image models like Stable Diffusion given just a few (3-5) images of a subject. It allows the model to generate contextualized images of the subject in different scenes, poses, and views.
 
 ![Dreambooth examples from the project's blog](https://dreambooth.github.io/DreamBooth_files/teaser_static.jpg)
@@ -130,7 +128,8 @@ python train_dreambooth_flax.py \
   --resolution=512 \
   --train_batch_size=1 \
   --learning_rate=5e-6 \
-  --max_train_steps=400
+  --max_train_steps=400 \
+  --push_to_hub
 ```
 </jax>
 </frameworkcontent>
@@ -187,7 +186,8 @@ python train_dreambooth_flax.py \
   --train_batch_size=1 \
   --learning_rate=5e-6 \
   --num_class_images=200 \
-  --max_train_steps=800
+  --max_train_steps=800 \
+  --push_to_hub
 ```
 </jax>
 </frameworkcontent>
@@ -223,7 +223,7 @@ accelerate launch train_dreambooth.py \
   --class_prompt="a photo of dog" \
   --resolution=512 \
   --train_batch_size=1 \
-  --use_8bit_adam
+  --use_8bit_adam \
   --gradient_checkpointing \
   --learning_rate=2e-6 \
   --lr_scheduler="constant" \
@@ -253,7 +253,8 @@ python train_dreambooth_flax.py \
   --train_batch_size=1 \
   --learning_rate=2e-6 \
   --num_class_images=200 \
-  --max_train_steps=800
+  --max_train_steps=800 \
+  --push_to_hub
 ```
 </jax>
 </frameworkcontent>
@@ -499,9 +500,68 @@ You may also run inference from any of the [saved training checkpoints](#inferen
 
 ## IF
 
-You can use the lora and full dreambooth scripts to also train the text to image [IF model](https://huggingface.co/DeepFloyd/IF-I-XL-v1.0). A few alternative cli flags are needed due to the model size, the expected input resolution, and the text encoder conventions.
+You can use the lora and full dreambooth scripts to train the text to image [IF model](https://huggingface.co/DeepFloyd/IF-I-XL-v1.0) and the stage II upscaler 
+[IF model](https://huggingface.co/DeepFloyd/IF-II-L-v1.0).
 
-### LoRA Dreambooth
+Note that IF has a predicted variance, and our finetuning scripts only train the models predicted error, so for finetuned IF models we switch to a fixed
+variance schedule. The full finetuning scripts will update the scheduler config for the full saved model. However, when loading saved LoRA weights, you
+must also update the pipeline's scheduler config.
+
+```py
+from diffusers import DiffusionPipeline
+
+pipe = DiffusionPipeline.from_pretrained("DeepFloyd/IF-I-XL-v1.0")
+
+pipe.load_lora_weights("<lora weights path>")
+
+# Update scheduler config to fixed variance schedule
+pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type="fixed_small")
+```
+
+Additionally, a few alternative cli flags are needed for IF.
+
+`--resolution=64`: IF is a pixel space diffusion model. In order to operate on un-compressed pixels, the input images are of a much smaller resolution. 
+
+`--pre_compute_text_embeddings`: IF uses [T5](https://huggingface.co/docs/transformers/model_doc/t5) for its text encoder. In order to save GPU memory, we pre compute all text embeddings and then de-allocate
+T5.
+
+`--tokenizer_max_length=77`: T5 has a longer default text length, but the default IF encoding procedure uses a smaller number.
+
+`--text_encoder_use_attention_mask`: T5 passes the attention mask to the text encoder.
+
+### Tips and Tricks
+We find LoRA to be sufficient for finetuning the stage I model as the low resolution of the model makes representing finegrained detail hard regardless.
+
+For common and/or not-visually complex object concepts, you can get away with not-finetuning the upscaler. Just be sure to adjust the prompt passed to the
+upscaler to remove the new token from the instance prompt. I.e. if your stage I prompt is "a sks dog", use "a dog" for your stage II prompt.
+
+For finegrained detail like faces that aren't present in the original training set, we find that full finetuning of the stage II upscaler is better than 
+LoRA finetuning stage II.
+
+For finegrained detail like faces, we find that lower learning rates along with larger batch sizes work best.
+
+For stage II, we find that lower learning rates are also needed.
+
+We found experimentally that the DDPM scheduler with the default larger number of denoising steps to sometimes work better than the DPM Solver scheduler
+used in the training scripts.
+
+### Stage II additional validation images
+
+The stage II validation requires images to upscale, we can download a downsized version of the training set:
+
+```py
+from huggingface_hub import snapshot_download
+
+local_dir = "./dog_downsized"
+snapshot_download(
+    "diffusers/dog-example-downsized",
+    local_dir=local_dir,
+    repo_type="dataset",
+    ignore_patterns=".gitattributes",
+)
+```
+
+### IF stage I LoRA Dreambooth
 This training configuration requires ~28 GB VRAM.
 
 ```sh
@@ -515,7 +575,7 @@ accelerate launch train_dreambooth_lora.py \
   --instance_data_dir=$INSTANCE_DIR \
   --output_dir=$OUTPUT_DIR \
   --instance_prompt="a sks dog" \
-  --resolution=64 \ # The input resolution of the IF unet is 64x64
+  --resolution=64 \
   --train_batch_size=4 \
   --gradient_accumulation_steps=1 \
   --learning_rate=5e-6 \
@@ -524,16 +584,58 @@ accelerate launch train_dreambooth_lora.py \
   --validation_prompt="a sks dog" \
   --validation_epochs=25 \
   --checkpointing_steps=100 \
-  --pre_compute_text_embeddings \ # Pre compute text embeddings to that T5 doesn't have to be kept in memory
-  --tokenizer_max_length=77 \ # IF expects an override of the max token length
-  --text_encoder_use_attention_mask # IF expects attention mask for text embeddings
+  --pre_compute_text_embeddings \
+  --tokenizer_max_length=77 \
+  --text_encoder_use_attention_mask
 ```
 
-### Full Dreambooth
-Due to the size of the optimizer states, we recommend training the full XL IF model with 8bit adam. 
-Using 8bit adam and the rest of the following config, the model can be trained in ~48 GB VRAM.
+### IF stage II LoRA Dreambooth
 
-For full dreambooth, IF requires very low learning rates. With higher learning rates model quality will degrade.
+`--validation_images`: These images are upscaled during validation steps.
+
+`--class_labels_conditioning=timesteps`: Pass additional conditioning to the UNet needed for stage II.
+
+`--learning_rate=1e-6`: Lower learning rate than stage I.
+
+`--resolution=256`: The upscaler expects higher resolution inputs
+
+```sh
+export MODEL_NAME="DeepFloyd/IF-II-L-v1.0"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="dreambooth_dog_upscale"
+export VALIDATION_IMAGES="dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"
+
+python train_dreambooth_lora.py \
+    --report_to wandb \
+    --pretrained_model_name_or_path=$MODEL_NAME \
+    --instance_data_dir=$INSTANCE_DIR \
+    --output_dir=$OUTPUT_DIR \
+    --instance_prompt="a sks dog" \
+    --resolution=256 \
+    --train_batch_size=4 \
+    --gradient_accumulation_steps=1 \
+    --learning_rate=1e-6 \ 
+    --max_train_steps=2000 \
+    --validation_prompt="a sks dog" \
+    --validation_epochs=100 \
+    --checkpointing_steps=500 \
+    --pre_compute_text_embeddings \
+    --tokenizer_max_length=77 \
+    --text_encoder_use_attention_mask \
+    --validation_images $VALIDATION_IMAGES \
+    --class_labels_conditioning=timesteps
+```
+
+### IF Stage I Full Dreambooth
+`--skip_save_text_encoder`: When training the full model, this will skip saving the entire T5 with the finetuned model. You can still load the pipeline
+with a T5 loaded from the original model.
+
+`use_8bit_adam`: Due to the size of the optimizer states, we recommend training the full XL IF model with 8bit adam. 
+
+`--learning_rate=1e-7`: For full dreambooth, IF requires very low learning rates. With higher learning rates model quality will degrade. Note that it is 
+likely the learning rate can be increased with larger batch sizes.
+
+Using 8bit adam and a batch size of 4, the model can be trained in ~48 GB VRAM.
 
 ```sh
 export MODEL_NAME="DeepFloyd/IF-I-XL-v1.0"
@@ -546,17 +648,56 @@ accelerate launch train_dreambooth.py \
   --instance_data_dir=$INSTANCE_DIR \
   --output_dir=$OUTPUT_DIR \
   --instance_prompt="a photo of sks dog" \
-  --resolution=64 \ # The input resolution of the IF unet is 64x64
+  --resolution=64 \
   --train_batch_size=4 \
   --gradient_accumulation_steps=1 \
   --learning_rate=1e-7 \
   --max_train_steps=150 \
   --validation_prompt "a photo of sks dog" \
   --validation_steps 25 \
-  --text_encoder_use_attention_mask \ # IF expects attention mask for text embeddings
-  --tokenizer_max_length 77 \ # IF expects an override of the max token length
-  --pre_compute_text_embeddings \ # Pre compute text embeddings to that T5 doesn't have to be kept in memory
-  --use_8bit_adam \ # 
+  --text_encoder_use_attention_mask \
+  --tokenizer_max_length 77 \
+  --pre_compute_text_embeddings \
+  --use_8bit_adam \
   --set_grads_to_none \
-  --skip_save_text_encoder # do not save the full T5 text encoder with the model
-```
+  --skip_save_text_encoder \
+  --push_to_hub
+```
+
+### IF Stage II Full Dreambooth
+
+`--learning_rate=5e-6`: With a smaller effective batch size of 4, we found that we required learning rates as low as
+1e-8.
+
+`--resolution=256`: The upscaler expects higher resolution inputs
+
+`--train_batch_size=2` and `--gradient_accumulation_steps=6`: We found that full training of stage II particularly with
+faces required large effective batch sizes.
+
+```sh
+export MODEL_NAME="DeepFloyd/IF-II-L-v1.0"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="dreambooth_dog_upscale"
+export VALIDATION_IMAGES="dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"
+
+accelerate launch train_dreambooth.py \
+  --report_to wandb \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --instance_data_dir=$INSTANCE_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --instance_prompt="a sks dog" \
+  --resolution=256 \
+  --train_batch_size=2 \
+  --gradient_accumulation_steps=6 \
+  --learning_rate=5e-6 \
+  --max_train_steps=2000 \
+  --validation_prompt="a sks dog" \
+  --validation_steps=150 \
+  --checkpointing_steps=500 \
+  --pre_compute_text_embeddings \
+  --tokenizer_max_length=77 \
+  --text_encoder_use_attention_mask \
+  --validation_images $VALIDATION_IMAGES \
+  --class_labels_conditioning timesteps \
+  --push_to_hub
+```

docs/source/en/training/instructpix2pix.mdx

@@ -100,7 +100,8 @@ accelerate launch --mixed_precision="fp16" train_instruct_pix2pix.py \
     --learning_rate=5e-05 --max_grad_norm=1 --lr_warmup_steps=0 \
     --conditioning_dropout_prob=0.05 \
     --mixed_precision=fp16 \
-    --seed=42 
+    --seed=42 \
+    --push_to_hub
 ```
 
 Additionally, we support performing validation inference to monitor training progress
@@ -121,7 +122,8 @@ accelerate launch --mixed_precision="fp16" train_instruct_pix2pix.py \
     --val_image_url="https://hf.co/datasets/diffusers/diffusers-images-docs/resolve/main/mountain.png" \
     --validation_prompt="make the mountains snowy" \
     --seed=42 \
-    --report_to=wandb 
+    --report_to=wandb \
+    --push_to_hub
  ```
 
  We recommend this type of validation as it can be useful for model debugging. Note that you need `wandb` installed to use this. You can install `wandb` by running `pip install wandb`. 
@@ -148,7 +150,8 @@ accelerate launch --mixed_precision="fp16" --multi_gpu train_instruct_pix2pix.py
  --learning_rate=5e-05 --lr_warmup_steps=0 \
  --conditioning_dropout_prob=0.05 \
  --mixed_precision=fp16 \
- --seed=42 
+ --seed=42 \
+ --push_to_hub
 ```
 
  ## Inference
@@ -204,3 +207,5 @@ speed and quality during performance:
 
 Particularly, `image_guidance_scale` and `guidance_scale` can have a profound impact
 on the generated ("edited") image (see [here](https://twitter.com/RisingSayak/status/1628392199196151808?s=20) for an example).
+
+If you're looking for some interesting ways to use the InstructPix2Pix training methodology, we welcome you to check out this blog post: [Instruction-tuning Stable Diffusion with InstructPix2Pix](https://huggingface.co/blog/instruction-tuning-sd). 

docs/source/en/training/lora.mdx

@@ -12,8 +12,6 @@ specific language governing permissions and limitations under the License.
 
 # Low-Rank Adaptation of Large Language Models (LoRA)
 
-[[open-in-colab]]
-
 <Tip warning={true}>
 
 Currently, LoRA is only supported for the attention layers of the [`UNet2DConditionalModel`]. We also 
@@ -260,6 +258,14 @@ pipe.load_lora_weights(lora_model_id)
 image = pipe("A picture of a sks dog in a bucket", num_inference_steps=25).images[0]
 ```
 
+<Tip>
+
+If your LoRA parameters involve the UNet as well as the Text Encoder, then passing
+`cross_attention_kwargs={"scale": 0.5}` will apply the `scale` value to both the UNet 
+and the Text Encoder. 
+
+</Tip>
+
 Note that the use of [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] is preferred to [`~diffusers.loaders.UNet2DConditionLoadersMixin.load_attn_procs`] for loading LoRA parameters. This is because
 [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] can handle the following situations:
 
@@ -272,4 +278,75 @@ Note that the use of [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] is
 * LoRA parameters that have separate identifiers for the UNet and the text encoder such as: [`"sayakpaul/dreambooth"`](https://huggingface.co/sayakpaul/dreambooth).
 
 **Note** that it is possible to provide a local directory path to [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] as well as [`~diffusers.loaders.UNet2DConditionLoadersMixin.load_attn_procs`]. To know about the supported inputs,
-refer to the respective docstrings.
+refer to the respective docstrings.
+
+## Supporting A1111 themed LoRA checkpoints from Diffusers
+
+To provide seamless interoperability with A1111 to our users, we support loading A1111 formatted
+LoRA checkpoints using [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] in a limited capacity.
+In this section, we explain how to load an A1111 formatted LoRA checkpoint from [CivitAI](https://civitai.com/)
+in Diffusers and perform inference with it. 
+
+First, download a checkpoint. We'll use
+[this one](https://civitai.com/models/13239/light-and-shadow) for demonstration purposes. 
+
+```bash
+wget https://civitai.com/api/download/models/15603 -O light_and_shadow.safetensors
+```
+
+Next, we initialize a [`~DiffusionPipeline`]:
+
+```python 
+import torch
+
+from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
+
+pipeline = StableDiffusionPipeline.from_pretrained(
+    "gsdf/Counterfeit-V2.5", torch_dtype=torch.float16, safety_checker=None
+).to("cuda")
+pipeline.scheduler = DPMSolverMultistepScheduler.from_config(
+    pipeline.scheduler.config, use_karras_sigmas=True
+)
+```
+
+We then load the checkpoint downloaded from CivitAI: 
+
+```python 
+pipeline.load_lora_weights(".", weight_name="light_and_shadow.safetensors")
+```
+
+<Tip warning={true}>
+
+If you're loading a checkpoint in the `safetensors` format, please ensure you have `safetensors` installed.
+
+</Tip>
+
+And then it's time for running inference: 
+
+```python 
+prompt = "masterpiece, best quality, 1girl, at dusk"
+negative_prompt = ("(low quality, worst quality:1.4), (bad anatomy), (inaccurate limb:1.2), "
+                   "bad composition, inaccurate eyes, extra digit, fewer digits, (extra arms:1.2), large breasts")
+
+images = pipeline(prompt=prompt, 
+    negative_prompt=negative_prompt, 
+    width=512, 
+    height=768, 
+    num_inference_steps=15, 
+    num_images_per_prompt=4,
+    generator=torch.manual_seed(0)
+).images
+```
+
+Below is a comparison between the LoRA and the non-LoRA results:
+
+![lora_non_lora](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lora_non_lora_comparison.png)
+
+You have a similar checkpoint stored on the Hugging Face Hub, you can load it
+directly with [`~diffusers.loaders.LoraLoaderMixin.load_lora_weights`] like so: 
+
+```python 
+lora_model_id = "sayakpaul/civitai-light-shadow-lora"
+lora_filename = "light_and_shadow.safetensors"
+pipeline.load_lora_weights(lora_model_id, weight_name=lora_filename)
+```

docs/source/en/training/text2image.mdx

@@ -76,13 +76,25 @@ Launch the [PyTorch training script](https://github.com/huggingface/diffusers/bl
 
 Specify the `MODEL_NAME` environment variable (either a Hub model repository id or a path to the directory containing the model weights) and pass it to the [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) argument.
 
-<literalinclude>
-{"path": "../../../../examples/text_to_image/README.md",
-"language": "bash",
-"start-after": "accelerate_snippet_start",
-"end-before": "accelerate_snippet_end",
-"dedent": 0}
-</literalinclude>
+```bash
+export MODEL_NAME="CompVis/stable-diffusion-v1-4"
+export dataset_name="lambdalabs/pokemon-blip-captions"
+
+accelerate launch --mixed_precision="fp16"  train_text_to_image.py \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --dataset_name=$dataset_name \
+  --use_ema \
+  --resolution=512 --center_crop --random_flip \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --max_train_steps=15000 \
+  --learning_rate=1e-05 \
+  --max_grad_norm=1 \
+  --lr_scheduler="constant" --lr_warmup_steps=0 \
+  --output_dir="sd-pokemon-model" \
+  --push_to_hub
+```
 
 To finetune on your own dataset, prepare the dataset according to the format required by 🤗 [Datasets](https://huggingface.co/docs/datasets/index). You can [upload your dataset to the Hub](https://huggingface.co/docs/datasets/image_dataset#upload-dataset-to-the-hub), or you can [prepare a local folder with your files](https://huggingface.co/docs/datasets/image_dataset#imagefolder).
 
@@ -105,8 +117,10 @@ accelerate launch train_text_to_image.py \
   --max_train_steps=15000 \
   --learning_rate=1e-05 \
   --max_grad_norm=1 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --output_dir=${OUTPUT_DIR}
+  --lr_scheduler="constant" 
+  --lr_warmup_steps=0 \
+  --output_dir=${OUTPUT_DIR} \
+  --push_to_hub
 ```
 
 #### Training with multiple GPUs
@@ -129,8 +143,10 @@ accelerate launch --mixed_precision="fp16" --multi_gpu  train_text_to_image.py \
   --max_train_steps=15000 \ 
   --learning_rate=1e-05 \
   --max_grad_norm=1 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --output_dir="sd-pokemon-model" 
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --output_dir="sd-pokemon-model" \
+  --push_to_hub
 ```
 
 </pt>
@@ -159,7 +175,8 @@ python train_text_to_image_flax.py \
   --max_train_steps=15000 \
   --learning_rate=1e-05 \
   --max_grad_norm=1 \
-  --output_dir="sd-pokemon-model" 
+  --output_dir="sd-pokemon-model" \
+  --push_to_hub
 ```
 
 To finetune on your own dataset, prepare the dataset according to the format required by 🤗 [Datasets](https://huggingface.co/docs/datasets/index). You can [upload your dataset to the Hub](https://huggingface.co/docs/datasets/image_dataset#upload-dataset-to-the-hub), or you can [prepare a local folder with your files](https://huggingface.co/docs/datasets/image_dataset#imagefolder).
@@ -179,7 +196,8 @@ python train_text_to_image_flax.py \
   --max_train_steps=15000 \
   --learning_rate=1e-05 \
   --max_grad_norm=1 \
-  --output_dir="sd-pokemon-model"
+  --output_dir="sd-pokemon-model" \
+  --push_to_hub
 ```
 </jax>
 </frameworkcontent>

docs/source/en/training/text_inversion.mdx

@@ -14,8 +14,6 @@ specific language governing permissions and limitations under the License.
 
 # Textual Inversion
 
-[[open-in-colab]]
-
 [Textual Inversion](https://arxiv.org/abs/2208.01618) is a technique for capturing novel concepts from a small number of example images. While the technique was originally demonstrated with a [latent diffusion model](https://github.com/CompVis/latent-diffusion), it has since been applied to other model variants like [Stable Diffusion](https://huggingface.co/docs/diffusers/main/en/conceptual/stable_diffusion). The learned concepts can be used to better control the images generated from text-to-image pipelines. It learns new "words" in the text encoder's embedding space, which are used within text prompts for personalized image generation.
 
 ![Textual Inversion example](https://textual-inversion.github.io/static/images/editing/colorful_teapot.JPG)
@@ -120,7 +118,8 @@ accelerate launch textual_inversion.py \
   --learning_rate=5.0e-04 --scale_lr \
   --lr_scheduler="constant" \
   --lr_warmup_steps=0 \
-  --output_dir="textual_inversion_cat"
+  --output_dir="textual_inversion_cat" \
+  --push_to_hub
 ```
 
 <Tip>
@@ -161,7 +160,8 @@ python textual_inversion_flax.py \
   --train_batch_size=1 \
   --max_train_steps=3000 \
   --learning_rate=5.0e-04 --scale_lr \
-  --output_dir="textual_inversion_cat"
+  --output_dir="textual_inversion_cat" \
+  --push_to_hub
 ```
 </jax>
 </frameworkcontent>

docs/source/en/training/unconditional_training.mdx

@@ -141,5 +141,6 @@ accelerate launch --mixed_precision="fp16" --multi_gpu train_unconditional.py \
   --learning_rate=1e-4 \
   --lr_warmup_steps=500 \
   --mixed_precision="fp16" \
-  --logger="wandb"
+  --logger="wandb" \
+  --push_to_hub
 ```

docs/source/en/tutorials/basic_training.mdx

@@ -26,8 +26,9 @@ This tutorial will teach you how to train a [`UNet2DModel`] from scratch on a su
 
 Before you begin, make sure you have 🤗 Datasets installed to load and preprocess image datasets, and 🤗 Accelerate, to simplify training on any number of GPUs. The following command will also install [TensorBoard](https://www.tensorflow.org/tensorboard) to visualize training metrics (you can also use [Weights & Biases](https://docs.wandb.ai/) to track your training).
 
-```bash
-!pip install diffusers[training]
+```py
+# uncomment to install the necessary libraries in Colab
+#!pip install diffusers[training]
 ```
 
 We encourage you to share your model with the community, and in order to do that, you'll need to login to your Hugging Face account (create one [here](https://hf.co/join) if you don't already have one!). You can login from a notebook and enter your token when prompted:
@@ -312,7 +313,7 @@ Now you can wrap all these components together in a training loop with 🤗 Acce
 ...         mixed_precision=config.mixed_precision,
 ...         gradient_accumulation_steps=config.gradient_accumulation_steps,
 ...         log_with="tensorboard",
-...         logging_dir=os.path.join(config.output_dir, "logs"),
+...         project_dir=os.path.join(config.output_dir, "logs"),
 ...     )
 ...     if accelerator.is_main_process:
 ...         if config.push_to_hub:
@@ -407,9 +408,9 @@ Once training is complete, take a look at the final 🦋 images 🦋 generated b
 
 ## Next steps
 
-Unconditional image generation is one example of a task that can be trained. You can explore other tasks and training techniques by visiting the [🧨 Diffusers Training Examples](./training/overview) page. Here are some examples of what you can learn:
+Unconditional image generation is one example of a task that can be trained. You can explore other tasks and training techniques by visiting the [🧨 Diffusers Training Examples](../training/overview) page. Here are some examples of what you can learn:
 
-* [Textual Inversion](./training/text_inversion), an algorithm that teaches a model a specific visual concept and integrates it into the generated image.
-* [DreamBooth](./training/dreambooth), a technique for generating personalized images of a subject given several input images of the subject.
-* [Guide](./training/text2image) to finetuning a Stable Diffusion model on your own dataset.
-* [Guide](./training/lora) to using LoRA, a memory-efficient technique for finetuning really large models faster.
+* [Textual Inversion](../training/text_inversion), an algorithm that teaches a model a specific visual concept and integrates it into the generated image.
+* [DreamBooth](../training/dreambooth), a technique for generating personalized images of a subject given several input images of the subject.
+* [Guide](../training/text2image) to finetuning a Stable Diffusion model on your own dataset.
+* [Guide](../training/lora) to using LoRA, a memory-efficient technique for finetuning really large models faster.

docs/source/en/using-diffusers/conditional_image_generation.mdx

@@ -20,12 +20,12 @@ The [`DiffusionPipeline`] is the easiest way to use a pre-trained diffusion syst
 
 Start by creating an instance of [`DiffusionPipeline`] and specify which pipeline [checkpoint](https://huggingface.co/models?library=diffusers&sort=downloads) you would like to download.
 
-In this guide, you'll use [`DiffusionPipeline`] for text-to-image generation with [Latent Diffusion](https://huggingface.co/CompVis/ldm-text2im-large-256):
+In this guide, you'll use [`DiffusionPipeline`] for text-to-image generation with [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5):
 
 ```python
 >>> from diffusers import DiffusionPipeline
 
->>> generator = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
+>>> generator = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
 ```
 
 The [`DiffusionPipeline`] downloads and caches all modeling, tokenization, and scheduling components. 

docs/source/en/using-diffusers/control_brightness.mdx

@@ -0,0 +1,45 @@
+# Control image brightness
+
+The Stable Diffusion pipeline is mediocre at generating images that are either very bright or dark as explained in the [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://huggingface.co/papers/2305.08891) paper. The solutions proposed in the paper are currently implemented in the [`DDIMScheduler`] which you can use to improve the lighting in your images.
+
+<Tip>
+
+💡 Take a look at the paper linked above for more details about the proposed solutions!
+
+</Tip>
+
+One of the solutions is to train a model with *v prediction* and *v loss*. Add the following flag to the [`train_text_to_image.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image.py) or [`train_text_to_image_lora.py`](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_lora.py) scripts to enable `v_prediction`:
+
+```bash
+--prediction_type="v_prediction"
+```
+
+For example, let's use the [`ptx0/pseudo-journey-v2`](https://huggingface.co/ptx0/pseudo-journey-v2) checkpoint which has been finetuned with `v_prediction`.
+
+Next, configure the following parameters in the [`DDIMScheduler`]:
+
+1. `rescale_betas_zero_snr=True`, rescales the noise schedule to zero terminal signal-to-noise ratio (SNR)
+2. `timestep_spacing="trailing"`, starts sampling from the last timestep
+
+```py
+>>> from diffusers import DiffusionPipeline, DDIMScheduler
+
+>>> pipeline = DiffusionPipeline.from_pretrained("ptx0/pseudo-journey-v2")
+# switch the scheduler in the pipeline to use the DDIMScheduler
+
+>>> pipeline.scheduler = DDIMScheduler.from_config(
+...     pipeline.scheduler.config, rescale_betas_zero_snr=True, timestep_spacing="trailing"
+... )
+>>> pipeline.to("cuda")
+```
+
+Finally, in your call to the pipeline, set `guidance_rescale` to prevent overexposure:
+
+```py
+prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intricate detail, octane render, 8k"
+image = pipeline(prompt, guidance_rescale=0.7).images[0]
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/zero_snr.png"/>
+</div>

docs/source/en/using-diffusers/custom_pipeline_examples.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Community pipelines
 
+[[open-in-colab]]
+
 > **For more information about community pipelines, please have a look at [this issue](https://github.com/huggingface/diffusers/issues/841).**
 
 **Community** examples consist of both inference and training examples that have been added by the community.

docs/source/en/using-diffusers/custom_pipeline_overview.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Load community pipelines
 
+[[open-in-colab]]
+
 Community pipelines are any [`DiffusionPipeline`] class that are different from the original implementation as specified in their paper (for example, the [`StableDiffusionControlNetPipeline`] corresponds to the [Text-to-Image Generation with ControlNet Conditioning](https://arxiv.org/abs/2302.05543) paper). They provide additional functionality or extend the original implementation of a pipeline.
 
 There are many cool community pipelines like [Speech to Image](https://github.com/huggingface/diffusers/tree/main/examples/community#speech-to-image) or [Composable Stable Diffusion](https://github.com/huggingface/diffusers/tree/main/examples/community#composable-stable-diffusion), and you can find all the official community pipelines [here](https://github.com/huggingface/diffusers/tree/main/examples/community).

docs/source/en/using-diffusers/img2img.mdx

@@ -18,8 +18,9 @@ The [`StableDiffusionImg2ImgPipeline`] lets you pass a text prompt and an initia
 
 Before you begin, make sure you have all the necessary libraries installed:
 
-```bash
-!pip install diffusers transformers ftfy accelerate
+```py
+# uncomment to install the necessary libraries in Colab
+#!pip install diffusers transformers ftfy accelerate
 ```
 
 Get started by creating a [`StableDiffusionImg2ImgPipeline`] with a pretrained Stable Diffusion model like [`nitrosocke/Ghibli-Diffusion`](https://huggingface.co/nitrosocke/Ghibli-Diffusion).

docs/source/en/using-diffusers/kerascv.mdx

@@ -1,179 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Using KerasCV Stable Diffusion Checkpoints in Diffusers
-
-<Tip warning={true}>
-
-This is an experimental feature.
-
-</Tip>
-
-[KerasCV](https://github.com/keras-team/keras-cv/) provides APIs for implementing various computer vision workflows. It
-also provides the Stable Diffusion [v1 and v2](https://github.com/keras-team/keras-cv/blob/master/keras_cv/models/stable_diffusion)
-models. Many practitioners find it easy to fine-tune the Stable Diffusion models shipped by KerasCV. However, as of this writing, KerasCV offers limited support to experiment with Stable Diffusion models for inference and deployment. On the other hand,
-Diffusers provides tooling dedicated to this purpose (and more), such as different [noise schedulers](https://huggingface.co/docs/diffusers/using-diffusers/schedulers), [flash attention](https://huggingface.co/docs/diffusers/optimization/xformers), and [other 
-optimization techniques](https://huggingface.co/docs/diffusers/optimization/fp16).
-
-How about fine-tuning Stable Diffusion models in KerasCV and exporting them such that they become compatible with Diffusers to combine the
-best of both worlds? We have created a [tool](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers) that
-lets you do just that! It takes KerasCV Stable Diffusion checkpoints and exports them to Diffusers-compatible checkpoints.
-More specifically, it first converts the checkpoints to PyTorch and then wraps them into a
-[`StableDiffusionPipeline`](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/overview) which is ready
-for inference. Finally, it pushes the converted checkpoints to a repository on the Hugging Face Hub. 
-
-We welcome you to try out the tool [here](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers)
-and share feedback via [discussions](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers/discussions/new). 
-
-## Getting Started 
-
-First, you need to obtain the fine-tuned KerasCV Stable Diffusion checkpoints. We provide an
-overview of the different ways Stable Diffusion models can be fine-tuned [using `diffusers`](https://huggingface.co/docs/diffusers/training/overview). For the Keras implementation of some of these methods, you can check out these resources:
-
-* [Teach StableDiffusion new concepts via Textual Inversion](https://keras.io/examples/generative/fine_tune_via_textual_inversion/)
-* [Fine-tuning Stable Diffusion](https://keras.io/examples/generative/finetune_stable_diffusion/)
-* [DreamBooth](https://keras.io/examples/generative/dreambooth/)
-* [Prompt-to-Prompt editing](https://github.com/miguelCalado/prompt-to-prompt-tensorflow)
-
-Stable Diffusion is comprised of the following models:
-
-* Text encoder 
-* UNet 
-* VAE 
-
-Depending on the fine-tuning task, we may fine-tune one or more of these components (the VAE is almost always left untouched). Here are some common combinations:
-
-* DreamBooth: UNet and text encoder 
-* Classical text to image fine-tuning: UNet 
-* Textual Inversion: Just the newly initialized embeddings in the text encoder
-
-### Performing the Conversion
-
-Let's use [this checkpoint](https://huggingface.co/sayakpaul/textual-inversion-kerasio/resolve/main/textual_inversion_kerasio.h5) which was generated
-by conducting Textual Inversion with the following "placeholder token": `<my-funny-cat-token>`. 
-
-On the tool, we supply the following things: 
-
-* Path(s) to download the fine-tuned checkpoint(s) (KerasCV)
-* An HF token 
-* Placeholder token (only applicable for Textual Inversion)
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/space_snap.png"/>
-</div> 
-
-As soon as you hit "Submit", the conversion process will begin. Once it's complete, you should see the following:
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/model_push_success.png"/>
-</div> 
-
-If you click the [link](https://huggingface.co/sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline/tree/main), you
-should see something like so: 
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/model_repo_contents.png"/>
-</div> 
-
-If you head over to the [model card of the repository](https://huggingface.co/sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline), the
-following should appear: 
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/model_card.png"/>
-</div> 
-
-<Tip>
-
-Note that we're not specifying the UNet weights here since the UNet is not fine-tuned during Textual Inversion.
-
-</Tip>
-
-And that's it! You now have your fine-tuned KerasCV Stable Diffusion model in Diffusers 🧨.
-
-## Using the Converted Model in Diffusers 
-
-Just beside the model card of the [repository](https://huggingface.co/sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline), 
-you'd notice an inference widget to try out the model directly from the UI 🤗
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inference_widget_output.png"/>
-</div> 
-
-On the top right hand side, we provide a "Use in Diffusers" button. If you click the button, you should see the following code-snippet: 
-
-```py 
-from diffusers import DiffusionPipeline
-
-pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
-```
-
-The model is in standard `diffusers` format. Let's perform inference!
-
-```py
-from diffusers import DiffusionPipeline
-
-pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
-pipeline.to("cuda")
-
-placeholder_token = "<my-funny-cat-token>"
-prompt = f"two {placeholder_token} getting married, photorealistic, high quality"
-image = pipeline(prompt, num_inference_steps=50).images[0]
-```
-
-And we get: 
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/diffusers_output_one.png"/>
-</div> 
-
-_**Note that if you specified a `placeholder_token` while performing the conversion, the tool will log it accordingly. Refer
-to the model card of [this repository](https://huggingface.co/sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline)
-as an example.**_
-
-We welcome you to use the tool for various Stable Diffusion fine-tuning scenarios and let us know your feedback! Here are some examples 
-of Diffusers checkpoints that were obtained using the tool: 
-
-* [sayakpaul/text-unet-dogs-kerascv_sd_diffusers_pipeline](https://huggingface.co/sayakpaul/text-unet-dogs-kerascv_sd_diffusers_pipeline) (DreamBooth with both the text encoder and UNet fine-tuned)
-* [sayakpaul/unet-dogs-kerascv_sd_diffusers_pipeline](https://huggingface.co/sayakpaul/unet-dogs-kerascv_sd_diffusers_pipeline) (DreamBooth with only the UNet fine-tuned)
-
-## Incorporating Diffusers Goodies 🎁
-
-Diffusers provides various options that one can leverage to experiment with different inference setups. One particularly
-useful option is the use of a different noise scheduler during inference other than what was used during fine-tuning. 
-Let's try out the [`DPMSolverMultistepScheduler`](https://huggingface.co/docs/diffusers/main/en/api/schedulers/multistep_dpm_solver)
-which is different from the one ([`DDPMScheduler`](https://huggingface.co/docs/diffusers/main/en/api/schedulers/ddpm)) used during
-fine-tuning.
-
-You can read more details about this process in [this section](https://huggingface.co/docs/diffusers/using-diffusers/schedulers).
-
-```py
-from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
-
-pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
-pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
-pipeline.to("cuda")
-
-placeholder_token = "<my-funny-cat-token>"
-prompt = f"two {placeholder_token} getting married, photorealistic, high quality"
-image = pipeline(prompt, num_inference_steps=50).images[0]
-```
-
-<div align="center">
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/diffusers_output_two.png"/>
-</div> 
-
-One can also continue fine-tuning from these Diffusers checkpoints by leveraging some relevant tools from Diffusers. Refer [here](https://huggingface.co/docs/diffusers/training/overview) for 
-more details. For inference-specific optimizations, refer [here](https://huggingface.co/docs/diffusers/main/en/optimization/fp16).
-
-## Known Limitations 
-
-* Only Stable Diffusion v1 checkpoints are supported for conversion in this tool. 

docs/source/en/using-diffusers/loading.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Load pipelines, models, and schedulers
 
+[[open-in-colab]]
+
 Having an easy way to use a diffusion system for inference is essential to 🧨 Diffusers. Diffusion systems often consist of multiple components like parameterized models, tokenizers, and schedulers that interact in complex ways. That is why we designed the [`DiffusionPipeline`] to wrap the complexity of the entire diffusion system into an easy-to-use API, while remaining flexible enough to be adapted for other use cases, such as loading each component individually as building blocks to assemble your own diffusion system.
 
 Everything you need for inference or training is accessible with the `from_pretrained()` method.

docs/source/en/using-diffusers/other-formats.mdx

@@ -0,0 +1,194 @@
+<!--Copyright 2023 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+-->
+
+# Load different Stable Diffusion formats
+
+[[open-in-colab]]
+
+Stable Diffusion models are available in different formats depending on the framework they're trained and saved with, and where you download them from. Converting these formats for use in 🤗 Diffusers allows you to use all the features supported by the library, such as [using different schedulers](schedulers) for inference, [building your custom pipeline](write_own_pipeline), and a variety of techniques and methods for [optimizing inference speed](./optimization/opt_overview).
+
+<Tip>
+
+We highly recommend using the `.safetensors` format because it is more secure than traditional pickled files which are vulnerable and can be exploited to execute any code on your machine (learn more in the [Load safetensors](using_safetensors) guide).
+
+</Tip>
+
+This guide will show you how to convert other Stable Diffusion formats to be compatible with 🤗 Diffusers.
+
+## PyTorch .ckpt
+
+The checkpoint - or `.ckpt` - format is commonly used to store and save models. The `.ckpt` file contains the entire model and is typically several GBs in size. While you can load and use a `.ckpt` file directly with the [`~StableDiffusionPipeline.from_ckpt`] method, it is generally better to convert the `.ckpt` file to 🤗 Diffusers so both formats are available.
+
+There are two options for converting a `.ckpt` file; use a Space to convert the checkpoint or convert the `.ckpt` file with a script.
+
+### Convert with a Space
+
+The easiest and most convenient way to convert a `.ckpt` file is to use the [SD to Diffusers](https://huggingface.co/spaces/diffusers/sd-to-diffusers) Space. You can follow the instructions on the Space to convert the `.ckpt` file.
+
+This approach works well for basic models, but it may struggle with more customized models. You'll know the Space failed if it returns an empty pull request or error. In this case, you can try converting the `.ckpt` file with a script. 
+
+### Convert with a script
+
+🤗 Diffusers provides a [conversion script](https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py) for converting `.ckpt` files. This approach is more reliable than the Space above. 
+
+Before you start, make sure you have a local clone of 🤗 Diffusers to run the script and log in to your Hugging Face account so you can open pull requests and push your converted model to the Hub.
+
+```bash
+huggingface-cli login
+```
+
+To use the script:
+
+1. Git clone the repository containing the `.ckpt` file you want to convert. For this example, let's convert this [TemporalNet](https://huggingface.co/CiaraRowles/TemporalNet) `.ckpt` file:
+
+```bash
+git lfs install
+git clone https://huggingface.co/CiaraRowles/TemporalNet
+```
+
+2. Open a pull request on the repository where you're converting the checkpoint from:
+
+```bash
+cd TemporalNet && git fetch origin refs/pr/13:pr/13
+git checkout pr/13
+```
+
+3. There are several input arguments to configure in the conversion script, but the most important ones are:
+
+    - `checkpoint_path`: the path to the `.ckpt` file to convert.
+    - `original_config_file`: a YAML file defining the configuration of the original architecture. If you can't find this file, try searching for the YAML file in the GitHub repository where you found the `.ckpt` file.
+    - `dump_path`: the path to the converted model.
+
+        For example, you can take the `cldm_v15.yaml` file from the [ControlNet](https://github.com/lllyasviel/ControlNet/tree/main/models) repository because the TemporalNet model is a Stable Diffusion v1.5 and ControlNet model.
+
+4. Now you can run the script to convert the `.ckpt` file:
+
+```bash
+python ../diffusers/scripts/convert_original_stable_diffusion_to_diffusers.py --checkpoint_path temporalnetv3.ckpt --original_config_file cldm_v15.yaml --dump_path ./ --controlnet
+```
+
+5. Once the conversion is done, upload your converted model and test out the resulting [pull request](https://huggingface.co/CiaraRowles/TemporalNet/discussions/13)!
+
+```bash
+git push origin pr/13:refs/pr/13
+```
+
+## Keras .pb or .h5
+
+<Tip warning={true}>
+
+🧪 This is an experimental feature. Only Stable Diffusion v1 checkpoints are supported by the Convert KerasCV Space at the moment. 
+
+</Tip>
+
+[KerasCV](https://keras.io/keras_cv/) supports training for [Stable Diffusion](https://github.com/keras-team/keras-cv/blob/master/keras_cv/models/stable_diffusion) v1 and v2. However, it offers limited support for experimenting with Stable Diffusion models for inference and deployment whereas 🤗 Diffusers has a more complete set of features for this purpose, such as different [noise schedulers](https://huggingface.co/docs/diffusers/using-diffusers/schedulers), [flash attention](https://huggingface.co/docs/diffusers/optimization/xformers), and [other 
+optimization techniques](https://huggingface.co/docs/diffusers/optimization/fp16).
+
+The [Convert KerasCV](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers) Space converts `.pb` or `.h5` files to PyTorch, and then wraps them in a [`StableDiffusionPipeline`] so it is ready for inference. The converted checkpoint is stored in a repository on the Hugging Face Hub.
+
+For this example, let's convert the [`sayakpaul/textual-inversion-kerasio`](https://huggingface.co/sayakpaul/textual-inversion-kerasio/tree/main) checkpoint which was trained with Textual Inversion. It uses the special token `<my-funny-cat>` to personalize images with cats.
+
+The Convert KerasCV Space allows you to input the following:
+
+* Your Hugging Face token.
+* Paths to download the UNet and text encoder weights from. Depending on how the model was trained, you don't necessarily need to provide the paths to both the UNet and text encoder. For example, Textual Inversion only requires the embeddings from the text encoder and a text-to-image model only requires the UNet weights.
+* Placeholder token is only applicable for textual inversion models.
+* The `output_repo_prefix` is the name of the repository where the converted model is stored.
+
+Click the **Submit** button to automatically convert the KerasCV checkpoint! Once the checkpoint is successfully converted, you'll see a link to the new repository containing the converted checkpoint. Follow the link to the new repository, and you'll see the Convert KerasCV Space generated a model card with an inference widget to try out the converted model.
+
+If you prefer to run inference with code, click on the **Use in Diffusers** button in the upper right corner of the model card to copy and paste the code snippet:
+
+```py
+from diffusers import DiffusionPipeline
+
+pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
+```
+
+Then you can generate an image like:
+
+```py
+from diffusers import DiffusionPipeline
+
+pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
+pipeline.to("cuda")
+
+placeholder_token = "<my-funny-cat-token>"
+prompt = f"two {placeholder_token} getting married, photorealistic, high quality"
+image = pipeline(prompt, num_inference_steps=50).images[0]
+```
+
+## A1111 LoRA files
+
+[Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) (A1111) is a popular web UI for Stable Diffusion that supports model sharing platforms like [Civitai](https://civitai.com/). Models trained with the Low-Rank Adaptation (LoRA) technique are especially popular because they're fast to train and have a much smaller file size than a fully finetuned model. 🤗 Diffusers supports loading A1111 LoRA checkpoints with [`~loaders.LoraLoaderMixin.load_lora_weights`]:
+
+```py
+from diffusers import DiffusionPipeline, UniPCMultistepScheduler
+import torch
+
+pipeline = DiffusionPipeline.from_pretrained(
+    "andite/anything-v4.0", torch_dtype=torch.float16, safety_checker=None
+).to("cuda")
+pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config)
+```
+
+Download a LoRA checkpoint from Civitai; this example uses the [Howls Moving Castle,Interior/Scenery LoRA (Ghibli Stlye)](https://civitai.com/models/14605?modelVersionId=19998) checkpoint, but feel free to try out any LoRA checkpoint!
+
+```py
+# uncomment to download the safetensor weights
+#!wget https://civitai.com/api/download/models/19998 -O howls_moving_castle.safetensors
+```
+
+Load the LoRA checkpoint into the pipeline with the [`~loaders.LoraLoaderMixin.load_lora_weights`] method:
+
+```py
+pipeline.load_lora_weights(".", weight_name="howls_moving_castle.safetensors")
+```
+
+Now you can use the pipeline to generate images:
+
+```py
+prompt = "masterpiece, illustration, ultra-detailed, cityscape, san francisco, golden gate bridge, california, bay area, in the snow, beautiful detailed starry sky"
+negative_prompt = "lowres, cropped, worst quality, low quality, normal quality, artifacts, signature, watermark, username, blurry, more than one bridge, bad architecture"
+
+images = pipeline(
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    width=512,
+    height=512,
+    num_inference_steps=25,
+    num_images_per_prompt=4,
+    generator=torch.manual_seed(0),
+).images
+```
+
+Finally, create a helper function to display the images:
+
+```py
+from PIL import Image
+
+
+def image_grid(imgs, rows=2, cols=2):
+    w, h = imgs[0].size
+    grid = Image.new("RGB", size=(cols * w, rows * h))
+
+    for i, img in enumerate(imgs):
+        grid.paste(img, box=(i % cols * w, i // cols * h))
+    return grid
+
+
+image_grid(images)
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/a1111-lora-sf.png"/>
+</div>

docs/source/en/using-diffusers/reproducibility.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Create reproducible pipelines
 
+[[open-in-colab]]
+
 Reproducibility is important for testing, replicating results, and can even be used to [improve image quality](reusing_seeds). However, the randomness in diffusion models is a desired property because it allows the pipeline to generate different images every time it is run. While you can't expect to get the exact same results across platforms, you can expect results to be reproducible across releases and platforms within a certain tolerance range. Even then, tolerance varies depending on the diffusion pipeline and checkpoint.
 
 This is why it's important to understand how to control sources of randomness in diffusion models or use deterministic algorithms.
@@ -111,7 +113,7 @@ print(np.abs(image).sum())
 
 The result is not the same even though you're using an identical seed because the GPU uses a different random number generator than the CPU.
 
-To circumvent this problem, 🧨 Diffusers has a [`randn_tensor`](#diffusers.utils.randn_tensor) function for creating random noise on the CPU, and then moving the tensor to a GPU if necessary. The `randn_tensor` function is used everywhere inside the pipeline, allowing the user to **always** pass a CPU `Generator` even if the pipeline is run on a GPU. 
+To circumvent this problem, 🧨 Diffusers has a [`~diffusers.utils.randn_tensor`] function for creating random noise on the CPU, and then moving the tensor to a GPU if necessary. The `randn_tensor` function is used everywhere inside the pipeline, allowing the user to **always** pass a CPU `Generator` even if the pipeline is run on a GPU. 
 
 You'll see the results are much closer now!
 
@@ -147,9 +149,6 @@ susceptible to precision error propagation. Don't expect similar results across
 different GPU hardware or PyTorch versions. In this case, you'll need to run 
 exactly the same hardware and PyTorch version for full reproducibility.
 
-### randn_tensor
-[[autodoc]] diffusers.utils.randn_tensor
-
 ## Deterministic algorithms
 
 You can also configure PyTorch to use deterministic algorithms to create a reproducible pipeline. However, you should be aware that deterministic algorithms may be slower than nondeterministic ones and you may observe a decrease in performance. But if reproducibility is important to you, then this is the way to go!

docs/source/en/using-diffusers/reusing_seeds.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Improve image quality with deterministic generation
 
+[[open-in-colab]]
+
 A common way to improve the quality of generated images is with *deterministic batch generation*, generate a batch of images and select one image to improve with a more detailed prompt in a second round of inference. The key is to pass a list of [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html#generator)'s to the pipeline for batched image generation, and tie each `Generator` to a seed so you can reuse it for an image.
 
 Let's use [`runwayml/stable-diffusion-v1-5`](runwayml/stable-diffusion-v1-5) for example, and generate several versions of the following prompt:

docs/source/en/using-diffusers/schedulers.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Schedulers
 
+[[open-in-colab]]
+
 Diffusion pipelines are inherently a collection of diffusion models and schedulers that are partly independent from each other. This means that one is able to switch out parts of the pipeline to better customize 
 a pipeline to one's use case. The best example of this is the [Schedulers](../api/schedulers/overview.mdx).
 
@@ -28,18 +30,15 @@ The following paragraphs show how to do so with the 🧨 Diffusers library.
 
 ## Load pipeline
 
-Let's start by loading the stable diffusion pipeline.
-Remember that you have to be a registered user on the 🤗 Hugging Face Hub, and have "click-accepted" the [license](https://huggingface.co/runwayml/stable-diffusion-v1-5) in order to use stable diffusion.
+Let's start by loading the [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) model in the [`DiffusionPipeline`]:
 
 ```python
 from huggingface_hub import login
 from diffusers import DiffusionPipeline
 import torch
 
-# first we need to login with our access token
 login()
 
-# Now we can download the pipeline
 pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16)
 ```
 

docs/source/en/using-diffusers/stable_diffusion_jax_how_to.mdx

@@ -14,9 +14,10 @@ Note that JAX is not exclusive to TPUs, but it shines on that hardware because e
 
 First make sure diffusers is installed.
 
-```bash
-!pip install jax==0.3.25 jaxlib==0.3.25 flax transformers ftfy
-!pip install diffusers
+```py
+# uncomment to install the necessary libraries in Colab
+#!pip install jax==0.3.25 jaxlib==0.3.25 flax transformers ftfy
+#!pip install diffusers
 ```
 
 ```python

docs/source/en/using-diffusers/using_safetensors.mdx

@@ -1,11 +1,14 @@
 # Load safetensors
 
+[[open-in-colab]]
+
 [safetensors](https://github.com/huggingface/safetensors) is a safe and fast file format for storing and loading tensors. Typically, PyTorch model weights are saved or *pickled* into a `.bin` file with Python's [`pickle`](https://docs.python.org/3/library/pickle.html) utility. However, `pickle` is not secure and pickled files may contain malicious code that can be executed. safetensors is a secure alternative to `pickle`, making it ideal for sharing model weights.
 
 This guide will show you how you load `.safetensor` files, and how to convert Stable Diffusion model weights stored in other formats to `.safetensor`. Before you start, make sure you have safetensors installed:
 
-```bash
-!pip install safetensors
+```py
+# uncomment to install the necessary libraries in Colab
+#!pip install safetensors
 ```
 
 If you look at the [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5/tree/main) repository, you'll see weights inside the `text_encoder`, `unet` and `vae` subfolders are stored in the `.safetensors` format. By default, 🤗 Diffusers automatically loads these `.safetensors` files from their subfolders if they're available in the model repository.
@@ -30,14 +33,7 @@ pipeline = StableDiffusionPipeline.from_ckpt(
 
 ## Convert to safetensors
 
-Not all weights on the Hub are available in the `.safetensors` format, and you may encounter weights stored as `.bin`. In this case, use the Space below to convert the weights to `.safetensors`. The Convert Space downloads the pickled weights, converts them, and opens a Pull Request to upload the newly converted `.safetensors` file on the Hub. This way, if there is any malicious code contained in the pickled files, they're uploaded to the Hub - which has a [security scanner](https://huggingface.co/docs/hub/security-pickle#hubs-security-scanner) to detect unsafe files and suspicious pickle imports - instead of your computer.
-
-<iframe
-	src="https://safetensors-convert.hf.space"
-	frameborder="0"
-	width="850"
-	height="450"
-></iframe>
+Not all weights on the Hub are available in the `.safetensors` format, and you may encounter weights stored as `.bin`. In this case, use the [Convert Space](https://huggingface.co/spaces/diffusers/convert) to convert the weights to `.safetensors`. The Convert Space downloads the pickled weights, converts them, and opens a Pull Request to upload the newly converted `.safetensors` file on the Hub. This way, if there is any malicious code contained in the pickled files, they're uploaded to the Hub - which has a [security scanner](https://huggingface.co/docs/hub/security-pickle#hubs-security-scanner) to detect unsafe files and suspicious pickle imports - instead of your computer.
 
 You can use the model with the new `.safetensors` weights by specifying the reference to the Pull Request in the `revision` parameter (you can also test it in this [Check PR](https://huggingface.co/spaces/diffusers/check_pr) Space on the Hub), for example `refs/pr/22`:
 

docs/source/en/using-diffusers/weighted_prompts.mdx

@@ -12,6 +12,8 @@ specific language governing permissions and limitations under the License.
 
 # Weighting prompts
 
+[[open-in-colab]]
+
 Text-guided diffusion models generate images based on a given text prompt. The text prompt
 can include multiple concepts that the model should generate and it's often desirable to weight
 certain parts of the prompt more or less. 
@@ -94,5 +96,15 @@ a try!
 If your favorite pipeline does not have a `prompt_embeds` input, please make sure to open an issue, the 
 diffusers team tries to be as responsive as possible.
 
+Compel 1.1.6 adds a utility class to simplify using textual inversions.  Instantiate a `DiffusersTextualInversionManager` and pass it to Compel init:
+
+```
+textual_inversion_manager = DiffusersTextualInversionManager(pipe)
+compel = Compel(
+    tokenizer=pipe.tokenizer,
+    text_encoder=pipe.text_encoder,
+    textual_inversion_manager=textual_inversion_manager)
+```
+
 Also, please check out the documentation of the [compel](https://github.com/damian0815/compel) library for 
 more information.

docs/source/en/using-diffusers/write_own_pipeline.mdx

@@ -36,69 +36,69 @@ A pipeline is a quick and easy way to run a model for inference, requiring no mo
 
 That was super easy, but how did the pipeline do that? Let's breakdown the pipeline and take a look at what's happening under the hood.
 
-In the example above, the pipeline contains a UNet model and a DDPM scheduler. The pipeline denoises an image by taking random noise the size of the desired output and passing it through the model several times. At each timestep, the model predicts the *noise residual* and the scheduler uses it to predict a less noisy image. The pipeline repeats this process until it reaches the end of the specified number of inference steps. 
+In the example above, the pipeline contains a [`UNet2DModel`] model and a [`DDPMScheduler`]. The pipeline denoises an image by taking random noise the size of the desired output and passing it through the model several times. At each timestep, the model predicts the *noise residual* and the scheduler uses it to predict a less noisy image. The pipeline repeats this process until it reaches the end of the specified number of inference steps. 
 
 To recreate the pipeline with the model and scheduler separately, let's write our own denoising process.
 
 1. Load the model and scheduler:
 
-    ```py
-    >>> from diffusers import DDPMScheduler, UNet2DModel
+```py
+>>> from diffusers import DDPMScheduler, UNet2DModel
 
-    >>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cat-256")
-    >>> model = UNet2DModel.from_pretrained("google/ddpm-cat-256").to("cuda")
-    ```
+>>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cat-256")
+>>> model = UNet2DModel.from_pretrained("google/ddpm-cat-256").to("cuda")
+```
 
 2. Set the number of timesteps to run the denoising process for:
 
-    ```py
-    >>> scheduler.set_timesteps(50)
-    ```
+```py
+>>> scheduler.set_timesteps(50)
+```
 
 3. Setting the scheduler timesteps creates a tensor with evenly spaced elements in it, 50 in this example. Each element corresponds to a timestep at which the model denoises an image. When you create the denoising loop later, you'll iterate over this tensor to denoise an image:
 
-    ```py
-    >>> scheduler.timesteps
-    tensor([980, 960, 940, 920, 900, 880, 860, 840, 820, 800, 780, 760, 740, 720,
-        700, 680, 660, 640, 620, 600, 580, 560, 540, 520, 500, 480, 460, 440,
-        420, 400, 380, 360, 340, 320, 300, 280, 260, 240, 220, 200, 180, 160,
-        140, 120, 100,  80,  60,  40,  20,   0])
-    ```
+```py
+>>> scheduler.timesteps
+tensor([980, 960, 940, 920, 900, 880, 860, 840, 820, 800, 780, 760, 740, 720,
+    700, 680, 660, 640, 620, 600, 580, 560, 540, 520, 500, 480, 460, 440,
+    420, 400, 380, 360, 340, 320, 300, 280, 260, 240, 220, 200, 180, 160,
+    140, 120, 100,  80,  60,  40,  20,   0])
+```
 
 4. Create some random noise with the same shape as the desired output:
 
-    ```py
-    >>> import torch
+```py
+>>> import torch
 
-    >>> sample_size = model.config.sample_size
-    >>> noise = torch.randn((1, 3, sample_size, sample_size)).to("cuda")
-    ```
+>>> sample_size = model.config.sample_size
+>>> noise = torch.randn((1, 3, sample_size, sample_size)).to("cuda")
+```
 
-4. Now write a loop to iterate over the timesteps. At each timestep, the model does a [`UNet2DModel.forward`] pass and returns the noisy residual. The scheduler's [`~DDPMScheduler.step`] method takes the noisy residual, timestep, and input and it predicts the image at the previous timestep. This output becomes the next input to the model in the denoising loop, and it'll repeat until it reaches the end of the `timesteps` array.
+5. Now write a loop to iterate over the timesteps. At each timestep, the model does a [`UNet2DModel.forward`] pass and returns the noisy residual. The scheduler's [`~DDPMScheduler.step`] method takes the noisy residual, timestep, and input and it predicts the image at the previous timestep. This output becomes the next input to the model in the denoising loop, and it'll repeat until it reaches the end of the `timesteps` array.
 
-    ```py
-    >>> input = noise
+```py
+>>> input = noise
 
-    >>> for t in scheduler.timesteps:
-    ...     with torch.no_grad():
-    ...         noisy_residual = model(input, t).sample
-    ...     previous_noisy_sample = scheduler.step(noisy_residual, t, input).prev_sample
-    ...     input = previous_noisy_sample
-    ```
+>>> for t in scheduler.timesteps:
+...     with torch.no_grad():
+...         noisy_residual = model(input, t).sample
+...     previous_noisy_sample = scheduler.step(noisy_residual, t, input).prev_sample
+...     input = previous_noisy_sample
+```
 
-    This is the entire denoising process, and you can use this same pattern to write any diffusion system.
+This is the entire denoising process, and you can use this same pattern to write any diffusion system.
 
-5. The last step is to convert the denoised output into an image:
+6. The last step is to convert the denoised output into an image:
 
-    ```py
-    >>> from PIL import Image
-    >>> import numpy as np
+```py
+>>> from PIL import Image
+>>> import numpy as np
 
-    >>> image = (input / 2 + 0.5).clamp(0, 1)
-    >>> image = image.cpu().permute(0, 2, 3, 1).numpy()[0]
-    >>> image = Image.fromarray((image * 255).round().astype("uint8"))
-    >>> image
-    ```
+>>> image = (input / 2 + 0.5).clamp(0, 1)
+>>> image = image.cpu().permute(0, 2, 3, 1).numpy()[0]
+>>> image = Image.fromarray((image * 255).round().astype("uint8"))
+>>> image
+```
 
 In the next section, you'll put your skills to the test and breakdown the more complex Stable Diffusion pipeline. The steps are more or less the same. You'll initialize the necessary components, and set the number of timesteps to create a `timestep` array. The `timestep` array is used in the denoising loop, and for each element in this array, the model predicts a less noisy image. The denoising loop iterates over the `timestep`'s, and at each timestep, it outputs a noisy residual and the scheduler uses it to predict a less noisy image at the previous timestep. This process is repeated until you reach the end of the `timestep` array.
 
@@ -286,5 +286,5 @@ This is really what 🧨 Diffusers is designed for: to make it intuitive and eas
 
 For your next steps, feel free to:
 
-* Learn how to [build and contribute a pipeline](using-diffusers/#contribute_pipeline) to 🧨 Diffusers. We can't wait and see what you'll come up with!
-* Explore [existing pipelines](./api/pipelines/overview) in the library, and see if you can deconstruct and build a pipeline from scratch using the models and schedulers separately.
+* Learn how to [build and contribute a pipeline](contribute_pipeline) to 🧨 Diffusers. We can't wait and see what you'll come up with!
+* Explore [existing pipelines](../api/pipelines/overview) in the library, and see if you can deconstruct and build a pipeline from scratch using the models and schedulers separately.

docs/source/ko/_toctree.yml

@@ -45,4 +45,4 @@
     title: MPS
   - local: optimization/habana
     title: Habana Gaudi
-  title: 최적화/특수 하드웨어
+  title: 최적화/특수 하드웨어

docs/source/zh/_toctree.yml

@@ -3,272 +3,6 @@
     title: 🧨 Diffusers
   - local: quicktour
     title: 快速入门
-  - local: stable_diffusion
-    title: Effective and efficient diffusion
   - local: installation
     title: 安装
   title: 开始
-- sections:
-  - local: tutorials/tutorial_overview
-    title: Overview
-  - local: using-diffusers/write_own_pipeline
-    title: Understanding models and schedulers
-  - local: tutorials/basic_training
-    title: Train a diffusion model
-  title: Tutorials
-- sections:
-  - sections:
-    - local: using-diffusers/loading_overview
-      title: Overview
-    - local: using-diffusers/loading
-      title: Load pipelines, models, and schedulers
-    - local: using-diffusers/schedulers
-      title: Load and compare different schedulers
-    - local: using-diffusers/custom_pipeline_overview
-      title: Load community pipelines
-    - local: using-diffusers/kerascv
-      title: Load KerasCV Stable Diffusion checkpoints
-    title: Loading & Hub
-  - sections:
-    - local: using-diffusers/pipeline_overview
-      title: Overview
-    - local: using-diffusers/unconditional_image_generation
-      title: Unconditional image generation
-    - local: using-diffusers/conditional_image_generation
-      title: Text-to-image generation
-    - local: using-diffusers/img2img
-      title: Text-guided image-to-image
-    - local: using-diffusers/inpaint
-      title: Text-guided image-inpainting
-    - local: using-diffusers/depth2img
-      title: Text-guided depth-to-image
-    - local: using-diffusers/reusing_seeds
-      title: Improve image quality with deterministic generation
-    - local: using-diffusers/reproducibility
-      title: Create reproducible pipelines
-    - local: using-diffusers/custom_pipeline_examples
-      title: Community pipelines
-    - local: using-diffusers/contribute_pipeline
-      title: How to contribute a community pipeline
-    - local: using-diffusers/using_safetensors
-      title: Using safetensors
-    - local: using-diffusers/stable_diffusion_jax_how_to
-      title: Stable Diffusion in JAX/Flax
-    - local: using-diffusers/weighted_prompts
-      title: Weighting Prompts
-    title: Pipelines for Inference
-  - sections:
-    - local: training/overview
-      title: Overview
-    - local: training/unconditional_training
-      title: Unconditional image generation
-    - local: training/text_inversion
-      title: Textual Inversion
-    - local: training/dreambooth
-      title: DreamBooth
-    - local: training/text2image
-      title: Text-to-image
-    - local: training/lora
-      title: Low-Rank Adaptation of Large Language Models (LoRA)
-    - local: training/controlnet
-      title: ControlNet
-    - local: training/instructpix2pix
-      title: InstructPix2Pix Training
-    - local: training/custom_diffusion
-      title: Custom Diffusion
-    title: Training
-  - sections:
-    - local: using-diffusers/rl
-      title: Reinforcement Learning
-    - local: using-diffusers/audio
-      title: Audio
-    - local: using-diffusers/other-modalities
-      title: Other Modalities
-    title: Taking Diffusers Beyond Images
-  title: Using Diffusers
-- sections:
-  - local: optimization/opt_overview
-    title: Overview
-  - local: optimization/fp16
-    title: Memory and Speed
-  - local: optimization/torch2.0
-    title: Torch2.0 support
-  - local: optimization/xformers
-    title: xFormers
-  - local: optimization/onnx
-    title: ONNX
-  - local: optimization/open_vino
-    title: OpenVINO
-  - local: optimization/coreml
-    title: Core ML
-  - local: optimization/mps
-    title: MPS
-  - local: optimization/habana
-    title: Habana Gaudi
-  - local: optimization/tome
-    title: Token Merging
-  title: Optimization/Special Hardware
-- sections:
-  - local: conceptual/philosophy
-    title: Philosophy
-  - local: using-diffusers/controlling_generation
-    title: Controlled generation
-  - local: conceptual/contribution
-    title: How to contribute?
-  - local: conceptual/ethical_guidelines
-    title: Diffusers' Ethical Guidelines
-  - local: conceptual/evaluation
-    title: Evaluating Diffusion Models
-  title: Conceptual Guides
-- sections:
-  - sections:
-    - local: api/models
-      title: Models
-    - local: api/diffusion_pipeline
-      title: Diffusion Pipeline
-    - local: api/logging
-      title: Logging
-    - local: api/configuration
-      title: Configuration
-    - local: api/outputs
-      title: Outputs
-    - local: api/loaders
-      title: Loaders
-    title: Main Classes
-  - sections:
-    - local: api/pipelines/overview
-      title: Overview
-    - local: api/pipelines/alt_diffusion
-      title: AltDiffusion
-    - local: api/pipelines/audio_diffusion
-      title: Audio Diffusion
-    - local: api/pipelines/audioldm
-      title: AudioLDM
-    - local: api/pipelines/cycle_diffusion
-      title: Cycle Diffusion
-    - local: api/pipelines/dance_diffusion
-      title: Dance Diffusion
-    - local: api/pipelines/ddim
-      title: DDIM
-    - local: api/pipelines/ddpm
-      title: DDPM
-    - local: api/pipelines/dit
-      title: DiT
-    - local: api/pipelines/if
-      title: IF
-    - local: api/pipelines/latent_diffusion
-      title: Latent Diffusion
-    - local: api/pipelines/paint_by_example
-      title: PaintByExample
-    - local: api/pipelines/pndm
-      title: PNDM
-    - local: api/pipelines/repaint
-      title: RePaint
-    - local: api/pipelines/stable_diffusion_safe
-      title: Safe Stable Diffusion
-    - local: api/pipelines/score_sde_ve
-      title: Score SDE VE
-    - local: api/pipelines/semantic_stable_diffusion
-      title: Semantic Guidance
-    - local: api/pipelines/spectrogram_diffusion
-      title: "Spectrogram Diffusion"
-    - sections:
-      - local: api/pipelines/stable_diffusion/overview
-        title: Overview
-      - local: api/pipelines/stable_diffusion/text2img
-        title: Text-to-Image
-      - local: api/pipelines/stable_diffusion/img2img
-        title: Image-to-Image
-      - local: api/pipelines/stable_diffusion/inpaint
-        title: Inpaint
-      - local: api/pipelines/stable_diffusion/depth2img
-        title: Depth-to-Image
-      - local: api/pipelines/stable_diffusion/image_variation
-        title: Image-Variation
-      - local: api/pipelines/stable_diffusion/upscale
-        title: Super-Resolution
-      - local: api/pipelines/stable_diffusion/latent_upscale
-        title: Stable-Diffusion-Latent-Upscaler
-      - local: api/pipelines/stable_diffusion/pix2pix
-        title: InstructPix2Pix
-      - local: api/pipelines/stable_diffusion/attend_and_excite
-        title: Attend and Excite
-      - local: api/pipelines/stable_diffusion/pix2pix_zero
-        title: Pix2Pix Zero
-      - local: api/pipelines/stable_diffusion/self_attention_guidance
-        title: Self-Attention Guidance
-      - local: api/pipelines/stable_diffusion/panorama
-        title: MultiDiffusion Panorama
-      - local: api/pipelines/stable_diffusion/controlnet
-        title: Text-to-Image Generation with ControlNet Conditioning
-      - local: api/pipelines/stable_diffusion/model_editing
-        title: Text-to-Image Model Editing
-      title: Stable Diffusion
-    - local: api/pipelines/stable_diffusion_2
-      title: Stable Diffusion 2
-    - local: api/pipelines/stable_unclip
-      title: Stable unCLIP
-    - local: api/pipelines/stochastic_karras_ve
-      title: Stochastic Karras VE
-    - local: api/pipelines/text_to_video
-      title: Text-to-Video
-    - local: api/pipelines/text_to_video_zero
-      title: Text-to-Video Zero
-    - local: api/pipelines/unclip
-      title: UnCLIP
-    - local: api/pipelines/latent_diffusion_uncond
-      title: Unconditional Latent Diffusion
-    - local: api/pipelines/versatile_diffusion
-      title: Versatile Diffusion
-    - local: api/pipelines/vq_diffusion
-      title: VQ Diffusion
-    title: Pipelines
-  - sections:
-    - local: api/schedulers/overview
-      title: Overview
-    - local: api/schedulers/ddim
-      title: DDIM
-    - local: api/schedulers/ddim_inverse
-      title: DDIMInverse
-    - local: api/schedulers/ddpm
-      title: DDPM
-    - local: api/schedulers/deis
-      title: DEIS
-    - local: api/schedulers/dpm_discrete
-      title: DPM Discrete Scheduler
-    - local: api/schedulers/dpm_discrete_ancestral
-      title: DPM Discrete Scheduler with ancestral sampling
-    - local: api/schedulers/euler_ancestral
-      title: Euler Ancestral Scheduler
-    - local: api/schedulers/euler
-      title: Euler scheduler
-    - local: api/schedulers/heun
-      title: Heun Scheduler
-    - local: api/schedulers/ipndm
-      title: IPNDM
-    - local: api/schedulers/lms_discrete
-      title: Linear Multistep
-    - local: api/schedulers/multistep_dpm_solver
-      title: Multistep DPM-Solver
-    - local: api/schedulers/pndm
-      title: PNDM
-    - local: api/schedulers/repaint
-      title: RePaint Scheduler
-    - local: api/schedulers/singlestep_dpm_solver
-      title: Singlestep DPM-Solver
-    - local: api/schedulers/stochastic_karras_ve
-      title: Stochastic Kerras VE
-    - local: api/schedulers/unipc
-      title: UniPCMultistepScheduler
-    - local: api/schedulers/score_sde_ve
-      title: VE-SDE
-    - local: api/schedulers/score_sde_vp
-      title: VP-SDE
-    - local: api/schedulers/vq_diffusion
-      title: VQDiffusionScheduler
-    title: Schedulers
-  - sections:
-    - local: api/experimental/rl
-      title: RL Planning
-    title: Experimental Features
-  title: API

examples/community/README.md

@@ -36,6 +36,8 @@ If a community doesn't work as expected, please open an issue and ping the autho
 | Stable Diffusion RePaint                                                                                                              | Stable Diffusion pipeline using [RePaint](https://arxiv.org/abs/2201.0986) for inpainting.                                                                                                                                                                                                                                                                                                                                                                                                               | [Stable Diffusion RePaint](#stable-diffusion-repaint )                                    | - |                  [Markus Pobitzer](https://github.com/Markus-Pobitzer) | 
 | TensorRT Stable Diffusion Image to Image Pipeline                                                                                                    | Accelerates the Stable Diffusion Image2Image Pipeline using TensorRT                                                                                                                                                                                                                                                                                                                                                                                                                                      | [TensorRT Stable Diffusion Image to Image Pipeline](#tensorrt-image2image-stable-diffusion-pipeline)      | - |              [Asfiya Baig](https://github.com/asfiyab-nvidia) |
 | Stable Diffusion IPEX Pipeline | Accelerate Stable Diffusion inference pipeline with BF16/FP32 precision on Intel Xeon CPUs with [IPEX](https://github.com/intel/intel-extension-for-pytorch) | [Stable Diffusion on IPEX](#stable-diffusion-on-ipex) | - | [Yingjie Han](https://github.com/yingjie-han/) | 
+| CLIP Guided Images Mixing Stable Diffusion Pipeline | Сombine images using usual diffusion models. | [CLIP Guided Images Mixing Using Stable Diffusion](#clip-guided-images-mixing-with-stable-diffusion) | - | [Karachev Denis](https://github.com/TheDenk) |  
+| TensorRT Stable Diffusion Inpainting Pipeline                                                                                                    | Accelerates the Stable Diffusion Inpainting Pipeline using TensorRT                                                                                                                                                                                                                                                                                                                                                                                                                                      | [TensorRT Stable Diffusion Inpainting Pipeline](#tensorrt-inpainting-stable-diffusion-pipeline)      | - |              [Asfiya Baig](https://github.com/asfiyab-nvidia) |
 
 To load a custom pipeline you just need to pass the `custom_pipeline` argument to `DiffusionPipeline`, as one of the files in `diffusers/examples/community`. Feel free to send a PR with your own pipelines, we will merge them quickly.
 ```py
@@ -1326,6 +1328,8 @@ image.save('tensorrt_img2img_new_zealand_hills.png')
 
 This pipeline uses the Reference Control. Refer to the [sd-webui-controlnet discussion: Reference-only Control](https://github.com/Mikubill/sd-webui-controlnet/discussions/1236)[sd-webui-controlnet discussion: Reference-adain Control](https://github.com/Mikubill/sd-webui-controlnet/discussions/1280).
 
+Based on [this issue](https://github.com/huggingface/diffusers/issues/3566),
+- `EulerAncestralDiscreteScheduler` got poor results.
 
 ```py
 import torch
@@ -1369,6 +1373,9 @@ Output Image of `reference_attn=True` and `reference_adain=True`
 
 This pipeline uses the Reference Control with ControlNet. Refer to the [sd-webui-controlnet discussion: Reference-only Control](https://github.com/Mikubill/sd-webui-controlnet/discussions/1236)[sd-webui-controlnet discussion: Reference-adain Control](https://github.com/Mikubill/sd-webui-controlnet/discussions/1280).
 
+Based on [this issue](https://github.com/huggingface/diffusers/issues/3566),
+- `EulerAncestralDiscreteScheduler` got poor results.
+- `guess_mode=True` works well for ControlNet v1.1
 
 ```py
 import cv2
@@ -1510,3 +1517,193 @@ latency = elapsed_time(pipe4)
 print("Latency of StableDiffusionPipeline--fp32",latency)
 
 ```
+  
+### CLIP Guided Images Mixing With Stable Diffusion
+
+![clip_guided_images_mixing_examples](https://huggingface.co/datasets/TheDenk/images_mixing/resolve/main/main.png)
+
+CLIP guided stable diffusion images mixing pipline allows to combine two images using standard diffusion models.  
+This approach is using (optional) CoCa model to avoid writing image description.  
+[More code examples](https://github.com/TheDenk/images_mixing)
+
+## Example Images Mixing (with CoCa)
+```python
+import requests
+from io import BytesIO
+
+import PIL
+import torch
+import open_clip
+from open_clip import SimpleTokenizer
+from diffusers import DiffusionPipeline
+from transformers import CLIPFeatureExtractor, CLIPModel
+
+
+def download_image(url):
+    response = requests.get(url)
+    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+# Loading additional models
+feature_extractor = CLIPFeatureExtractor.from_pretrained(
+    "laion/CLIP-ViT-B-32-laion2B-s34B-b79K"
+)
+clip_model = CLIPModel.from_pretrained(
+    "laion/CLIP-ViT-B-32-laion2B-s34B-b79K", torch_dtype=torch.float16
+)
+coca_model = open_clip.create_model('coca_ViT-L-14', pretrained='laion2B-s13B-b90k').to('cuda')
+coca_model.dtype = torch.float16
+coca_transform = open_clip.image_transform(
+    coca_model.visual.image_size,
+    is_train = False,
+    mean = getattr(coca_model.visual, 'image_mean', None),
+    std = getattr(coca_model.visual, 'image_std', None),
+)
+coca_tokenizer = SimpleTokenizer()
+
+# Pipline creating
+mixing_pipeline = DiffusionPipeline.from_pretrained(
+    "CompVis/stable-diffusion-v1-4",
+    custom_pipeline="clip_guided_images_mixing_stable_diffusion",
+    clip_model=clip_model,
+    feature_extractor=feature_extractor,
+    coca_model=coca_model,
+    coca_tokenizer=coca_tokenizer,
+    coca_transform=coca_transform,
+    torch_dtype=torch.float16,
+)
+mixing_pipeline.enable_attention_slicing()
+mixing_pipeline = mixing_pipeline.to("cuda")
+
+# Pipline running
+generator = torch.Generator(device="cuda").manual_seed(17) 
+
+def download_image(url):
+    response = requests.get(url)
+    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+content_image = download_image("https://huggingface.co/datasets/TheDenk/images_mixing/resolve/main/boromir.jpg")
+style_image = download_image("https://huggingface.co/datasets/TheDenk/images_mixing/resolve/main/gigachad.jpg")
+
+pipe_images = mixing_pipeline(
+    num_inference_steps=50,
+    content_image=content_image,
+    style_image=style_image,
+    noise_strength=0.65,
+    slerp_latent_style_strength=0.9,
+    slerp_prompt_style_strength=0.1,
+    slerp_clip_image_style_strength=0.1,
+    guidance_scale=9.0,
+    batch_size=1,
+    clip_guidance_scale=100,
+    generator=generator,
+).images
+```
+
+![image_mixing_result](https://huggingface.co/datasets/TheDenk/images_mixing/resolve/main/boromir_gigachad.png)
+
+### Stable Diffusion Mixture Tiling
+
+This pipeline uses the Mixture. Refer to the [Mixture](https://arxiv.org/abs/2302.02412) paper for more details.
+    
+```python
+from diffusers import LMSDiscreteScheduler, DiffusionPipeline
+
+# Creater scheduler and model (similar to StableDiffusionPipeline)
+scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
+pipeline = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", scheduler=scheduler, custom_pipeline="mixture_tiling")
+pipeline.to("cuda")
+
+# Mixture of Diffusers generation
+image = pipeline(
+    prompt=[[
+        "A charming house in the countryside, by jakub rozalski, sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece",
+        "A dirt road in the countryside crossing pastures, by jakub rozalski, sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece",
+        "An old and rusty giant robot lying on a dirt road, by jakub rozalski, dark sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece"
+    ]],
+    tile_height=640,
+    tile_width=640,
+    tile_row_overlap=0,
+    tile_col_overlap=256,
+    guidance_scale=8,
+    seed=7178915308,
+    num_inference_steps=50,
+)["images"][0]
+```
+![mixture_tiling_results](https://huggingface.co/datasets/kadirnar/diffusers_readme_images/resolve/main/mixture_tiling.png)
+
+### TensorRT Inpainting Stable Diffusion Pipeline
+
+The TensorRT Pipeline can be used to accelerate the Inpainting Stable Diffusion Inference run.
+
+NOTE: The ONNX conversions and TensorRT engine build may take up to 30 minutes.
+
+```python
+import requests
+from io import BytesIO
+from PIL import Image
+import torch
+from diffusers import PNDMScheduler
+from diffusers.pipelines.stable_diffusion import StableDiffusionImg2ImgPipeline
+
+# Use the PNDMScheduler scheduler here instead
+scheduler = PNDMScheduler.from_pretrained("stabilityai/stable-diffusion-2-inpainting", subfolder="scheduler")
+
+
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained("stabilityai/stable-diffusion-2-inpainting",
+    custom_pipeline="stable_diffusion_tensorrt_inpaint",
+    revision='fp16',
+    torch_dtype=torch.float16,
+    scheduler=scheduler,
+    )
+
+# re-use cached folder to save ONNX models and TensorRT Engines
+pipe.set_cached_folder("stabilityai/stable-diffusion-2-inpainting", revision='fp16',)
+
+pipe = pipe.to("cuda")
+
+url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+response = requests.get(url)
+input_image = Image.open(BytesIO(response.content)).convert("RGB")
+
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+response = requests.get(mask_url)
+mask_image = Image.open(BytesIO(response.content)).convert("RGB")
+
+prompt = "a mecha robot sitting on a bench"
+image = pipe(prompt, image=input_image, mask_image=mask_image, strength=0.75,).images[0]
+image.save('tensorrt_inpaint_mecha_robot.png')
+```
+
+### Stable Diffusion Mixture Canvas
+
+This pipeline uses the Mixture. Refer to the [Mixture](https://arxiv.org/abs/2302.02412) paper for more details.
+    
+```python
+from PIL import Image
+from diffusers import LMSDiscreteScheduler, DiffusionPipeline
+from diffusers.pipelines.pipeline_utils import Image2ImageRegion, Text2ImageRegion, preprocess_image
+
+
+# Load and preprocess guide image
+iic_image = preprocess_image(Image.open("input_image.png").convert("RGB"))
+
+# Creater scheduler and model (similar to StableDiffusionPipeline)
+scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
+pipeline = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", scheduler=scheduler).to("cuda:0", custom_pipeline="mixture_canvas")
+pipeline.to("cuda")
+
+# Mixture of Diffusers generation
+output = pipeline(
+    canvas_height=800,
+    canvas_width=352,
+    regions=[
+        Text2ImageRegion(0, 800, 0, 352, guidance_scale=8,
+            prompt=f"best quality, masterpiece, WLOP, sakimichan, art contest winner on pixiv, 8K, intricate details, wet effects, rain drops, ethereal, mysterious, futuristic, UHD, HDR, cinematic lighting, in a beautiful forest, rainy day, award winning, trending on artstation, beautiful confident cheerful young woman, wearing a futuristic sleeveless dress, ultra beautiful detailed  eyes, hyper-detailed face, complex,  perfect, model,  textured,  chiaroscuro, professional make-up, realistic, figure in frame, "),
+        Image2ImageRegion(352-800, 352, 0, 352, reference_image=iic_image, strength=1.0),
+    ],
+    num_inference_steps=100,
+    seed=5525475061,
+)["images"][0]
+```
+![Input_Image](https://huggingface.co/datasets/kadirnar/diffusers_readme_images/resolve/main/input_image.png)
+![mixture_canvas_results](https://huggingface.co/datasets/kadirnar/diffusers_readme_images/resolve/main/canvas.png)

examples/community/clip_guided_images_mixing_stable_diffusion.py

@@ -0,0 +1,456 @@
+# -*- coding: utf-8 -*-
+import inspect
+from typing import Optional, Union
+
+import numpy as np
+import PIL
+import torch
+from torch.nn import functional as F
+from torchvision import transforms
+from transformers import CLIPFeatureExtractor, CLIPModel, CLIPTextModel, CLIPTokenizer
+
+from diffusers import (
+    AutoencoderKL,
+    DDIMScheduler,
+    DiffusionPipeline,
+    DPMSolverMultistepScheduler,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+    UNet2DConditionModel,
+)
+from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion import StableDiffusionPipelineOutput
+from diffusers.utils import (
+    PIL_INTERPOLATION,
+    randn_tensor,
+)
+
+
+def preprocess(image, w, h):
+    if isinstance(image, torch.Tensor):
+        return image
+    elif isinstance(image, PIL.Image.Image):
+        image = [image]
+
+    if isinstance(image[0], PIL.Image.Image):
+        image = [np.array(i.resize((w, h), resample=PIL_INTERPOLATION["lanczos"]))[None, :] for i in image]
+        image = np.concatenate(image, axis=0)
+        image = np.array(image).astype(np.float32) / 255.0
+        image = image.transpose(0, 3, 1, 2)
+        image = 2.0 * image - 1.0
+        image = torch.from_numpy(image)
+    elif isinstance(image[0], torch.Tensor):
+        image = torch.cat(image, dim=0)
+    return image
+
+
+def slerp(t, v0, v1, DOT_THRESHOLD=0.9995):
+    if not isinstance(v0, np.ndarray):
+        inputs_are_torch = True
+        input_device = v0.device
+        v0 = v0.cpu().numpy()
+        v1 = v1.cpu().numpy()
+
+    dot = np.sum(v0 * v1 / (np.linalg.norm(v0) * np.linalg.norm(v1)))
+    if np.abs(dot) > DOT_THRESHOLD:
+        v2 = (1 - t) * v0 + t * v1
+    else:
+        theta_0 = np.arccos(dot)
+        sin_theta_0 = np.sin(theta_0)
+        theta_t = theta_0 * t
+        sin_theta_t = np.sin(theta_t)
+        s0 = np.sin(theta_0 - theta_t) / sin_theta_0
+        s1 = sin_theta_t / sin_theta_0
+        v2 = s0 * v0 + s1 * v1
+
+    if inputs_are_torch:
+        v2 = torch.from_numpy(v2).to(input_device)
+
+    return v2
+
+
+def spherical_dist_loss(x, y):
+    x = F.normalize(x, dim=-1)
+    y = F.normalize(y, dim=-1)
+    return (x - y).norm(dim=-1).div(2).arcsin().pow(2).mul(2)
+
+
+def set_requires_grad(model, value):
+    for param in model.parameters():
+        param.requires_grad = value
+
+
+class CLIPGuidedImagesMixingStableDiffusion(DiffusionPipeline):
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        clip_model: CLIPModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: Union[PNDMScheduler, LMSDiscreteScheduler, DDIMScheduler, DPMSolverMultistepScheduler],
+        feature_extractor: CLIPFeatureExtractor,
+        coca_model=None,
+        coca_tokenizer=None,
+        coca_transform=None,
+    ):
+        super().__init__()
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            clip_model=clip_model,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            feature_extractor=feature_extractor,
+            coca_model=coca_model,
+            coca_tokenizer=coca_tokenizer,
+            coca_transform=coca_transform,
+        )
+        self.feature_extractor_size = (
+            feature_extractor.size
+            if isinstance(feature_extractor.size, int)
+            else feature_extractor.size["shortest_edge"]
+        )
+        self.normalize = transforms.Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std)
+        set_requires_grad(self.text_encoder, False)
+        set_requires_grad(self.clip_model, False)
+
+    def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
+        if slice_size == "auto":
+            # half the attention head size is usually a good trade-off between
+            # speed and memory
+            slice_size = self.unet.config.attention_head_dim // 2
+        self.unet.set_attention_slice(slice_size)
+
+    def disable_attention_slicing(self):
+        self.enable_attention_slicing(None)
+
+    def freeze_vae(self):
+        set_requires_grad(self.vae, False)
+
+    def unfreeze_vae(self):
+        set_requires_grad(self.vae, True)
+
+    def freeze_unet(self):
+        set_requires_grad(self.unet, False)
+
+    def unfreeze_unet(self):
+        set_requires_grad(self.unet, True)
+
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+        t_start = max(num_inference_steps - init_timestep, 0)
+        timesteps = self.scheduler.timesteps[t_start:]
+
+        return timesteps, num_inference_steps - t_start
+
+    def prepare_latents(self, image, timestep, batch_size, dtype, device, generator=None):
+        if not isinstance(image, torch.Tensor):
+            raise ValueError(f"`image` has to be of type `torch.Tensor` but is {type(image)}")
+
+        image = image.to(device=device, dtype=dtype)
+
+        if isinstance(generator, list):
+            init_latents = [
+                self.vae.encode(image[i : i + 1]).latent_dist.sample(generator[i]) for i in range(batch_size)
+            ]
+            init_latents = torch.cat(init_latents, dim=0)
+        else:
+            init_latents = self.vae.encode(image).latent_dist.sample(generator)
+
+        # Hardcode 0.18215 because stable-diffusion-2-base has not self.vae.config.scaling_factor
+        init_latents = 0.18215 * init_latents
+        init_latents = init_latents.repeat_interleave(batch_size, dim=0)
+
+        noise = randn_tensor(init_latents.shape, generator=generator, device=device, dtype=dtype)
+
+        # get latents
+        init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+        latents = init_latents
+
+        return latents
+
+    def get_image_description(self, image):
+        transformed_image = self.coca_transform(image).unsqueeze(0)
+        with torch.no_grad(), torch.cuda.amp.autocast():
+            generated = self.coca_model.generate(transformed_image.to(device=self.device, dtype=self.coca_model.dtype))
+        generated = self.coca_tokenizer.decode(generated[0].cpu().numpy())
+        return generated.split("<end_of_text>")[0].replace("<start_of_text>", "").rstrip(" .,")
+
+    def get_clip_image_embeddings(self, image, batch_size):
+        clip_image_input = self.feature_extractor.preprocess(image)
+        clip_image_features = torch.from_numpy(clip_image_input["pixel_values"][0]).unsqueeze(0).to(self.device).half()
+        image_embeddings_clip = self.clip_model.get_image_features(clip_image_features)
+        image_embeddings_clip = image_embeddings_clip / image_embeddings_clip.norm(p=2, dim=-1, keepdim=True)
+        image_embeddings_clip = image_embeddings_clip.repeat_interleave(batch_size, dim=0)
+        return image_embeddings_clip
+
+    @torch.enable_grad()
+    def cond_fn(
+        self,
+        latents,
+        timestep,
+        index,
+        text_embeddings,
+        noise_pred_original,
+        original_image_embeddings_clip,
+        clip_guidance_scale,
+    ):
+        latents = latents.detach().requires_grad_()
+
+        latent_model_input = self.scheduler.scale_model_input(latents, timestep)
+
+        # predict the noise residual
+        noise_pred = self.unet(latent_model_input, timestep, encoder_hidden_states=text_embeddings).sample
+
+        if isinstance(self.scheduler, (PNDMScheduler, DDIMScheduler, DPMSolverMultistepScheduler)):
+            alpha_prod_t = self.scheduler.alphas_cumprod[timestep]
+            beta_prod_t = 1 - alpha_prod_t
+            # compute predicted original sample from predicted noise also called
+            # "predicted x_0" of formula (12) from https://arxiv.org/pdf/2010.02502.pdf
+            pred_original_sample = (latents - beta_prod_t ** (0.5) * noise_pred) / alpha_prod_t ** (0.5)
+
+            fac = torch.sqrt(beta_prod_t)
+            sample = pred_original_sample * (fac) + latents * (1 - fac)
+        elif isinstance(self.scheduler, LMSDiscreteScheduler):
+            sigma = self.scheduler.sigmas[index]
+            sample = latents - sigma * noise_pred
+        else:
+            raise ValueError(f"scheduler type {type(self.scheduler)} not supported")
+
+        # Hardcode 0.18215 because stable-diffusion-2-base has not self.vae.config.scaling_factor
+        sample = 1 / 0.18215 * sample
+        image = self.vae.decode(sample).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+
+        image = transforms.Resize(self.feature_extractor_size)(image)
+        image = self.normalize(image).to(latents.dtype)
+
+        image_embeddings_clip = self.clip_model.get_image_features(image)
+        image_embeddings_clip = image_embeddings_clip / image_embeddings_clip.norm(p=2, dim=-1, keepdim=True)
+
+        loss = spherical_dist_loss(image_embeddings_clip, original_image_embeddings_clip).mean() * clip_guidance_scale
+
+        grads = -torch.autograd.grad(loss, latents)[0]
+
+        if isinstance(self.scheduler, LMSDiscreteScheduler):
+            latents = latents.detach() + grads * (sigma**2)
+            noise_pred = noise_pred_original
+        else:
+            noise_pred = noise_pred_original - torch.sqrt(beta_prod_t) * grads
+        return noise_pred, latents
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        style_image: Union[torch.FloatTensor, PIL.Image.Image],
+        content_image: Union[torch.FloatTensor, PIL.Image.Image],
+        style_prompt: Optional[str] = None,
+        content_prompt: Optional[str] = None,
+        height: Optional[int] = 512,
+        width: Optional[int] = 512,
+        noise_strength: float = 0.6,
+        num_inference_steps: Optional[int] = 50,
+        guidance_scale: Optional[float] = 7.5,
+        batch_size: Optional[int] = 1,
+        eta: float = 0.0,
+        clip_guidance_scale: Optional[float] = 100,
+        generator: Optional[torch.Generator] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        slerp_latent_style_strength: float = 0.8,
+        slerp_prompt_style_strength: float = 0.1,
+        slerp_clip_image_style_strength: float = 0.1,
+    ):
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(f"You have passed {batch_size} batch_size, but only {len(generator)} generators.")
+
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if isinstance(generator, torch.Generator) and batch_size > 1:
+            generator = [generator] + [None] * (batch_size - 1)
+
+        coca_is_none = [
+            ("model", self.coca_model is None),
+            ("tokenizer", self.coca_tokenizer is None),
+            ("transform", self.coca_transform is None),
+        ]
+        coca_is_none = [x[0] for x in coca_is_none if x[1]]
+        coca_is_none_str = ", ".join(coca_is_none)
+        # generate prompts with coca model if prompt is None
+        if content_prompt is None:
+            if len(coca_is_none):
+                raise ValueError(
+                    f"Content prompt is None and CoCa [{coca_is_none_str}] is None."
+                    f"Set prompt or pass Coca [{coca_is_none_str}] to DiffusionPipeline."
+                )
+            content_prompt = self.get_image_description(content_image)
+        if style_prompt is None:
+            if len(coca_is_none):
+                raise ValueError(
+                    f"Style prompt is None and CoCa [{coca_is_none_str}] is None."
+                    f" Set prompt or pass Coca [{coca_is_none_str}] to DiffusionPipeline."
+                )
+            style_prompt = self.get_image_description(style_image)
+
+        # get prompt text embeddings for content and style
+        content_text_input = self.tokenizer(
+            content_prompt,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+        content_text_embeddings = self.text_encoder(content_text_input.input_ids.to(self.device))[0]
+
+        style_text_input = self.tokenizer(
+            style_prompt,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+        style_text_embeddings = self.text_encoder(style_text_input.input_ids.to(self.device))[0]
+
+        text_embeddings = slerp(slerp_prompt_style_strength, content_text_embeddings, style_text_embeddings)
+
+        # duplicate text embeddings for each generation per prompt
+        text_embeddings = text_embeddings.repeat_interleave(batch_size, dim=0)
+
+        # set timesteps
+        accepts_offset = "offset" in set(inspect.signature(self.scheduler.set_timesteps).parameters.keys())
+        extra_set_kwargs = {}
+        if accepts_offset:
+            extra_set_kwargs["offset"] = 1
+
+        self.scheduler.set_timesteps(num_inference_steps, **extra_set_kwargs)
+        # Some schedulers like PNDM have timesteps as arrays
+        # It's more optimized to move all timesteps to correct device beforehand
+        self.scheduler.timesteps.to(self.device)
+
+        timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, noise_strength, self.device)
+        latent_timestep = timesteps[:1].repeat(batch_size)
+
+        # Preprocess image
+        preprocessed_content_image = preprocess(content_image, width, height)
+        content_latents = self.prepare_latents(
+            preprocessed_content_image, latent_timestep, batch_size, text_embeddings.dtype, self.device, generator
+        )
+
+        preprocessed_style_image = preprocess(style_image, width, height)
+        style_latents = self.prepare_latents(
+            preprocessed_style_image, latent_timestep, batch_size, text_embeddings.dtype, self.device, generator
+        )
+
+        latents = slerp(slerp_latent_style_strength, content_latents, style_latents)
+
+        if clip_guidance_scale > 0:
+            content_clip_image_embedding = self.get_clip_image_embeddings(content_image, batch_size)
+            style_clip_image_embedding = self.get_clip_image_embeddings(style_image, batch_size)
+            clip_image_embeddings = slerp(
+                slerp_clip_image_style_strength, content_clip_image_embedding, style_clip_image_embedding
+            )
+
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance:
+            max_length = content_text_input.input_ids.shape[-1]
+            uncond_input = self.tokenizer([""], padding="max_length", max_length=max_length, return_tensors="pt")
+            uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
+            # duplicate unconditional embeddings for each generation per prompt
+            uncond_embeddings = uncond_embeddings.repeat_interleave(batch_size, dim=0)
+
+            # For classifier free guidance, we need to do two forward passes.
+            # Here we concatenate the unconditional and text embeddings into a single batch
+            # to avoid doing two forward passes
+            text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
+
+        # get the initial random noise unless the user supplied it
+
+        # Unlike in other pipelines, latents need to be generated in the target device
+        # for 1-to-1 results reproducibility with the CompVis implementation.
+        # However this currently doesn't work in `mps`.
+        latents_shape = (batch_size, self.unet.config.in_channels, height // 8, width // 8)
+        latents_dtype = text_embeddings.dtype
+        if latents is None:
+            if self.device.type == "mps":
+                # randn does not work reproducibly on mps
+                latents = torch.randn(latents_shape, generator=generator, device="cpu", dtype=latents_dtype).to(
+                    self.device
+                )
+            else:
+                latents = torch.randn(latents_shape, generator=generator, device=self.device, dtype=latents_dtype)
+        else:
+            if latents.shape != latents_shape:
+                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
+            latents = latents.to(self.device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+
+        with self.progress_bar(total=num_inference_steps):
+            for i, t in enumerate(timesteps):
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+                # predict the noise residual
+                noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample
+
+                # perform classifier free guidance
+                if do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                # perform clip guidance
+                if clip_guidance_scale > 0:
+                    text_embeddings_for_guidance = (
+                        text_embeddings.chunk(2)[1] if do_classifier_free_guidance else text_embeddings
+                    )
+                    noise_pred, latents = self.cond_fn(
+                        latents,
+                        t,
+                        i,
+                        text_embeddings_for_guidance,
+                        noise_pred,
+                        clip_image_embeddings,
+                        clip_guidance_scale,
+                    )
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+        # Hardcode 0.18215 because stable-diffusion-2-base has not self.vae.config.scaling_factor
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+
+        image = (image / 2 + 0.5).clamp(0, 1)
+        image = image.cpu().permute(0, 2, 3, 1).numpy()
+
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+
+        if not return_dict:
+            return (image, None)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=None)

examples/community/mixture_canvas.py

@@ -0,0 +1,503 @@
+import re
+from copy import deepcopy
+from dataclasses import asdict, dataclass
+from enum import Enum
+from typing import List, Optional, Union
+
+import numpy as np
+import torch
+from numpy import exp, pi, sqrt
+from torchvision.transforms.functional import resize
+from tqdm.auto import tqdm
+from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+
+from diffusers.models import AutoencoderKL, UNet2DConditionModel
+from diffusers.pipeline_utils import DiffusionPipeline
+from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
+from diffusers.schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
+
+
+def preprocess_image(image):
+    from PIL import Image
+
+    """Preprocess an input image
+
+    Same as
+    https://github.com/huggingface/diffusers/blob/1138d63b519e37f0ce04e027b9f4a3261d27c628/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py#L44
+    """
+    w, h = image.size
+    w, h = (x - x % 32 for x in (w, h))  # resize to integer multiple of 32
+    image = image.resize((w, h), resample=Image.LANCZOS)
+    image = np.array(image).astype(np.float32) / 255.0
+    image = image[None].transpose(0, 3, 1, 2)
+    image = torch.from_numpy(image)
+    return 2.0 * image - 1.0
+
+
+@dataclass
+class CanvasRegion:
+    """Class defining a rectangular region in the canvas"""
+
+    row_init: int  # Region starting row in pixel space (included)
+    row_end: int  # Region end row in pixel space (not included)
+    col_init: int  # Region starting column in pixel space (included)
+    col_end: int  # Region end column in pixel space (not included)
+    region_seed: int = None  # Seed for random operations in this region
+    noise_eps: float = 0.0  # Deviation of a zero-mean gaussian noise to be applied over the latents in this region. Useful for slightly "rerolling" latents
+
+    def __post_init__(self):
+        # Initialize arguments if not specified
+        if self.region_seed is None:
+            self.region_seed = np.random.randint(9999999999)
+        # Check coordinates are non-negative
+        for coord in [self.row_init, self.row_end, self.col_init, self.col_end]:
+            if coord < 0:
+                raise ValueError(
+                    f"A CanvasRegion must be defined with non-negative indices, found ({self.row_init}, {self.row_end}, {self.col_init}, {self.col_end})"
+                )
+        # Check coordinates are divisible by 8, else we end up with nasty rounding error when mapping to latent space
+        for coord in [self.row_init, self.row_end, self.col_init, self.col_end]:
+            if coord // 8 != coord / 8:
+                raise ValueError(
+                    f"A CanvasRegion must be defined with locations divisible by 8, found ({self.row_init}-{self.row_end}, {self.col_init}-{self.col_end})"
+                )
+        # Check noise eps is non-negative
+        if self.noise_eps < 0:
+            raise ValueError(f"A CanvasRegion must be defined noises eps non-negative, found {self.noise_eps}")
+        # Compute coordinates for this region in latent space
+        self.latent_row_init = self.row_init // 8
+        self.latent_row_end = self.row_end // 8
+        self.latent_col_init = self.col_init // 8
+        self.latent_col_end = self.col_end // 8
+
+    @property
+    def width(self):
+        return self.col_end - self.col_init
+
+    @property
+    def height(self):
+        return self.row_end - self.row_init
+
+    def get_region_generator(self, device="cpu"):
+        """Creates a torch.Generator based on the random seed of this region"""
+        # Initialize region generator
+        return torch.Generator(device).manual_seed(self.region_seed)
+
+    @property
+    def __dict__(self):
+        return asdict(self)
+
+
+class MaskModes(Enum):
+    """Modes in which the influence of diffuser is masked"""
+
+    CONSTANT = "constant"
+    GAUSSIAN = "gaussian"
+    QUARTIC = "quartic"  # See https://en.wikipedia.org/wiki/Kernel_(statistics)
+
+
+@dataclass
+class DiffusionRegion(CanvasRegion):
+    """Abstract class defining a region where some class of diffusion process is acting"""
+
+    pass
+
+
+@dataclass
+class Text2ImageRegion(DiffusionRegion):
+    """Class defining a region where a text guided diffusion process is acting"""
+
+    prompt: str = ""  # Text prompt guiding the diffuser in this region
+    guidance_scale: float = 7.5  # Guidance scale of the diffuser in this region. If None, randomize
+    mask_type: MaskModes = MaskModes.GAUSSIAN.value  # Kind of weight mask applied to this region
+    mask_weight: float = 1.0  # Global weights multiplier of the mask
+    tokenized_prompt = None  # Tokenized prompt
+    encoded_prompt = None  # Encoded prompt
+
+    def __post_init__(self):
+        super().__post_init__()
+        # Mask weight cannot be negative
+        if self.mask_weight < 0:
+            raise ValueError(
+                f"A Text2ImageRegion must be defined with non-negative mask weight, found {self.mask_weight}"
+            )
+        # Mask type must be an actual known mask
+        if self.mask_type not in [e.value for e in MaskModes]:
+            raise ValueError(
+                f"A Text2ImageRegion was defined with mask {self.mask_type}, which is not an accepted mask ({[e.value for e in MaskModes]})"
+            )
+        # Randomize arguments if given as None
+        if self.guidance_scale is None:
+            self.guidance_scale = np.random.randint(5, 30)
+        # Clean prompt
+        self.prompt = re.sub(" +", " ", self.prompt).replace("\n", " ")
+
+    def tokenize_prompt(self, tokenizer):
+        """Tokenizes the prompt for this diffusion region using a given tokenizer"""
+        self.tokenized_prompt = tokenizer(
+            self.prompt,
+            padding="max_length",
+            max_length=tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+
+    def encode_prompt(self, text_encoder, device):
+        """Encodes the previously tokenized prompt for this diffusion region using a given encoder"""
+        assert self.tokenized_prompt is not None, ValueError(
+            "Prompt in diffusion region must be tokenized before encoding"
+        )
+        self.encoded_prompt = text_encoder(self.tokenized_prompt.input_ids.to(device))[0]
+
+
+@dataclass
+class Image2ImageRegion(DiffusionRegion):
+    """Class defining a region where an image guided diffusion process is acting"""
+
+    reference_image: torch.FloatTensor = None
+    strength: float = 0.8  # Strength of the image
+
+    def __post_init__(self):
+        super().__post_init__()
+        if self.reference_image is None:
+            raise ValueError("Must provide a reference image when creating an Image2ImageRegion")
+        if self.strength < 0 or self.strength > 1:
+            raise ValueError(f"The value of strength should in [0.0, 1.0] but is {self.strength}")
+        # Rescale image to region shape
+        self.reference_image = resize(self.reference_image, size=[self.height, self.width])
+
+    def encode_reference_image(self, encoder, device, generator, cpu_vae=False):
+        """Encodes the reference image for this Image2Image region into the latent space"""
+        # Place encoder in CPU or not following the parameter cpu_vae
+        if cpu_vae:
+            # Note here we use mean instead of sample, to avoid moving also generator to CPU, which is troublesome
+            self.reference_latents = encoder.cpu().encode(self.reference_image).latent_dist.mean.to(device)
+        else:
+            self.reference_latents = encoder.encode(self.reference_image.to(device)).latent_dist.sample(
+                generator=generator
+            )
+        self.reference_latents = 0.18215 * self.reference_latents
+
+    @property
+    def __dict__(self):
+        # This class requires special casting to dict because of the reference_image tensor. Otherwise it cannot be casted to JSON
+
+        # Get all basic fields from parent class
+        super_fields = {key: getattr(self, key) for key in DiffusionRegion.__dataclass_fields__.keys()}
+        # Pack other fields
+        return {**super_fields, "reference_image": self.reference_image.cpu().tolist(), "strength": self.strength}
+
+
+class RerollModes(Enum):
+    """Modes in which the reroll regions operate"""
+
+    RESET = "reset"  # Completely reset the random noise in the region
+    EPSILON = "epsilon"  # Alter slightly the latents in the region
+
+
+@dataclass
+class RerollRegion(CanvasRegion):
+    """Class defining a rectangular canvas region in which initial latent noise will be rerolled"""
+
+    reroll_mode: RerollModes = RerollModes.RESET.value
+
+
+@dataclass
+class MaskWeightsBuilder:
+    """Auxiliary class to compute a tensor of weights for a given diffusion region"""
+
+    latent_space_dim: int  # Size of the U-net latent space
+    nbatch: int = 1  # Batch size in the U-net
+
+    def compute_mask_weights(self, region: DiffusionRegion) -> torch.tensor:
+        """Computes a tensor of weights for a given diffusion region"""
+        MASK_BUILDERS = {
+            MaskModes.CONSTANT.value: self._constant_weights,
+            MaskModes.GAUSSIAN.value: self._gaussian_weights,
+            MaskModes.QUARTIC.value: self._quartic_weights,
+        }
+        return MASK_BUILDERS[region.mask_type](region)
+
+    def _constant_weights(self, region: DiffusionRegion) -> torch.tensor:
+        """Computes a tensor of constant for a given diffusion region"""
+        latent_width = region.latent_col_end - region.latent_col_init
+        latent_height = region.latent_row_end - region.latent_row_init
+        return torch.ones(self.nbatch, self.latent_space_dim, latent_height, latent_width) * region.mask_weight
+
+    def _gaussian_weights(self, region: DiffusionRegion) -> torch.tensor:
+        """Generates a gaussian mask of weights for tile contributions"""
+        latent_width = region.latent_col_end - region.latent_col_init
+        latent_height = region.latent_row_end - region.latent_row_init
+
+        var = 0.01
+        midpoint = (latent_width - 1) / 2  # -1 because index goes from 0 to latent_width - 1
+        x_probs = [
+            exp(-(x - midpoint) * (x - midpoint) / (latent_width * latent_width) / (2 * var)) / sqrt(2 * pi * var)
+            for x in range(latent_width)
+        ]
+        midpoint = (latent_height - 1) / 2
+        y_probs = [
+            exp(-(y - midpoint) * (y - midpoint) / (latent_height * latent_height) / (2 * var)) / sqrt(2 * pi * var)
+            for y in range(latent_height)
+        ]
+
+        weights = np.outer(y_probs, x_probs) * region.mask_weight
+        return torch.tile(torch.tensor(weights), (self.nbatch, self.latent_space_dim, 1, 1))
+
+    def _quartic_weights(self, region: DiffusionRegion) -> torch.tensor:
+        """Generates a quartic mask of weights for tile contributions
+
+        The quartic kernel has bounded support over the diffusion region, and a smooth decay to the region limits.
+        """
+        quartic_constant = 15.0 / 16.0
+
+        support = (np.array(range(region.latent_col_init, region.latent_col_end)) - region.latent_col_init) / (
+            region.latent_col_end - region.latent_col_init - 1
+        ) * 1.99 - (1.99 / 2.0)
+        x_probs = quartic_constant * np.square(1 - np.square(support))
+        support = (np.array(range(region.latent_row_init, region.latent_row_end)) - region.latent_row_init) / (
+            region.latent_row_end - region.latent_row_init - 1
+        ) * 1.99 - (1.99 / 2.0)
+        y_probs = quartic_constant * np.square(1 - np.square(support))
+
+        weights = np.outer(y_probs, x_probs) * region.mask_weight
+        return torch.tile(torch.tensor(weights), (self.nbatch, self.latent_space_dim, 1, 1))
+
+
+class StableDiffusionCanvasPipeline(DiffusionPipeline):
+    """Stable Diffusion pipeline that mixes several diffusers in the same canvas"""
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: Union[DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler],
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+    ):
+        super().__init__()
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+
+    def decode_latents(self, latents, cpu_vae=False):
+        """Decodes a given array of latents into pixel space"""
+        # scale and decode the image latents with vae
+        if cpu_vae:
+            lat = deepcopy(latents).cpu()
+            vae = deepcopy(self.vae).cpu()
+        else:
+            lat = latents
+            vae = self.vae
+
+        lat = 1 / 0.18215 * lat
+        image = vae.decode(lat).sample
+
+        image = (image / 2 + 0.5).clamp(0, 1)
+        image = image.cpu().permute(0, 2, 3, 1).numpy()
+
+        return self.numpy_to_pil(image)
+
+    def get_latest_timestep_img2img(self, num_inference_steps, strength):
+        """Finds the latest timesteps where an img2img strength does not impose latents anymore"""
+        # get the original timestep using init_timestep
+        offset = self.scheduler.config.get("steps_offset", 0)
+        init_timestep = int(num_inference_steps * (1 - strength)) + offset
+        init_timestep = min(init_timestep, num_inference_steps)
+
+        t_start = min(max(num_inference_steps - init_timestep + offset, 0), num_inference_steps - 1)
+        latest_timestep = self.scheduler.timesteps[t_start]
+
+        return latest_timestep
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        canvas_height: int,
+        canvas_width: int,
+        regions: List[DiffusionRegion],
+        num_inference_steps: Optional[int] = 50,
+        seed: Optional[int] = 12345,
+        reroll_regions: Optional[List[RerollRegion]] = None,
+        cpu_vae: Optional[bool] = False,
+        decode_steps: Optional[bool] = False,
+    ):
+        if reroll_regions is None:
+            reroll_regions = []
+        batch_size = 1
+
+        if decode_steps:
+            steps_images = []
+
+        # Prepare scheduler
+        self.scheduler.set_timesteps(num_inference_steps, device=self.device)
+
+        # Split diffusion regions by their kind
+        text2image_regions = [region for region in regions if isinstance(region, Text2ImageRegion)]
+        image2image_regions = [region for region in regions if isinstance(region, Image2ImageRegion)]
+
+        # Prepare text embeddings
+        for region in text2image_regions:
+            region.tokenize_prompt(self.tokenizer)
+            region.encode_prompt(self.text_encoder, self.device)
+
+        # Create original noisy latents using the timesteps
+        latents_shape = (batch_size, self.unet.config.in_channels, canvas_height // 8, canvas_width // 8)
+        generator = torch.Generator(self.device).manual_seed(seed)
+        init_noise = torch.randn(latents_shape, generator=generator, device=self.device)
+
+        # Reset latents in seed reroll regions, if requested
+        for region in reroll_regions:
+            if region.reroll_mode == RerollModes.RESET.value:
+                region_shape = (
+                    latents_shape[0],
+                    latents_shape[1],
+                    region.latent_row_end - region.latent_row_init,
+                    region.latent_col_end - region.latent_col_init,
+                )
+                init_noise[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ] = torch.randn(region_shape, generator=region.get_region_generator(self.device), device=self.device)
+
+        # Apply epsilon noise to regions: first diffusion regions, then reroll regions
+        all_eps_rerolls = regions + [r for r in reroll_regions if r.reroll_mode == RerollModes.EPSILON.value]
+        for region in all_eps_rerolls:
+            if region.noise_eps > 0:
+                region_noise = init_noise[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ]
+                eps_noise = (
+                    torch.randn(
+                        region_noise.shape, generator=region.get_region_generator(self.device), device=self.device
+                    )
+                    * region.noise_eps
+                )
+                init_noise[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ] += eps_noise
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = init_noise * self.scheduler.init_noise_sigma
+
+        # Get unconditional embeddings for classifier free guidance in text2image regions
+        for region in text2image_regions:
+            max_length = region.tokenized_prompt.input_ids.shape[-1]
+            uncond_input = self.tokenizer(
+                [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt"
+            )
+            uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
+
+            # For classifier free guidance, we need to do two forward passes.
+            # Here we concatenate the unconditional and text embeddings into a single batch
+            # to avoid doing two forward passes
+            region.encoded_prompt = torch.cat([uncond_embeddings, region.encoded_prompt])
+
+        # Prepare image latents
+        for region in image2image_regions:
+            region.encode_reference_image(self.vae, device=self.device, generator=generator)
+
+        # Prepare mask of weights for each region
+        mask_builder = MaskWeightsBuilder(latent_space_dim=self.unet.config.in_channels, nbatch=batch_size)
+        mask_weights = [mask_builder.compute_mask_weights(region).to(self.device) for region in text2image_regions]
+
+        # Diffusion timesteps
+        for i, t in tqdm(enumerate(self.scheduler.timesteps)):
+            # Diffuse each region
+            noise_preds_regions = []
+
+            # text2image regions
+            for region in text2image_regions:
+                region_latents = latents[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ]
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([region_latents] * 2)
+                # scale model input following scheduler rules
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+                # predict the noise residual
+                noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=region.encoded_prompt)["sample"]
+                # perform guidance
+                noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                noise_pred_region = noise_pred_uncond + region.guidance_scale * (noise_pred_text - noise_pred_uncond)
+                noise_preds_regions.append(noise_pred_region)
+
+            # Merge noise predictions for all tiles
+            noise_pred = torch.zeros(latents.shape, device=self.device)
+            contributors = torch.zeros(latents.shape, device=self.device)
+            # Add each tile contribution to overall latents
+            for region, noise_pred_region, mask_weights_region in zip(
+                text2image_regions, noise_preds_regions, mask_weights
+            ):
+                noise_pred[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ] += (
+                    noise_pred_region * mask_weights_region
+                )
+                contributors[
+                    :,
+                    :,
+                    region.latent_row_init : region.latent_row_end,
+                    region.latent_col_init : region.latent_col_end,
+                ] += mask_weights_region
+            # Average overlapping areas with more than 1 contributor
+            noise_pred /= contributors
+            noise_pred = torch.nan_to_num(
+                noise_pred
+            )  # Replace NaNs by zeros: NaN can appear if a position is not covered by any DiffusionRegion
+
+            # compute the previous noisy sample x_t -> x_t-1
+            latents = self.scheduler.step(noise_pred, t, latents).prev_sample
+
+            # Image2Image regions: override latents generated by the scheduler
+            for region in image2image_regions:
+                influence_step = self.get_latest_timestep_img2img(num_inference_steps, region.strength)
+                # Only override in the timesteps before the last influence step of the image (given by its strength)
+                if t > influence_step:
+                    timestep = t.repeat(batch_size)
+                    region_init_noise = init_noise[
+                        :,
+                        :,
+                        region.latent_row_init : region.latent_row_end,
+                        region.latent_col_init : region.latent_col_end,
+                    ]
+                    region_latents = self.scheduler.add_noise(region.reference_latents, region_init_noise, timestep)
+                    latents[
+                        :,
+                        :,
+                        region.latent_row_init : region.latent_row_end,
+                        region.latent_col_init : region.latent_col_end,
+                    ] = region_latents
+
+            if decode_steps:
+                steps_images.append(self.decode_latents(latents, cpu_vae))
+
+        # scale and decode the image latents with vae
+        image = self.decode_latents(latents, cpu_vae)
+
+        output = {"images": image}
+        if decode_steps:
+            output = {**output, "steps_images": steps_images}
+        return output

examples/community/mixture_tiling.py

@@ -0,0 +1,405 @@
+import inspect
+from copy import deepcopy
+from enum import Enum
+from typing import List, Optional, Tuple, Union
+
+import torch
+from tqdm.auto import tqdm
+
+from diffusers.models import AutoencoderKL, UNet2DConditionModel
+from diffusers.pipeline_utils import DiffusionPipeline
+from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
+from diffusers.schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
+from diffusers.utils import logging
+
+
+try:
+    from ligo.segments import segment
+    from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+except ImportError:
+    raise ImportError("Please install transformers and ligo-segments to use the mixture pipeline")
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> from diffusers import LMSDiscreteScheduler, DiffusionPipeline
+
+        >>> scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
+        >>> pipeline = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", scheduler=scheduler, custom_pipeline="mixture_tiling")
+        >>> pipeline.to("cuda")
+
+        >>> image = pipeline(
+        >>>     prompt=[[
+        >>>         "A charming house in the countryside, by jakub rozalski, sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece",
+        >>>         "A dirt road in the countryside crossing pastures, by jakub rozalski, sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece",
+        >>>         "An old and rusty giant robot lying on a dirt road, by jakub rozalski, dark sunset lighting, elegant, highly detailed, smooth, sharp focus, artstation, stunning masterpiece"
+        >>>     ]],
+        >>>     tile_height=640,
+        >>>     tile_width=640,
+        >>>     tile_row_overlap=0,
+        >>>     tile_col_overlap=256,
+        >>>     guidance_scale=8,
+        >>>     seed=7178915308,
+        >>>     num_inference_steps=50,
+    >>> )["images"][0]
+        ```
+"""
+
+
+def _tile2pixel_indices(tile_row, tile_col, tile_width, tile_height, tile_row_overlap, tile_col_overlap):
+    """Given a tile row and column numbers returns the range of pixels affected by that tiles in the overall image
+
+    Returns a tuple with:
+        - Starting coordinates of rows in pixel space
+        - Ending coordinates of rows in pixel space
+        - Starting coordinates of columns in pixel space
+        - Ending coordinates of columns in pixel space
+    """
+    px_row_init = 0 if tile_row == 0 else tile_row * (tile_height - tile_row_overlap)
+    px_row_end = px_row_init + tile_height
+    px_col_init = 0 if tile_col == 0 else tile_col * (tile_width - tile_col_overlap)
+    px_col_end = px_col_init + tile_width
+    return px_row_init, px_row_end, px_col_init, px_col_end
+
+
+def _pixel2latent_indices(px_row_init, px_row_end, px_col_init, px_col_end):
+    """Translates coordinates in pixel space to coordinates in latent space"""
+    return px_row_init // 8, px_row_end // 8, px_col_init // 8, px_col_end // 8
+
+
+def _tile2latent_indices(tile_row, tile_col, tile_width, tile_height, tile_row_overlap, tile_col_overlap):
+    """Given a tile row and column numbers returns the range of latents affected by that tiles in the overall image
+
+    Returns a tuple with:
+        - Starting coordinates of rows in latent space
+        - Ending coordinates of rows in latent space
+        - Starting coordinates of columns in latent space
+        - Ending coordinates of columns in latent space
+    """
+    px_row_init, px_row_end, px_col_init, px_col_end = _tile2pixel_indices(
+        tile_row, tile_col, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+    )
+    return _pixel2latent_indices(px_row_init, px_row_end, px_col_init, px_col_end)
+
+
+def _tile2latent_exclusive_indices(
+    tile_row, tile_col, tile_width, tile_height, tile_row_overlap, tile_col_overlap, rows, columns
+):
+    """Given a tile row and column numbers returns the range of latents affected only by that tile in the overall image
+
+    Returns a tuple with:
+        - Starting coordinates of rows in latent space
+        - Ending coordinates of rows in latent space
+        - Starting coordinates of columns in latent space
+        - Ending coordinates of columns in latent space
+    """
+    row_init, row_end, col_init, col_end = _tile2latent_indices(
+        tile_row, tile_col, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+    )
+    row_segment = segment(row_init, row_end)
+    col_segment = segment(col_init, col_end)
+    # Iterate over the rest of tiles, clipping the region for the current tile
+    for row in range(rows):
+        for column in range(columns):
+            if row != tile_row and column != tile_col:
+                clip_row_init, clip_row_end, clip_col_init, clip_col_end = _tile2latent_indices(
+                    row, column, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+                )
+                row_segment = row_segment - segment(clip_row_init, clip_row_end)
+                col_segment = col_segment - segment(clip_col_init, clip_col_end)
+    # return row_init, row_end, col_init, col_end
+    return row_segment[0], row_segment[1], col_segment[0], col_segment[1]
+
+
+class StableDiffusionExtrasMixin:
+    """Mixin providing additional convenience method to Stable Diffusion pipelines"""
+
+    def decode_latents(self, latents, cpu_vae=False):
+        """Decodes a given array of latents into pixel space"""
+        # scale and decode the image latents with vae
+        if cpu_vae:
+            lat = deepcopy(latents).cpu()
+            vae = deepcopy(self.vae).cpu()
+        else:
+            lat = latents
+            vae = self.vae
+
+        lat = 1 / 0.18215 * lat
+        image = vae.decode(lat).sample
+
+        image = (image / 2 + 0.5).clamp(0, 1)
+        image = image.cpu().permute(0, 2, 3, 1).numpy()
+
+        return self.numpy_to_pil(image)
+
+
+class StableDiffusionTilingPipeline(DiffusionPipeline, StableDiffusionExtrasMixin):
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: Union[DDIMScheduler, PNDMScheduler],
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+    ):
+        super().__init__()
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+
+    class SeedTilesMode(Enum):
+        """Modes in which the latents of a particular tile can be re-seeded"""
+
+        FULL = "full"
+        EXCLUSIVE = "exclusive"
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        prompt: Union[str, List[List[str]]],
+        num_inference_steps: Optional[int] = 50,
+        guidance_scale: Optional[float] = 7.5,
+        eta: Optional[float] = 0.0,
+        seed: Optional[int] = None,
+        tile_height: Optional[int] = 512,
+        tile_width: Optional[int] = 512,
+        tile_row_overlap: Optional[int] = 256,
+        tile_col_overlap: Optional[int] = 256,
+        guidance_scale_tiles: Optional[List[List[float]]] = None,
+        seed_tiles: Optional[List[List[int]]] = None,
+        seed_tiles_mode: Optional[Union[str, List[List[str]]]] = "full",
+        seed_reroll_regions: Optional[List[Tuple[int, int, int, int, int]]] = None,
+        cpu_vae: Optional[bool] = False,
+    ):
+        r"""
+        Function to run the diffusion pipeline with tiling support.
+
+        Args:
+            prompt: either a single string (no tiling) or a list of lists with all the prompts to use (one list for each row of tiles). This will also define the tiling structure.
+            num_inference_steps: number of diffusions steps.
+            guidance_scale: classifier-free guidance.
+            seed: general random seed to initialize latents.
+            tile_height: height in pixels of each grid tile.
+            tile_width: width in pixels of each grid tile.
+            tile_row_overlap: number of overlap pixels between tiles in consecutive rows.
+            tile_col_overlap: number of overlap pixels between tiles in consecutive columns.
+            guidance_scale_tiles: specific weights for classifier-free guidance in each tile.
+            guidance_scale_tiles: specific weights for classifier-free guidance in each tile. If None, the value provided in guidance_scale will be used.
+            seed_tiles: specific seeds for the initialization latents in each tile. These will override the latents generated for the whole canvas using the standard seed parameter.
+            seed_tiles_mode: either "full" "exclusive". If "full", all the latents affected by the tile be overriden. If "exclusive", only the latents that are affected exclusively by this tile (and no other tiles) will be overrriden.
+            seed_reroll_regions: a list of tuples in the form (start row, end row, start column, end column, seed) defining regions in pixel space for which the latents will be overriden using the given seed. Takes priority over seed_tiles.
+            cpu_vae: the decoder from latent space to pixel space can require too mucho GPU RAM for large images. If you find out of memory errors at the end of the generation process, try setting this parameter to True to run the decoder in CPU. Slower, but should run without memory issues.
+
+        Examples:
+
+        Returns:
+            A PIL image with the generated image.
+
+        """
+        if not isinstance(prompt, list) or not all(isinstance(row, list) for row in prompt):
+            raise ValueError(f"`prompt` has to be a list of lists but is {type(prompt)}")
+        grid_rows = len(prompt)
+        grid_cols = len(prompt[0])
+        if not all(len(row) == grid_cols for row in prompt):
+            raise ValueError("All prompt rows must have the same number of prompt columns")
+        if not isinstance(seed_tiles_mode, str) and (
+            not isinstance(seed_tiles_mode, list) or not all(isinstance(row, list) for row in seed_tiles_mode)
+        ):
+            raise ValueError(f"`seed_tiles_mode` has to be a string or list of lists but is {type(prompt)}")
+        if isinstance(seed_tiles_mode, str):
+            seed_tiles_mode = [[seed_tiles_mode for _ in range(len(row))] for row in prompt]
+
+        modes = [mode.value for mode in self.SeedTilesMode]
+        if any(mode not in modes for row in seed_tiles_mode for mode in row):
+            raise ValueError(f"Seed tiles mode must be one of {modes}")
+        if seed_reroll_regions is None:
+            seed_reroll_regions = []
+        batch_size = 1
+
+        # create original noisy latents using the timesteps
+        height = tile_height + (grid_rows - 1) * (tile_height - tile_row_overlap)
+        width = tile_width + (grid_cols - 1) * (tile_width - tile_col_overlap)
+        latents_shape = (batch_size, self.unet.config.in_channels, height // 8, width // 8)
+        generator = torch.Generator("cuda").manual_seed(seed)
+        latents = torch.randn(latents_shape, generator=generator, device=self.device)
+
+        # overwrite latents for specific tiles if provided
+        if seed_tiles is not None:
+            for row in range(grid_rows):
+                for col in range(grid_cols):
+                    if (seed_tile := seed_tiles[row][col]) is not None:
+                        mode = seed_tiles_mode[row][col]
+                        if mode == self.SeedTilesMode.FULL.value:
+                            row_init, row_end, col_init, col_end = _tile2latent_indices(
+                                row, col, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+                            )
+                        else:
+                            row_init, row_end, col_init, col_end = _tile2latent_exclusive_indices(
+                                row,
+                                col,
+                                tile_width,
+                                tile_height,
+                                tile_row_overlap,
+                                tile_col_overlap,
+                                grid_rows,
+                                grid_cols,
+                            )
+                        tile_generator = torch.Generator("cuda").manual_seed(seed_tile)
+                        tile_shape = (latents_shape[0], latents_shape[1], row_end - row_init, col_end - col_init)
+                        latents[:, :, row_init:row_end, col_init:col_end] = torch.randn(
+                            tile_shape, generator=tile_generator, device=self.device
+                        )
+
+        # overwrite again for seed reroll regions
+        for row_init, row_end, col_init, col_end, seed_reroll in seed_reroll_regions:
+            row_init, row_end, col_init, col_end = _pixel2latent_indices(
+                row_init, row_end, col_init, col_end
+            )  # to latent space coordinates
+            reroll_generator = torch.Generator("cuda").manual_seed(seed_reroll)
+            region_shape = (latents_shape[0], latents_shape[1], row_end - row_init, col_end - col_init)
+            latents[:, :, row_init:row_end, col_init:col_end] = torch.randn(
+                region_shape, generator=reroll_generator, device=self.device
+            )
+
+        # Prepare scheduler
+        accepts_offset = "offset" in set(inspect.signature(self.scheduler.set_timesteps).parameters.keys())
+        extra_set_kwargs = {}
+        if accepts_offset:
+            extra_set_kwargs["offset"] = 1
+        self.scheduler.set_timesteps(num_inference_steps, **extra_set_kwargs)
+        # if we use LMSDiscreteScheduler, let's make sure latents are multiplied by sigmas
+        if isinstance(self.scheduler, LMSDiscreteScheduler):
+            latents = latents * self.scheduler.sigmas[0]
+
+        # get prompts text embeddings
+        text_input = [
+            [
+                self.tokenizer(
+                    col,
+                    padding="max_length",
+                    max_length=self.tokenizer.model_max_length,
+                    truncation=True,
+                    return_tensors="pt",
+                )
+                for col in row
+            ]
+            for row in prompt
+        ]
+        text_embeddings = [[self.text_encoder(col.input_ids.to(self.device))[0] for col in row] for row in text_input]
+
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0  # TODO: also active if any tile has guidance scale
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance:
+            for i in range(grid_rows):
+                for j in range(grid_cols):
+                    max_length = text_input[i][j].input_ids.shape[-1]
+                    uncond_input = self.tokenizer(
+                        [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt"
+                    )
+                    uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
+
+                    # For classifier free guidance, we need to do two forward passes.
+                    # Here we concatenate the unconditional and text embeddings into a single batch
+                    # to avoid doing two forward passes
+                    text_embeddings[i][j] = torch.cat([uncond_embeddings, text_embeddings[i][j]])
+
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # Mask for tile weights strenght
+        tile_weights = self._gaussian_weights(tile_width, tile_height, batch_size)
+
+        # Diffusion timesteps
+        for i, t in tqdm(enumerate(self.scheduler.timesteps)):
+            # Diffuse each tile
+            noise_preds = []
+            for row in range(grid_rows):
+                noise_preds_row = []
+                for col in range(grid_cols):
+                    px_row_init, px_row_end, px_col_init, px_col_end = _tile2latent_indices(
+                        row, col, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+                    )
+                    tile_latents = latents[:, :, px_row_init:px_row_end, px_col_init:px_col_end]
+                    # expand the latents if we are doing classifier free guidance
+                    latent_model_input = torch.cat([tile_latents] * 2) if do_classifier_free_guidance else tile_latents
+                    latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+                    # predict the noise residual
+                    noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings[row][col])[
+                        "sample"
+                    ]
+                    # perform guidance
+                    if do_classifier_free_guidance:
+                        noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                        guidance = (
+                            guidance_scale
+                            if guidance_scale_tiles is None or guidance_scale_tiles[row][col] is None
+                            else guidance_scale_tiles[row][col]
+                        )
+                        noise_pred_tile = noise_pred_uncond + guidance * (noise_pred_text - noise_pred_uncond)
+                        noise_preds_row.append(noise_pred_tile)
+                noise_preds.append(noise_preds_row)
+            # Stitch noise predictions for all tiles
+            noise_pred = torch.zeros(latents.shape, device=self.device)
+            contributors = torch.zeros(latents.shape, device=self.device)
+            # Add each tile contribution to overall latents
+            for row in range(grid_rows):
+                for col in range(grid_cols):
+                    px_row_init, px_row_end, px_col_init, px_col_end = _tile2latent_indices(
+                        row, col, tile_width, tile_height, tile_row_overlap, tile_col_overlap
+                    )
+                    noise_pred[:, :, px_row_init:px_row_end, px_col_init:px_col_end] += (
+                        noise_preds[row][col] * tile_weights
+                    )
+                    contributors[:, :, px_row_init:px_row_end, px_col_init:px_col_end] += tile_weights
+            # Average overlapping areas with more than 1 contributor
+            noise_pred /= contributors
+
+            # compute the previous noisy sample x_t -> x_t-1
+            latents = self.scheduler.step(noise_pred, t, latents).prev_sample
+
+        # scale and decode the image latents with vae
+        image = self.decode_latents(latents, cpu_vae)
+
+        return {"images": image}
+
+    def _gaussian_weights(self, tile_width, tile_height, nbatches):
+        """Generates a gaussian mask of weights for tile contributions"""
+        import numpy as np
+        from numpy import exp, pi, sqrt
+
+        latent_width = tile_width // 8
+        latent_height = tile_height // 8
+
+        var = 0.01
+        midpoint = (latent_width - 1) / 2  # -1 because index goes from 0 to latent_width - 1
+        x_probs = [
+            exp(-(x - midpoint) * (x - midpoint) / (latent_width * latent_width) / (2 * var)) / sqrt(2 * pi * var)
+            for x in range(latent_width)
+        ]
+        midpoint = latent_height / 2
+        y_probs = [
+            exp(-(y - midpoint) * (y - midpoint) / (latent_height * latent_height) / (2 * var)) / sqrt(2 * pi * var)
+            for y in range(latent_height)
+        ]
+
+        weights = np.outer(y_probs, x_probs)
+        return torch.tile(torch.tensor(weights, device=self.device), (nbatches, self.unet.config.in_channels, 1, 1))

examples/community/stable_diffusion_controlnet_reference.py

@@ -1,6 +1,7 @@
 # Inspired by: https://github.com/Mikubill/sd-webui-controlnet/discussions/1236 and https://github.com/Mikubill/sd-webui-controlnet/discussions/1280
 from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
+import numpy as np
 import PIL.Image
 import torch
 
@@ -97,7 +98,14 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
     def __call__(
         self,
         prompt: Union[str, List[str]] = None,
-        image: Union[torch.FloatTensor, PIL.Image.Image, List[torch.FloatTensor], List[PIL.Image.Image]] = None,
+        image: Union[
+            torch.FloatTensor,
+            PIL.Image.Image,
+            np.ndarray,
+            List[torch.FloatTensor],
+            List[PIL.Image.Image],
+            List[np.ndarray],
+        ] = None,
         ref_image: Union[torch.FloatTensor, PIL.Image.Image] = None,
         height: Optional[int] = None,
         width: Optional[int] = None,
@@ -130,8 +138,8 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
             prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
                 instead.
-            image (`torch.FloatTensor`, `PIL.Image.Image`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`,
-                    `List[List[torch.FloatTensor]]`, or `List[List[PIL.Image.Image]]`):
+            image (`torch.FloatTensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.FloatTensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+                    `List[List[torch.FloatTensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
                 The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If
                 the type is specified as `Torch.FloatTensor`, it is passed to ControlNet as is. `PIL.Image.Image` can
                 also be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If
@@ -223,15 +231,12 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
             list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
             (nsfw) content, according to the `safety_checker`.
         """
-        # 0. Default height and width to unet
-        height, width = self._default_height_width(height, width, image)
+        assert reference_attn or reference_adain, "`reference_attn` or `reference_adain` must be True."
 
         # 1. Check inputs. Raise error if not correct
         self.check_inputs(
             prompt,
             image,
-            height,
-            width,
             callback_steps,
             negative_prompt,
             prompt_embeds,
@@ -266,6 +271,9 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
         guess_mode = guess_mode or global_pool_conditions
 
         # 3. Encode input prompt
+        text_encoder_lora_scale = (
+            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+        )
         prompt_embeds = self._encode_prompt(
             prompt,
             device,
@@ -274,6 +282,7 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
             negative_prompt,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
         )
 
         # 4. Prepare image
@@ -289,6 +298,7 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 do_classifier_free_guidance=do_classifier_free_guidance,
                 guess_mode=guess_mode,
             )
+            height, width = image.shape[-2:]
         elif isinstance(controlnet, MultiControlNetModel):
             images = []
 
@@ -308,6 +318,7 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 images.append(image_)
 
             image = images
+            height, width = image[0].shape[-2:]
         else:
             assert False
 
@@ -505,8 +516,8 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -545,8 +556,8 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -605,8 +616,8 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -642,8 +653,8 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -720,14 +731,15 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 # controlnet(s) inference
                 if guess_mode and do_classifier_free_guidance:
                     # Infer ControlNet only for the conditional batch.
-                    controlnet_latent_model_input = latents
+                    control_model_input = latents
+                    control_model_input = self.scheduler.scale_model_input(control_model_input, t)
                     controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
                 else:
-                    controlnet_latent_model_input = latent_model_input
+                    control_model_input = latent_model_input
                     controlnet_prompt_embeds = prompt_embeds
 
                 down_block_res_samples, mid_block_res_sample = self.controlnet(
-                    controlnet_latent_model_input,
+                    control_model_input,
                     t,
                     encoder_hidden_states=controlnet_prompt_embeds,
                     controlnet_cond=image,

examples/community/stable_diffusion_reference.py

@@ -9,6 +9,7 @@ from diffusers import StableDiffusionPipeline
 from diffusers.models.attention import BasicTransformerBlock
 from diffusers.models.unet_2d_blocks import CrossAttnDownBlock2D, CrossAttnUpBlock2D, DownBlock2D, UpBlock2D
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
+from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion import rescale_noise_cfg
 from diffusers.utils import PIL_INTERPOLATION, logging, randn_tensor
 
 
@@ -179,6 +180,7 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: int = 1,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        guidance_rescale: float = 0.0,
         attention_auto_machine_weight: float = 1.0,
         gn_auto_machine_weight: float = 1.0,
         style_fidelity: float = 0.5,
@@ -248,6 +250,11 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                 `self.processor` in
                 [diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
+            guidance_rescale (`float`, *optional*, defaults to 0.7):
+                Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
+                Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
+                [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+                Guidance rescale factor should fix overexposure when using zero terminal SNR.
             attention_auto_machine_weight (`float`):
                 Weight of using reference query for self attention's context.
                 If attention_auto_machine_weight=1.0, use reference query for all self attention's context.
@@ -295,6 +302,9 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
         do_classifier_free_guidance = guidance_scale > 1.0
 
         # 3. Encode input prompt
+        text_encoder_lora_scale = (
+            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+        )
         prompt_embeds = self._encode_prompt(
             prompt,
             device,
@@ -303,6 +313,7 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
             negative_prompt,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
         )
 
         # 4. Preprocess reference image
@@ -499,8 +510,8 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -539,8 +550,8 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -599,8 +610,8 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -636,8 +647,8 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 if MODE == "write":
                     if gn_auto_machine_weight >= self.gn_weight:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
-                        self.mean_bank.append(mean)
-                        self.var_bank.append(var)
+                        self.mean_bank.append([mean])
+                        self.var_bank.append([var])
                 if MODE == "read":
                     if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
                         var, mean = torch.var_mean(hidden_states, dim=(2, 3), keepdim=True, correction=0)
@@ -748,6 +759,10 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                     noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
+                if do_classifier_free_guidance and guidance_rescale > 0.0:
+                    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
                 # compute the previous noisy sample x_t -> x_t-1
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
 

examples/community/unclip_image_interpolation.py

@@ -376,14 +376,16 @@ class UnCLIPImageInterpolationPipeline(DiffusionPipeline):
         height = self.decoder.config.sample_size
         width = self.decoder.config.sample_size
 
+        # Get the decoder latents for 1 step and then repeat the same tensor for the entire batch to keep same noise across all interpolation steps.
         decoder_latents = self.prepare_latents(
-            (batch_size, num_channels_latents, height, width),
+            (1, num_channels_latents, height, width),
             text_encoder_hidden_states.dtype,
             device,
             generator,
             decoder_latents,
             self.decoder_scheduler,
         )
+        decoder_latents = decoder_latents.repeat((batch_size, 1, 1, 1))
 
         for i, t in enumerate(self.progress_bar(decoder_timesteps_tensor)):
             # expand the latents if we are doing classifier free guidance

examples/controlnet/train_controlnet.py

@@ -18,6 +18,7 @@ import logging
 import math
 import os
 import random
+import shutil
 from pathlib import Path
 
 import accelerate
@@ -55,7 +56,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.17.0.dev0")
+check_min_version("0.18.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -307,11 +308,7 @@ def parse_args(input_args=None):
         "--checkpoints_total_limit",
         type=int,
         default=None,
-        help=(
-            "Max number of checkpoints to store. Passed as `total_limit` to the `Accelerator` `ProjectConfiguration`."
-            " See Accelerator::save_state https://huggingface.co/docs/accelerate/package_reference/accelerator#accelerate.Accelerator.save_state"
-            " for more details"
-        ),
+        help=("Max number of checkpoints to store."),
     )
     parser.add_argument(
         "--resume_from_checkpoint",
@@ -716,13 +713,12 @@ def collate_fn(examples):
 def main(args):
     logging_dir = Path(args.output_dir, args.logging_dir)
 
-    accelerator_project_config = ProjectConfiguration(total_limit=args.checkpoints_total_limit)
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
 
     accelerator = Accelerator(
         gradient_accumulation_steps=args.gradient_accumulation_steps,
         mixed_precision=args.mixed_precision,
         log_with=args.report_to,
-        logging_dir=logging_dir,
         project_config=accelerator_project_config,
     )
 
@@ -1059,6 +1055,26 @@ def main(args):
 
                 if accelerator.is_main_process:
                     if global_step % args.checkpointing_steps == 0:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
                         save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
                         accelerator.save_state(save_path)
                         logger.info(f"Saved state to {save_path}")

examples/controlnet/train_controlnet_flax.py

@@ -59,7 +59,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.17.0.dev0")
+check_min_version("0.18.0.dev0")
 
 logger = logging.getLogger(__name__)
 

examples/custom_diffusion/train_custom_diffusion.py

@@ -21,6 +21,7 @@ import logging
 import math
 import os
 import random
+import shutil
 import warnings
 from pathlib import Path
 
@@ -56,7 +57,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.17.0.dev0")
+check_min_version("0.18.0.dev0")
 
 logger = get_logger(__name__)
 
@@ -446,11 +447,7 @@ def parse_args(input_args=None):
         "--checkpoints_total_limit",
         type=int,
         default=None,
-        help=(
-            "Max number of checkpoints to store. Passed as `total_limit` to the `Accelerator` `ProjectConfiguration`."
-            " See Accelerator::save_state https://huggingface.co/docs/accelerate/package_reference/accelerator#accelerate.Accelerator.save_state"
-            " for more docs"
-        ),
+        help=("Max number of checkpoints to store."),
     )
     parser.add_argument(
         "--resume_from_checkpoint",
@@ -637,13 +634,12 @@ def parse_args(input_args=None):
 def main(args):
     logging_dir = Path(args.output_dir, args.logging_dir)
 
-    accelerator_project_config = ProjectConfiguration(total_limit=args.checkpoints_total_limit)
+    accelerator_project_config = ProjectConfiguration(project_dir=args.output_dir, logging_dir=logging_dir)
 
     accelerator = Accelerator(
         gradient_accumulation_steps=args.gradient_accumulation_steps,
         mixed_precision=args.mixed_precision,
         log_with=args.report_to,
-        logging_dir=logging_dir,
         project_config=accelerator_project_config,
     )
 
@@ -1170,6 +1166,26 @@ def main(args):
 
                 if global_step % args.checkpointing_steps == 0:
                     if accelerator.is_main_process:
+                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
+                        if args.checkpoints_total_limit is not None:
+                            checkpoints = os.listdir(args.output_dir)
+                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
+                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
+
+                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
+                            if len(checkpoints) >= args.checkpoints_total_limit:
+                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
+                                removing_checkpoints = checkpoints[0:num_to_remove]
+
+                                logger.info(
+                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
+                                )
+                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
+
+                                for removing_checkpoint in removing_checkpoints:
+                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
+                                    shutil.rmtree(removing_checkpoint)
+
                         save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
                         accelerator.save_state(save_path)
                         logger.info(f"Saved state to {save_path}")

examples/dreambooth/README.md

@@ -536,9 +536,68 @@ You can refer to [this blog post](https://huggingface.co/blog/dreambooth) that d
 
 ## IF
 
-You can use the lora and full dreambooth scripts to also train the text to image [IF model](https://huggingface.co/DeepFloyd/IF-I-XL-v1.0). A few alternative cli flags are needed due to the model size, the expected input resolution, and the text encoder conventions.
+You can use the lora and full dreambooth scripts to train the text to image [IF model](https://huggingface.co/DeepFloyd/IF-I-XL-v1.0) and the stage II upscaler 
+[IF model](https://huggingface.co/DeepFloyd/IF-II-L-v1.0).
 
-### LoRA Dreambooth
+Note that IF has a predicted variance, and our finetuning scripts only train the models predicted error, so for finetuned IF models we switch to a fixed
+variance schedule. The full finetuning scripts will update the scheduler config for the full saved model. However, when loading saved LoRA weights, you
+must also update the pipeline's scheduler config.
+
+```py
+from diffusers import DiffusionPipeline
+
+pipe = DiffusionPipeline.from_pretrained("DeepFloyd/IF-I-XL-v1.0")
+
+pipe.load_lora_weights("<lora weights path>")
+
+# Update scheduler config to fixed variance schedule
+pipe.scheduler = pipe.scheduler.__class__.from_config(pipe.scheduler.config, variance_type="fixed_small")
+```
+
+Additionally, a few alternative cli flags are needed for IF.
+
+`--resolution=64`: IF is a pixel space diffusion model. In order to operate on un-compressed pixels, the input images are of a much smaller resolution. 
+
+`--pre_compute_text_embeddings`: IF uses [T5](https://huggingface.co/docs/transformers/model_doc/t5) for its text encoder. In order to save GPU memory, we pre compute all text embeddings and then de-allocate
+T5.
+
+`--tokenizer_max_length=77`: T5 has a longer default text length, but the default IF encoding procedure uses a smaller number.
+
+`--text_encoder_use_attention_mask`: T5 passes the attention mask to the text encoder.
+
+### Tips and Tricks
+We find LoRA to be sufficient for finetuning the stage I model as the low resolution of the model makes representing finegrained detail hard regardless.
+
+For common and/or not-visually complex object concepts, you can get away with not-finetuning the upscaler. Just be sure to adjust the prompt passed to the
+upscaler to remove the new token from the instance prompt. I.e. if your stage I prompt is "a sks dog", use "a dog" for your stage II prompt.
+
+For finegrained detail like faces that aren't present in the original training set, we find that full finetuning of the stage II upscaler is better than 
+LoRA finetuning stage II.
+
+For finegrained detail like faces, we find that lower learning rates along with larger batch sizes work best.
+
+For stage II, we find that lower learning rates are also needed.
+
+We found experimentally that the DDPM scheduler with the default larger number of denoising steps to sometimes work better than the DPM Solver scheduler
+used in the training scripts.
+
+### Stage II additional validation images
+
+The stage II validation requires images to upscale, we can download a downsized version of the training set:
+
+```py
+from huggingface_hub import snapshot_download
+
+local_dir = "./dog_downsized"
+snapshot_download(
+    "diffusers/dog-example-downsized",
+    local_dir=local_dir,
+    repo_type="dataset",
+    ignore_patterns=".gitattributes",
+)
+```
+
+### IF stage I LoRA Dreambooth
 This training configuration requires ~28 GB VRAM.
 
 ```sh
@@ -552,7 +611,7 @@ accelerate launch train_dreambooth_lora.py \
   --instance_data_dir=$INSTANCE_DIR \
   --output_dir=$OUTPUT_DIR \
   --instance_prompt="a sks dog" \
-  --resolution=64 \ # The input resolution of the IF unet is 64x64
+  --resolution=64 \
   --train_batch_size=4 \
   --gradient_accumulation_steps=1 \
   --learning_rate=5e-6 \
@@ -561,16 +620,58 @@ accelerate launch train_dreambooth_lora.py \
   --validation_prompt="a sks dog" \
   --validation_epochs=25 \
   --checkpointing_steps=100 \
-  --pre_compute_text_embeddings \ # Pre compute text embeddings to that T5 doesn't have to be kept in memory
-  --tokenizer_max_length=77 \ # IF expects an override of the max token length
-  --text_encoder_use_attention_mask # IF expects attention mask for text embeddings
+  --pre_compute_text_embeddings \
+  --tokenizer_max_length=77 \
+  --text_encoder_use_attention_mask
 ```
 
-### Full Dreambooth
-Due to the size of the optimizer states, we recommend training the full XL IF model with 8bit adam. 
-Using 8bit adam and the rest of the following config, the model can be trained in ~48 GB VRAM.
+### IF stage II LoRA Dreambooth
 
-For full dreambooth, IF requires very low learning rates. With higher learning rates model quality will degrade.
+`--validation_images`: These images are upscaled during validation steps.
+
+`--class_labels_conditioning=timesteps`: Pass additional conditioning to the UNet needed for stage II.
+
+`--learning_rate=1e-6`: Lower learning rate than stage I.
+
+`--resolution=256`: The upscaler expects higher resolution inputs
+
+```sh
+export MODEL_NAME="DeepFloyd/IF-II-L-v1.0"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="dreambooth_dog_upscale"
+export VALIDATION_IMAGES="dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"
+
+python train_dreambooth_lora.py \
+    --report_to wandb \
+    --pretrained_model_name_or_path=$MODEL_NAME \
+    --instance_data_dir=$INSTANCE_DIR \
+    --output_dir=$OUTPUT_DIR \
+    --instance_prompt="a sks dog" \
+    --resolution=256 \
+    --train_batch_size=4 \
+    --gradient_accumulation_steps=1 \
+    --learning_rate=1e-6 \ 
+    --max_train_steps=2000 \
+    --validation_prompt="a sks dog" \
+    --validation_epochs=100 \
+    --checkpointing_steps=500 \
+    --pre_compute_text_embeddings \
+    --tokenizer_max_length=77 \
+    --text_encoder_use_attention_mask \
+    --validation_images $VALIDATION_IMAGES \
+    --class_labels_conditioning=timesteps
+```
+
+### IF Stage I Full Dreambooth
+`--skip_save_text_encoder`: When training the full model, this will skip saving the entire T5 with the finetuned model. You can still load the pipeline
+with a T5 loaded from the original model.
+
+`use_8bit_adam`: Due to the size of the optimizer states, we recommend training the full XL IF model with 8bit adam. 
+
+`--learning_rate=1e-7`: For full dreambooth, IF requires very low learning rates. With higher learning rates model quality will degrade. Note that it is 
+likely the learning rate can be increased with larger batch sizes.
+
+Using 8bit adam and a batch size of 4, the model can be trained in ~48 GB VRAM.
 
 ```sh
 export MODEL_NAME="DeepFloyd/IF-I-XL-v1.0"
@@ -583,17 +684,56 @@ accelerate launch train_dreambooth.py \
   --instance_data_dir=$INSTANCE_DIR \
   --output_dir=$OUTPUT_DIR \
   --instance_prompt="a photo of sks dog" \
-  --resolution=64 \ # The input resolution of the IF unet is 64x64
+  --resolution=64 \
   --train_batch_size=4 \
   --gradient_accumulation_steps=1 \
   --learning_rate=1e-7 \
   --max_train_steps=150 \
   --validation_prompt "a photo of sks dog" \
   --validation_steps 25 \
-  --text_encoder_use_attention_mask \ # IF expects attention mask for text embeddings
-  --tokenizer_max_length 77 \ # IF expects an override of the max token length
-  --pre_compute_text_embeddings \ # Pre compute text embeddings to that T5 doesn't have to be kept in memory
-  --use_8bit_adam \ # 
+  --text_encoder_use_attention_mask \
+  --tokenizer_max_length 77 \
+  --pre_compute_text_embeddings \
+  --use_8bit_adam \
   --set_grads_to_none \
-  --skip_save_text_encoder # do not save the full T5 text encoder with the model
+  --skip_save_text_encoder \
+  --push_to_hub
+```
+
+### IF Stage II Full Dreambooth
+
+`--learning_rate=5e-6`: With a smaller effective batch size of 4, we found that we required learning rates as low as
+1e-8.
+
+`--resolution=256`: The upscaler expects higher resolution inputs
+
+`--train_batch_size=2` and `--gradient_accumulation_steps=6`: We found that full training of stage II particularly with
+faces required large effective batch sizes.
+
+```sh
+export MODEL_NAME="DeepFloyd/IF-II-L-v1.0"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="dreambooth_dog_upscale"
+export VALIDATION_IMAGES="dog_downsized/image_1.png dog_downsized/image_2.png dog_downsized/image_3.png dog_downsized/image_4.png"
+
+accelerate launch train_dreambooth.py \
+  --report_to wandb \
+  --pretrained_model_name_or_path=$MODEL_NAME \
+  --instance_data_dir=$INSTANCE_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --instance_prompt="a sks dog" \
+  --resolution=256 \
+  --train_batch_size=2 \
+  --gradient_accumulation_steps=6 \
+  --learning_rate=5e-6 \
+  --max_train_steps=2000 \
+  --validation_prompt="a sks dog" \
+  --validation_steps=150 \
+  --checkpointing_steps=500 \
+  --pre_compute_text_embeddings \
+  --tokenizer_max_length=77 \
+  --text_encoder_use_attention_mask \
+  --validation_images $VALIDATION_IMAGES \
+  --class_labels_conditioning timesteps \
+  --push_to_hub
 ```