trailing . fix

* fix imports

* better error

* up

* finish

.github/workflows/pr_tests.yml

@@ -60,6 +60,7 @@ jobs:
       run: |
         python -m pip install -e .[quality,test]
         python -m pip install git+https://github.com/huggingface/accelerate
+        python -m pip install -U git+https://github.com/huggingface/transformers
 
     - name: Environment
       run: |
@@ -127,6 +128,7 @@ jobs:
         ${CONDA_RUN} python -m pip install -e .[quality,test]
         ${CONDA_RUN} python -m pip install --pre torch==${MPS_TORCH_VERSION} --extra-index-url https://download.pytorch.org/whl/test/cpu
         ${CONDA_RUN} python -m pip install git+https://github.com/huggingface/accelerate
+        ${CONDA_RUN} python -m pip install -U git+https://github.com/huggingface/transformers
 
     - name: Environment
       shell: arch -arch arm64 bash {0}

.github/workflows/push_tests.yml

@@ -62,6 +62,7 @@ jobs:
       run: |
         python -m pip install -e .[quality,test]
         python -m pip install git+https://github.com/huggingface/accelerate
+        python -m pip install -U git+https://github.com/huggingface/transformers
 
     - name: Environment
       run: |
@@ -131,6 +132,7 @@ jobs:
       run: |
         python -m pip install -e .[quality,test,training]
         python -m pip install git+https://github.com/huggingface/accelerate
+        python -m pip install -U git+https://github.com/huggingface/transformers
 
     - name: Environment
       run: |

docs/source/_toctree.yml

@@ -106,6 +106,8 @@
       title: "Score SDE VE"
     - local: api/pipelines/stable_diffusion
       title: "Stable Diffusion"
+    - local: api/pipelines/stable_diffusion_safe
+      title: "Safe Stable Diffusion"
     - local: api/pipelines/stochastic_karras_ve
       title: "Stochastic Karras VE"
     - local: api/pipelines/dance_diffusion

docs/source/api/pipelines/overview.mdx

@@ -58,7 +58,11 @@ available a colab notebook to directly try them out.
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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/training_example.ipynb)
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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)
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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)
-| [stochastic_karras_ve](./api/pipelines/stochastic_karras_ve) | [**Elucidating the Design Space of Diffusion-Based Generative Models**](https://arxiv.org/abs/2206.00364) | Unconditional Image Generation | 
+| [stable_diffusion_safe](./api/pipelines/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)
+| [stochastic_karras_ve](./api/pipelines/stochastic_karras_ve) | [**Elucidating the Design Space of Diffusion-Based Generative Models**](https://arxiv.org/abs/2206.00364) | Unconditional Image Generation |
+| [versatile_diffusion](./api/pipelines/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](./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 | 
 
 

docs/source/api/pipelines/stable_diffusion.mdx

@@ -88,3 +88,10 @@ If you want to use all possible use cases in a single `DiffusionPipeline` you ca
 	- __call__
 	- enable_attention_slicing
 	- disable_attention_slicing
+
+
+## StableDiffusionImageVariationPipeline
+[[autodoc]] StableDiffusionImageVariationPipeline
+	- __call__
+	- enable_attention_slicing
+	- disable_attention_slicing

docs/source/api/pipelines/stable_diffusion_safe.mdx

@@ -0,0 +1,90 @@
+<!--Copyright 2022 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.
+-->
+
+# Safe Stable Diffusion
+
+Safe Stable Diffusion was proposed in [Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models](https://arxiv.org/abs/2211.05105) and mitigates the well known issue that models like Stable Diffusion that are trained on unfiltered, web-crawled datasets tend to suffer from inappropriate degeneration. For instance Stable Diffusion may unexpectedly generate nudity, violence, images depicting self-harm, or otherwise offensive content.
+Safe Stable Diffusion is an extension to the Stable Diffusion that drastically reduces content like this.
+
+The abstract of the paper is the following:
+
+*Text-conditioned image generation models have recently achieved astonishing results in image quality and text alignment and are consequently employed in a fast-growing number of applications. Since they are highly data-driven, relying on billion-sized datasets randomly scraped from the internet, they also suffer, as we demonstrate, from degenerated and biased human behavior. In turn, they may even reinforce such biases. To help combat these undesired side effects, we present safe latent diffusion (SLD). Specifically, to measure the inappropriate degeneration due to unfiltered and imbalanced training sets, we establish a novel image generation test bed-inappropriate image prompts (I2P)-containing dedicated, real-world image-to-text prompts covering concepts such as nudity and violence. As our exhaustive empirical evaluation demonstrates, the introduced SLD removes and suppresses inappropriate image parts during the diffusion process, with no additional training required and no adverse effect on overall image quality or text alignment.*
+
+
+*Overview*:
+
+| Pipeline | Tasks | Colab | Demo
+|---|---|:---:|:---:|
+| [pipeline_stable_diffusion_safe.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py) | *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) | -
+
+## Tips
+
+- Safe Stable Diffusion may also be used with weights of [Stable Diffusion](./api/pipelines/stable_diffusion).
+
+### Run Safe Stable Diffusion
+
+Safe Stable Diffusion can be tested very easily with the [`StableDiffusionPipelineSafe`], and the `"AIML-TUDA/stable-diffusion-safe"` checkpoint exactly in the same way it is shown in the [Conditional Image Generation Guide](./using-diffusers/conditional_image_generation).
+
+### Interacting with the Safety Concept
+
+To check and edit the currently used safety concept, use the `safety_concept` property of [`StableDiffusionPipelineSafe`]
+```python
+>>> from diffusers import StableDiffusionPipelineSafe
+
+>>> pipeline = StableDiffusionPipelineSafe.from_pretrained("AIML-TUDA/stable-diffusion-safe")
+>>> pipeline.safety_concept
+```
+For each image generation the active concept is also contained in [`StableDiffusionSafePipelineOutput`].
+
+### Using pre-defined safety configurations
+
+You may use the 4 configurations defined in the [Safe Latent Diffusion paper](https://arxiv.org/abs/2211.05105) as follows:
+
+```python
+>>> from diffusers import StableDiffusionPipelineSafe
+>>> from diffusers.pipelines.stable_diffusion_safe import SafetyConfig
+
+>>> pipeline = StableDiffusionPipelineSafe.from_pretrained("AIML-TUDA/stable-diffusion-safe")
+>>> prompt = "the four horsewomen of the apocalypse, painting by tom of finland, gaston bussiere, craig mullins, j. c. leyendecker"
+>>> out = pipeline(prompt=prompt, **SafetyConfig.MAX)
+```
+
+The following configurations are available: `SafetyConfig.WEAK`, `SafetyConfig.MEDIUM`, `SafetyConfig.STRONg`, and `SafetyConfig.MAX`.
+
+### How to load and use different schedulers.
+
+The safe stable diffusion pipeline uses [`PNDMScheduler`] scheduler by default. But `diffusers` provides many other schedulers that can be used with the stable diffusion pipeline such as [`DDIMScheduler`], [`LMSDiscreteScheduler`], [`EulerDiscreteScheduler`], [`EulerAncestralDiscreteScheduler`] etc.
+To use a different scheduler, you can either change it via the [`ConfigMixin.from_config`] method or pass the `scheduler` argument to the `from_pretrained` method of the pipeline. For example, to use the [`EulerDiscreteScheduler`], you can do the following:
+
+```python
+>>> from diffusers import StableDiffusionPipelineSafe, EulerDiscreteScheduler
+
+>>> pipeline = StableDiffusionPipelineSafe.from_pretrained("AIML-TUDA/stable-diffusion-safe")
+>>> pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
+
+>>> # or
+>>> euler_scheduler = EulerDiscreteScheduler.from_pretrained("AIML-TUDA/stable-diffusion-safe", subfolder="scheduler")
+>>> pipeline = StableDiffusionPipelineSafe.from_pretrained(
+...     "AIML-TUDA/stable-diffusion-safe", scheduler=euler_scheduler
+... )
+```
+
+
+## StableDiffusionSafePipelineOutput
+[[autodoc]] pipelines.stable_diffusion_safe.StableDiffusionSafePipelineOutput
+
+## StableDiffusionPipelineSafe
+[[autodoc]] StableDiffusionPipelineSafe
+	- __call__
+	- enable_attention_slicing
+	- disable_attention_slicing
+

docs/source/api/pipelines/versatile_diffusion.mdx

@@ -18,65 +18,56 @@ The abstract of the paper is the following:
 
 *The recent advances in diffusion models have set an impressive milestone in many generation tasks. Trending works such as DALL-E2, Imagen, and Stable Diffusion have attracted great interest in academia and industry. Despite the rapid landscape changes, recent new approaches focus on extensions and performance rather than capacity, thus requiring separate models for separate tasks. In this work, we expand the existing single-flow diffusion pipeline into a multi-flow network, dubbed Versatile Diffusion (VD), that handles text-to-image, image-to-text, image-variation, and text-variation in one unified model. Moreover, we generalize VD to a unified multi-flow multimodal diffusion framework with grouped layers, swappable streams, and other propositions that can process modalities beyond images and text. Through our experiments, we demonstrate that VD and its underlying framework have the following merits: a) VD handles all subtasks with competitive quality; b) VD initiates novel extensions and applications such as disentanglement of style and semantic, image-text dual-guided generation, etc.; c) Through these experiments and applications, VD provides more semantic insights of the generated outputs.*
 
-*Overview*:
-
-| Pipeline | Tasks | Colab | Demo
-|---|---|:---:|:---:|
-| [pipeline_alt_diffusion.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py) | *Text-to-Image Generation* | - | -
-| [pipeline_alt_diffusion_img2img.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py) | *Image-to-Image Text-Guided Generation* | - |-
-
 ## Tips
 
 - VersatileDiffusion is conceptually very similar as [Stable Diffusion](./api/pipelines/stable_diffusion),  but instead of providing just a image data stream conditioned on text, VersatileDiffusion provides both a image and text data stream and can be conditioned on both text and image.
 
-- *Run VersatileDiffusion*
+### *Run VersatileDiffusion*
 
-All task VersatileDiffusion can be tested very easily with the [`VersatileDiffusionPipeline`], [`VersatileDiffusionImg2ImgPipeline`] and the `"BAAI/VersatileDiffusion-m9"` checkpoint exactly in the same way it is shown in the [Conditional Image Generation Guide](./using-diffusers/conditional_image_generation) and the [Image-to-Image Generation Guide](./using-diffusers/img2img).
+You can both load the memory intensive "all-in-one" [`VersatileDiffusionPipeline`] that can run all tasks 
+with the same class as shown in [`VersatileDiffusionPipeline.text_to_image`], [`VersatileDiffusionPipeline.image_variation`], and [`VersatileDiffusionPipeline.dual_guided`]
 
-- *How to load and use different schedulers.*
+**or**
 
-The alt diffusion pipeline uses [`DDIMScheduler`] scheduler by default. But `diffusers` provides many other schedulers that can be used with the alt diffusion pipeline such as [`PNDMScheduler`], [`LMSDiscreteScheduler`], [`EulerDiscreteScheduler`], [`EulerAncestralDiscreteScheduler`] etc.
+You can run the individual pipelines which are much more memory efficient:
+
+- *Text-to-Image*: [`VersatileDiffusionTextToImagePipeline.__call__`]
+- *Image Variation*: [`VersatileDiffusionImageVariationPipeline.__call__`]
+- *Dual Text and Image Guided Generation*: [`VersatileDiffusionDualGuidedPipeline.__call__`]
+
+### *How to load and use different schedulers.*
+
+The versatile diffusion pipelines uses [`DDIMScheduler`] scheduler by default. But `diffusers` provides many other schedulers that can be used with the alt diffusion pipeline such as [`PNDMScheduler`], [`LMSDiscreteScheduler`], [`EulerDiscreteScheduler`], [`EulerAncestralDiscreteScheduler`] etc.
 To use a different scheduler, you can either change it via the [`ConfigMixin.from_config`] method or pass the `scheduler` argument to the `from_pretrained` method of the pipeline. For example, to use the [`EulerDiscreteScheduler`], you can do the following:
 
 ```python
 >>> from diffusers import VersatileDiffusionPipeline, EulerDiscreteScheduler
 
->>> pipeline = VersatileDiffusionPipeline.from_pretrained("BAAI/VersatileDiffusion-m9")
+>>> pipeline = VersatileDiffusionPipeline.from_pretrained("shi-labs/versatile-diffusion")
 >>> pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
 
 >>> # or
->>> euler_scheduler = EulerDiscreteScheduler.from_pretrained("BAAI/VersatileDiffusion-m9", subfolder="scheduler")
->>> pipeline = VersatileDiffusionPipeline.from_pretrained("BAAI/VersatileDiffusion-m9", scheduler=euler_scheduler)
+>>> euler_scheduler = EulerDiscreteScheduler.from_pretrained("shi-labs/versatile-diffusion", subfolder="scheduler")
+>>> pipeline = VersatileDiffusionPipeline.from_pretrained("shi-labs/versatile-diffusion", scheduler=euler_scheduler)
 ```
 
-
-- *How to conver all use cases with multiple or single pipeline*
-
-If you want to use all possible use cases in a single `DiffusionPipeline` we recommend using the `components` functionality to instantiate all components in the most memory-efficient way:
-
-```python
->>> from diffusers import (
-...     VersatileDiffusionPipeline,
-...     VersatileDiffusionImg2ImgPipeline,
-... )
-
->>> text2img = VersatileDiffusionPipeline.from_pretrained("BAAI/VersatileDiffusion-m9")
->>> img2img = VersatileDiffusionImg2ImgPipeline(**text2img.components)
-
->>> # now you can use text2img(...) and img2img(...) just like the call methods of each respective pipeline
-```
-
-## VersatileDiffusionPipelineOutput
-[[autodoc]] pipelines.alt_diffusion.VersatileDiffusionPipelineOutput
-
 ## VersatileDiffusionPipeline
 [[autodoc]] VersatileDiffusionPipeline
+
+## VersatileDiffusionTextToImagePipeline
+[[autodoc]] VersatileDiffusionTextToImagePipeline
 	- __call__
 	- enable_attention_slicing
 	- disable_attention_slicing
 
-## VersatileDiffusionImg2ImgPipeline
-[[autodoc]] VersatileDiffusionImg2ImgPipeline
+## VersatileDiffusionImageVariationPipeline
+[[autodoc]] VersatileDiffusionImageVariationPipeline
+	- __call__
+	- enable_attention_slicing
+	- disable_attention_slicing
+
+## VersatileDiffusionDualGuidedPipeline
+[[autodoc]] VersatileDiffusionDualGuidedPipeline
 	- __call__
 	- enable_attention_slicing
 	- disable_attention_slicing

docs/source/index.mdx

@@ -48,7 +48,11 @@ available a colab notebook to directly try them out.
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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/training_example.ipynb)
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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)
 | [stable_diffusion](./api/pipelines/stable_diffusion) | [**Stable Diffusion**](https://stability.ai/blog/stable-diffusion-public-release) | 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)
+| [stable_diffusion_safe](./api/pipelines/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)
 | [stochastic_karras_ve](./api/pipelines/stochastic_karras_ve) | [**Elucidating the Design Space of Diffusion-Based Generative Models**](https://arxiv.org/abs/2206.00364) | Unconditional Image Generation | 
+| [versatile_diffusion](./api/pipelines/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](./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 | 
 
 **Note**: Pipelines are simple examples of how to play around with the diffusion systems as described in the corresponding papers. 

examples/community/README.md

@@ -22,6 +22,7 @@ If a community doesn't work as expected, please open an issue and ping the autho
 | Image to Image Inpainting Stable Diffusion | Stable Diffusion Pipeline that enables the overlaying of two images and subsequent inpainting| [Image to Image Inpainting Stable Diffusion](#image-to-image-inpainting-stable-diffusion)                                                                 | -                                                                                                                                                                                                                  |                        [Alex McKinney](https://github.com/vvvm23) |
 | Text Based Inpainting Stable Diffusion | Stable Diffusion Inpainting Pipeline that enables passing a text prompt to generate the mask for inpainting| [Text Based Inpainting Stable Diffusion](#image-to-image-inpainting-stable-diffusion)                                                                 | -                                                                                                                                                                                                                  |                        [Dhruv Karan](https://github.com/unography) |
 | Bit Diffusion | Diffusion on discrete data | [Bit Diffusion](#bit-diffusion) | -  |[Stuti R.](https://github.com/kingstut) |
+| K-Diffusion Stable Diffusion | Run Stable Diffusion with any of [K-Diffusion's samplers](https://github.com/crowsonkb/k-diffusion/blob/master/k_diffusion/sampling.py) | [Stable Diffusion with K Diffusion](#stable-diffusion-with-k-diffusion) | -  | [Patrick von Platen](https://github.com/patrickvonplaten/) |
 
 
 
@@ -663,4 +664,65 @@ Based https://arxiv.org/abs/2208.04202, this is used for diffusion on discrete d
 from diffusers import DiffusionPipeline
 pipe = DiffusionPipeline.from_pretrained("google/ddpm-cifar10-32", custom_pipeline="bit_diffusion")
 image = pipe().images[0]
+
 ```
+
+### Stable Diffusion with K Diffusion
+
+Make sure you have @crowsonkb's https://github.com/crowsonkb/k-diffusion installed:
+
+```
+pip install k-diffusion
+```
+
+You can use the community pipeline as follows:
+
+```python
+from diffusers import DiffusionPipeline
+
+pipe = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", custom_pipeline="sd_text2img_k_diffusion")
+pipe = pipe.to("cuda")
+
+prompt = "an astronaut riding a horse on mars"
+pipe.set_sampler("sample_heun")
+generator = torch.Generator(device="cuda").manual_seed(seed)
+image = pipe(prompt, generator=generator, num_inference_steps=20).images[0]
+
+image.save("./astronaut_heun_k_diffusion.png")
+```
+
+To make sure that K Diffusion and `diffusers` yield the same results:
+
+**Diffusers**:
+```python
+from diffusers import DiffusionPipeline, EulerDiscreteScheduler
+
+seed = 33
+
+pipe = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
+pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config)
+pipe = pipe.to("cuda")
+
+generator = torch.Generator(device="cuda").manual_seed(seed)
+image = pipe(prompt, generator=generator, num_inference_steps=50).images[0]
+```
+
+![diffusers_euler](https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/k_diffusion/astronaut_euler.png)
+
+**K Diffusion**:
+```python
+from diffusers import DiffusionPipeline, EulerDiscreteScheduler
+
+seed = 33
+
+pipe = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", custom_pipeline="sd_text2img_k_diffusion")
+pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config)
+pipe = pipe.to("cuda")
+
+pipe.set_sampler("sample_euler")
+generator = torch.Generator(device="cuda").manual_seed(seed)
+image = pipe(prompt, generator=generator, num_inference_steps=50).images[0]
+```
+
+![diffusers_euler](https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/k_diffusion/astronaut_euler_k_diffusion.png)
+

examples/community/img2img_inpainting.py

@@ -110,7 +110,7 @@ class ImageToImageInpaintingPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/interpolate_stable_diffusion.py

@@ -101,7 +101,7 @@ class StableDiffusionWalkPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/lpw_stable_diffusion.py

@@ -469,7 +469,7 @@ class StableDiffusionLongPromptWeightingPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/multilingual_stable_diffusion.py

@@ -113,7 +113,7 @@ class MultilingualStableDiffusion(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/sd_text2img_k_diffusion.py

@@ -0,0 +1,479 @@
+# Copyright 2022 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.
+
+import importlib
+from typing import Callable, List, Optional, Union
+
+import torch
+
+from diffusers import LMSDiscreteScheduler
+from diffusers.pipeline_utils import DiffusionPipeline
+from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
+from diffusers.utils import is_accelerate_available, logging
+from k_diffusion.external import CompVisDenoiser
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+class ModelWrapper:
+    def __init__(self, model, alphas_cumprod):
+        self.model = model
+        self.alphas_cumprod = alphas_cumprod
+
+    def apply_model(self, *args, **kwargs):
+        return self.model(*args, **kwargs).sample
+
+
+class StableDiffusionPipeline(DiffusionPipeline):
+    r"""
+    Pipeline for text-to-image generation using Stable Diffusion.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`CLIPTextModel`]):
+            Frozen text-encoder. Stable Diffusion uses the text portion of
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offensive or harmful.
+            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
+        feature_extractor ([`CLIPFeatureExtractor`]):
+            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
+    """
+
+    def __init__(
+        self,
+        vae,
+        text_encoder,
+        tokenizer,
+        unet,
+        scheduler,
+        safety_checker,
+        feature_extractor,
+    ):
+        super().__init__()
+
+        if safety_checker is None:
+            logger.warning(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
+        # get correct sigmas from LMS
+        scheduler = LMSDiscreteScheduler.from_config(scheduler.config)
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+
+        model = ModelWrapper(unet, scheduler.alphas_cumprod)
+        self.k_diffusion_model = CompVisDenoiser(model)
+
+    def set_sampler(self, scheduler_type: str):
+        library = importlib.import_module("k_diffusion")
+        sampling = getattr(library, "sampling")
+        self.sampler = getattr(sampling, scheduler_type)
+
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(True)
+
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(False)
+
+    def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
+        r"""
+        Enable sliced attention computation.
+
+        When this option is enabled, the attention module will split the input tensor in slices, to compute attention
+        in several steps. This is useful to save some memory in exchange for a small speed decrease.
+
+        Args:
+            slice_size (`str` or `int`, *optional*, defaults to `"auto"`):
+                When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
+                a number is provided, uses as many slices as `attention_head_dim // slice_size`. In this case,
+                `attention_head_dim` must be a multiple of `slice_size`.
+        """
+        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):
+        r"""
+        Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
+        back to computing attention in one step.
+        """
+        # set slice_size = `None` to disable `attention slicing`
+        self.enable_attention_slicing(None)
+
+    def enable_sequential_cpu_offload(self, gpu_id=0):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        """
+        if is_accelerate_available():
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("Please install accelerate via `pip install accelerate`")
+
+        device = torch.device(f"cuda:{gpu_id}")
+
+        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.safety_checker]:
+            if cpu_offloaded_model is not None:
+                cpu_offload(cpu_offloaded_model, device)
+
+    @property
+    def _execution_device(self):
+        r"""
+        Returns the device on which the pipeline's models will be executed. After calling
+        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
+        hooks.
+        """
+        if self.device != torch.device("meta") or not hasattr(self.unet, "_hf_hook"):
+            return self.device
+        for module in self.unet.modules():
+            if (
+                hasattr(module, "_hf_hook")
+                and hasattr(module._hf_hook, "execution_device")
+                and module._hf_hook.execution_device is not None
+            ):
+                return torch.device(module._hf_hook.execution_device)
+        return self.device
+
+    def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+
+        Args:
+            prompt (`str` or `list(int)`):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            do_classifier_free_guidance (`bool`):
+                whether to use classifier free guidance or not
+            negative_prompt (`str` or `List[str]`):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+        """
+        batch_size = len(prompt) if isinstance(prompt, list) else 1
+
+        text_inputs = self.tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer(prompt, padding="max_length", return_tensors="pt").input_ids
+
+        if not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+            )
+
+        if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+            attention_mask = text_inputs.attention_mask.to(device)
+        else:
+            attention_mask = None
+
+        text_embeddings = self.text_encoder(
+            text_input_ids.to(device),
+            attention_mask=attention_mask,
+        )
+        text_embeddings = text_embeddings[0]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        bs_embed, seq_len, _ = text_embeddings.shape
+        text_embeddings = text_embeddings.repeat(1, num_images_per_prompt, 1)
+        text_embeddings = text_embeddings.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance:
+            uncond_tokens: List[str]
+            if negative_prompt is None:
+                uncond_tokens = [""] * batch_size
+            elif type(prompt) is not type(negative_prompt):
+                raise TypeError(
+                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                    f" {type(prompt)}."
+                )
+            elif isinstance(negative_prompt, str):
+                uncond_tokens = [negative_prompt]
+            elif batch_size != len(negative_prompt):
+                raise ValueError(
+                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                    " the batch size of `prompt`."
+                )
+            else:
+                uncond_tokens = negative_prompt
+
+            max_length = text_input_ids.shape[-1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = uncond_input.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            uncond_embeddings = self.text_encoder(
+                uncond_input.input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            uncond_embeddings = uncond_embeddings[0]
+
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = uncond_embeddings.shape[1]
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+            # 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])
+
+        return text_embeddings
+
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    def check_inputs(self, prompt, height, width, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        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 (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // 8, width // 8)
+        if latents is None:
+            if device.type == "mps":
+                # randn does not work reproducibly on mps
+                latents = torch.randn(shape, generator=generator, device="cpu", dtype=dtype).to(device)
+            else:
+                latents = torch.randn(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            if latents.shape != shape:
+                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        return latents
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        prompt: Union[str, List[str]],
+        height: int = 512,
+        width: int = 512,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 7.5,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[torch.Generator] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: Optional[int] = 1,
+        **kwargs,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`):
+                The prompt or prompts to guide the image generation.
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(prompt, height, width, callback_steps)
+
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
+        device = self._execution_device
+        # 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 = True
+        if guidance_scale <= 1.0:
+            raise ValueError("has to use guidance_scale")
+
+        # 3. Encode input prompt
+        text_embeddings = self._encode_prompt(
+            prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
+        )
+
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=text_embeddings.device)
+        sigmas = self.scheduler.sigmas
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            text_embeddings.dtype,
+            device,
+            generator,
+            latents,
+        )
+        latents = latents * sigmas[0]
+        self.k_diffusion_model.sigmas = self.k_diffusion_model.sigmas.to(latents.device)
+        self.k_diffusion_model.log_sigmas = self.k_diffusion_model.log_sigmas.to(latents.device)
+
+        def model_fn(x, t):
+            latent_model_input = torch.cat([x] * 2)
+
+            noise_pred = self.k_diffusion_model(latent_model_input, t, encoder_hidden_states=text_embeddings)
+
+            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+            noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+            return noise_pred
+
+        latents = self.sampler(model_fn, latents, sigmas)
+
+        # 8. Post-processing
+        image = self.decode_latents(latents)
+
+        # 9. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, text_embeddings.dtype)
+
+        # 10. Convert to PIL
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

examples/community/speech_to_image_diffusion.py

@@ -42,7 +42,7 @@ class SpeechToImagePipeline(DiffusionPipeline):
         super().__init__()
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/text_inpainting.py

@@ -99,7 +99,7 @@ class TextInpainting(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/community/wildcard_stable_diffusion.py

@@ -135,7 +135,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

examples/dreambooth/README.md

@@ -141,7 +141,7 @@ export INSTANCE_DIR="path-to-instance-images"
 export CLASS_DIR="path-to-class-images"
 export OUTPUT_DIR="path-to-save-model"
 
-accelerate launch train_dreambooth.py \
+accelerate launch --mixed_precision="fp16" train_dreambooth.py \
   --pretrained_model_name_or_path=$MODEL_NAME \
   --instance_data_dir=$INSTANCE_DIR \
   --class_data_dir=$CLASS_DIR \
@@ -157,8 +157,7 @@ accelerate launch train_dreambooth.py \
   --lr_scheduler="constant" \
   --lr_warmup_steps=0 \
   --num_class_images=200 \
-  --max_train_steps=800 \
-  --mixed_precision=fp16
+  --max_train_steps=800
 ```
 
 ### Fine-tune text encoder with the UNet.

examples/dreambooth/train_dreambooth.py

@@ -187,12 +187,12 @@ def parse_args(input_args=None):
     parser.add_argument(
         "--mixed_precision",
         type=str,
-        default="no",
+        default=None,
         choices=["no", "fp16", "bf16"],
         help=(
-            "Whether to use mixed precision. Choose"
-            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
-            "and an Nvidia Ampere GPU."
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
         ),
     )
     parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
@@ -538,9 +538,9 @@ def main(args):
         )
 
     weight_dtype = torch.float32
-    if args.mixed_precision == "fp16":
+    if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16
-    elif args.mixed_precision == "bf16":
+    elif accelerator.mixed_precision == "bf16":
         weight_dtype = torch.bfloat16
 
     # Move text_encode and vae to gpu.

examples/text_to_image/README.md

@@ -46,7 +46,7 @@ With `gradient_checkpointing` and `mixed_precision` it should be possible to fin
 export MODEL_NAME="CompVis/stable-diffusion-v1-4"
 export dataset_name="lambdalabs/pokemon-blip-captions"
 
-accelerate launch train_text_to_image.py \
+accelerate launch --mixed_precision="fp16"  train_text_to_image.py \
   --pretrained_model_name_or_path=$MODEL_NAME \
   --dataset_name=$dataset_name \
   --use_ema \
@@ -54,7 +54,6 @@ accelerate launch train_text_to_image.py \
   --train_batch_size=1 \
   --gradient_accumulation_steps=4 \
   --gradient_checkpointing \
-  --mixed_precision="fp16" \
   --max_train_steps=15000 \
   --learning_rate=1e-05 \
   --max_grad_norm=1 \
@@ -70,7 +69,7 @@ If you wish to use custom loading logic, you should modify the script, we have l
 export MODEL_NAME="CompVis/stable-diffusion-v1-4"
 export TRAIN_DIR="path_to_your_dataset"
 
-accelerate launch train_text_to_image.py \
+accelerate launch --mixed_precision="fp16" train_text_to_image.py \
   --pretrained_model_name_or_path=$MODEL_NAME \
   --train_data_dir=$TRAIN_DIR \
   --use_ema \
@@ -78,7 +77,6 @@ accelerate launch train_text_to_image.py \
   --train_batch_size=1 \
   --gradient_accumulation_steps=4 \
   --gradient_checkpointing \
-  --mixed_precision="fp16" \
   --max_train_steps=15000 \
   --learning_rate=1e-05 \
   --max_grad_norm=1 \

examples/text_to_image/train_text_to_image.py

@@ -186,12 +186,12 @@ def parse_args():
     parser.add_argument(
         "--mixed_precision",
         type=str,
-        default="no",
+        default=None,
         choices=["no", "fp16", "bf16"],
         help=(
-            "Whether to use mixed precision. Choose"
-            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
-            "and an Nvidia Ampere GPU."
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
         ),
     )
     parser.add_argument(
@@ -496,9 +496,9 @@ def main():
     )
 
     weight_dtype = torch.float32
-    if args.mixed_precision == "fp16":
+    if accelerator.mixed_precision == "fp16":
         weight_dtype = torch.float16
-    elif args.mixed_precision == "bf16":
+    elif accelerator.mixed_precision == "bf16":
         weight_dtype = torch.bfloat16
 
     # Move text_encode and vae to gpu.

setup.py

@@ -212,7 +212,7 @@ install_requires = [
 
 setup(
     name="diffusers",
-    version="0.8.0.dev0",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="0.8.1",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     description="Diffusers",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/diffusers/__init__.py

@@ -9,7 +9,7 @@ from .utils import (
 )
 
 
-__version__ = "0.8.0.dev0"
+__version__ = "0.8.1"
 
 from .configuration_utils import ConfigMixin
 from .onnx_utils import OnnxRuntimeModel
@@ -69,12 +69,13 @@ if is_torch_available() and is_transformers_available():
         AltDiffusionPipeline,
         CycleDiffusionPipeline,
         LDMTextToImagePipeline,
+        StableDiffusionImageVariationPipeline,
         StableDiffusionImg2ImgPipeline,
         StableDiffusionInpaintPipeline,
         StableDiffusionInpaintPipelineLegacy,
         StableDiffusionPipeline,
+        StableDiffusionPipelineSafe,
         VersatileDiffusionDualGuidedPipeline,
-        VersatileDiffusionImageToTextPipeline,
         VersatileDiffusionImageVariationPipeline,
         VersatileDiffusionPipeline,
         VersatileDiffusionTextToImagePipeline,

src/diffusers/modeling_utils.py

@@ -332,7 +332,7 @@ class ModelMixin(torch.nn.Module):
 
         if low_cpu_mem_usage and not is_accelerate_available():
             low_cpu_mem_usage = False
-            logger.warn(
+            logger.warning(
                 "Cannot initialize model with low cpu memory usage because `accelerate` was not found in the"
                 " environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install"
                 " `accelerate` for faster and less memory-intense model loading. You can do so with: \n```\npip"

src/diffusers/models/attention.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import math
+import warnings
 from dataclasses import dataclass
 from typing import Optional
 
@@ -22,7 +23,7 @@ from torch import nn
 from ..configuration_utils import ConfigMixin, register_to_config
 from ..modeling_utils import ModelMixin
 from ..models.embeddings import ImagePositionalEmbeddings
-from ..utils import CONFIG_NAME, BaseOutput
+from ..utils import BaseOutput
 from ..utils.import_utils import is_xformers_available
 
 
@@ -284,22 +285,52 @@ class AttentionBlock(nn.Module):
         key_proj = self.key(hidden_states)
         value_proj = self.value(hidden_states)
 
-        # transpose
-        query_states = self.transpose_for_scores(query_proj)
-        key_states = self.transpose_for_scores(key_proj)
-        value_states = self.transpose_for_scores(value_proj)
+        scale = 1 / math.sqrt(self.channels / self.num_heads)
 
         # get scores
-        scale = 1 / math.sqrt(math.sqrt(self.channels / self.num_heads))
-        attention_scores = torch.matmul(query_states * scale, key_states.transpose(-1, -2) * scale)  # TODO: use baddmm
+        if self.num_heads > 1:
+            query_states = self.transpose_for_scores(query_proj)
+            key_states = self.transpose_for_scores(key_proj)
+            value_states = self.transpose_for_scores(value_proj)
+
+            # TODO: is there a way to perform batched matmul (e.g. baddbmm) on 4D tensors?
+            #       or reformulate this into a 3D problem?
+            # TODO: measure whether on MPS device it would be faster to do this matmul via einsum
+            #       as some matmuls can be 1.94x slower than an equivalent einsum on MPS
+            #       https://gist.github.com/Birch-san/cba16789ec27bb20996a4b4831b13ce0
+            attention_scores = torch.matmul(query_states, key_states.transpose(-1, -2)) * scale
+        else:
+            query_states, key_states, value_states = query_proj, key_proj, value_proj
+
+            attention_scores = torch.baddbmm(
+                torch.empty(
+                    query_states.shape[0],
+                    query_states.shape[1],
+                    key_states.shape[1],
+                    dtype=query_states.dtype,
+                    device=query_states.device,
+                ),
+                query_states,
+                key_states.transpose(-1, -2),
+                beta=0,
+                alpha=scale,
+            )
+
         attention_probs = torch.softmax(attention_scores.float(), dim=-1).type(attention_scores.dtype)
 
         # compute attention output
-        hidden_states = torch.matmul(attention_probs, value_states)
-
-        hidden_states = hidden_states.permute(0, 2, 1, 3).contiguous()
-        new_hidden_states_shape = hidden_states.size()[:-2] + (self.channels,)
-        hidden_states = hidden_states.view(new_hidden_states_shape)
+        if self.num_heads > 1:
+            # TODO: is there a way to perform batched matmul (e.g. bmm) on 4D tensors?
+            #       or reformulate this into a 3D problem?
+            # TODO: measure whether on MPS device it would be faster to do this matmul via einsum
+            #       as some matmuls can be 1.94x slower than an equivalent einsum on MPS
+            #       https://gist.github.com/Birch-san/cba16789ec27bb20996a4b4831b13ce0
+            hidden_states = torch.matmul(attention_probs, value_states)
+            hidden_states = hidden_states.permute(0, 2, 1, 3).contiguous()
+            new_hidden_states_shape = hidden_states.size()[:-2] + (self.channels,)
+            hidden_states = hidden_states.view(new_hidden_states_shape)
+        else:
+            hidden_states = torch.bmm(attention_probs, value_states)
 
         # compute next hidden_states
         hidden_states = self.proj_attn(hidden_states)
@@ -366,6 +397,16 @@ class BasicTransformerBlock(nn.Module):
             self.norm2 = nn.LayerNorm(dim)
         self.norm3 = nn.LayerNorm(dim)
 
+        # if xformers is installed try to use memory_efficient_attention by default
+        if is_xformers_available():
+            try:
+                self._set_use_memory_efficient_attention_xformers(True)
+            except Exception as e:
+                warnings.warn(
+                    "Could not enable memory efficient attention. Make sure xformers is installed"
+                    f" correctly and a GPU is available: {e}"
+                )
+
     def _set_attention_slice(self, slice_size):
         self.attn1._slice_size = slice_size
         self.attn2._slice_size = slice_size
@@ -507,19 +548,17 @@ class CrossAttention(nn.Module):
         return hidden_states
 
     def _attention(self, query, key, value):
-        # TODO: use baddbmm for better performance
-        if query.device.type == "mps":
-            # Better performance on mps (~20-25%)
-            attention_scores = torch.einsum("b i d, b j d -> b i j", query, key) * self.scale
-        else:
-            attention_scores = torch.matmul(query, key.transpose(-1, -2)) * self.scale
+        attention_scores = torch.baddbmm(
+            torch.empty(query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device),
+            query,
+            key.transpose(-1, -2),
+            beta=0,
+            alpha=self.scale,
+        )
         attention_probs = attention_scores.softmax(dim=-1)
         # compute attention output
 
-        if query.device.type == "mps":
-            hidden_states = torch.einsum("b i j, b j d -> b i d", attention_probs, value)
-        else:
-            hidden_states = torch.matmul(attention_probs, value)
+        hidden_states = torch.bmm(attention_probs, value)
 
         # reshape hidden_states
         hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
@@ -534,21 +573,15 @@ class CrossAttention(nn.Module):
         for i in range(hidden_states.shape[0] // slice_size):
             start_idx = i * slice_size
             end_idx = (i + 1) * slice_size
-            if query.device.type == "mps":
-                # Better performance on mps (~20-25%)
-                attn_slice = (
-                    torch.einsum("b i d, b j d -> b i j", query[start_idx:end_idx], key[start_idx:end_idx])
-                    * self.scale
-                )
-            else:
-                attn_slice = (
-                    torch.matmul(query[start_idx:end_idx], key[start_idx:end_idx].transpose(1, 2)) * self.scale
-                )  # TODO: use baddbmm for better performance
+            attn_slice = torch.baddbmm(
+                torch.empty(slice_size, query.shape[1], key.shape[1], dtype=query.dtype, device=query.device),
+                query[start_idx:end_idx],
+                key[start_idx:end_idx].transpose(-1, -2),
+                beta=0,
+                alpha=self.scale,
+            )
             attn_slice = attn_slice.softmax(dim=-1)
-            if query.device.type == "mps":
-                attn_slice = torch.einsum("b i j, b j d -> b i d", attn_slice, value[start_idx:end_idx])
-            else:
-                attn_slice = torch.matmul(attn_slice, value[start_idx:end_idx])
+            attn_slice = torch.bmm(attn_slice, value[start_idx:end_idx])
 
             hidden_states[start_idx:end_idx] = attn_slice
 
@@ -731,12 +764,18 @@ class DualTransformer2DModel(nn.Module):
             ]
         )
 
+        # Variables that can be set by a pipeline:
+
         # The ratio of transformer1 to transformer2's output states to be combined during inference
         self.mix_ratio = 0.5
 
         # The shape of `encoder_hidden_states` is expected to be
-        # `(batch_size, num_condition_tokens[0]+num_condition_tokens[1], num_features)`
-        self.num_condition_tokens = (77, 257)
+        # `(batch_size, condition_lengths[0]+condition_lengths[1], num_features)`
+        self.condition_lengths = [77, 257]
+
+        # Which transformer to use to encode which condition.
+        # E.g. `(1, 0)` means that we'll use `transformers[1](conditions[0])` and `transformers[0](conditions[1])`
+        self.transformer_index_for_condition = [1, 0]
 
     def forward(self, hidden_states, encoder_hidden_states, timestep=None, return_dict: bool = True):
         """
@@ -763,10 +802,13 @@ class DualTransformer2DModel(nn.Module):
         tokens_start = 0
         for i in range(2):
             # for each of the two transformers, pass the corresponding condition tokens
-            condition_state = encoder_hidden_states[:, tokens_start : tokens_start + self.num_condition_tokens[i]]
-            encoded_state = self.transformers[i](input_states, condition_state, timestep, return_dict)[0]
+            condition_state = encoder_hidden_states[:, tokens_start : tokens_start + self.condition_lengths[i]]
+            transformer_index = self.transformer_index_for_condition[i]
+            encoded_state = self.transformers[transformer_index](input_states, condition_state, timestep, return_dict)[
+                0
+            ]
             encoded_states.append(encoded_state - input_states)
-            tokens_start += self.num_condition_tokens[i]
+            tokens_start += self.condition_lengths[i]
 
         output_states = encoded_states[0] * self.mix_ratio + encoded_states[1] * (1 - self.mix_ratio)
         output_states = output_states + input_states

src/diffusers/models/unet_2d.py

@@ -43,8 +43,8 @@ class UNet2DModel(ModelMixin, ConfigMixin):
     implements for all the model (such as downloading or saving, etc.)
 
     Parameters:
-        sample_size (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`, *optional*):
-            Input sample size.
+        sample_size (`int` or `Tuple[int, int]`, *optional*, defaults to `None`):
+            Height and width of input/output sample.
         in_channels (`int`, *optional*, defaults to 3): Number of channels in the input image.
         out_channels (`int`, *optional*, defaults to 3): Number of channels in the output.
         center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.
@@ -71,7 +71,7 @@ class UNet2DModel(ModelMixin, ConfigMixin):
     @register_to_config
     def __init__(
         self,
-        sample_size: Optional[int] = None,
+        sample_size: Optional[Union[int, Tuple[int, int]]] = None,
         in_channels: int = 3,
         out_channels: int = 3,
         center_input_sample: bool = False,
@@ -209,6 +209,11 @@ class UNet2DModel(ModelMixin, ConfigMixin):
         timesteps = timesteps * torch.ones(sample.shape[0], dtype=timesteps.dtype, device=timesteps.device)
 
         t_emb = self.time_proj(timesteps)
+
+        # timesteps does not contain any weights and will always return f32 tensors
+        # but time_embedding might actually be running in fp16. so we need to cast here.
+        # there might be better ways to encapsulate this.
+        t_emb = t_emb.to(dtype=self.dtype)
         emb = self.time_embedding(t_emb)
 
         # 2. pre-process
@@ -242,9 +247,7 @@ class UNet2DModel(ModelMixin, ConfigMixin):
                 sample = upsample_block(sample, res_samples, emb)
 
         # 6. post-process
-        # make sure hidden states is in float32
-        # when running in half-precision
-        sample = self.conv_norm_out(sample.float()).type(sample.dtype)
+        sample = self.conv_norm_out(sample)
         sample = self.conv_act(sample)
         sample = self.conv_out(sample)
 

src/diffusers/models/unet_2d_blocks.py

@@ -15,7 +15,7 @@ import numpy as np
 import torch
 from torch import nn
 
-from .attention import AttentionBlock, Transformer2DModel, DualTransformer2DModel
+from .attention import AttentionBlock, DualTransformer2DModel, Transformer2DModel
 from .resnet import Downsample2D, FirDownsample2D, FirUpsample2D, ResnetBlock2D, Upsample2D
 
 
@@ -169,6 +169,7 @@ def get_up_block(
             resnet_groups=resnet_groups,
             cross_attention_dim=cross_attention_dim,
             attn_num_head_channels=attn_num_head_channels,
+            dual_cross_attention=dual_cross_attention,
         )
     elif up_block_type == "AttnUpBlock2D":
         return AttnUpBlock2D(
@@ -352,16 +353,28 @@ class UNetMidBlock2DCrossAttn(nn.Module):
         attentions = []
 
         for _ in range(num_layers):
-            attentions.append(
-                Transformer2DModel(
-                    attn_num_head_channels,
-                    in_channels // attn_num_head_channels,
-                    in_channels=in_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
+            if not dual_cross_attention:
+                attentions.append(
+                    Transformer2DModel(
+                        attn_num_head_channels,
+                        in_channels // attn_num_head_channels,
+                        in_channels=in_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
+                )
+            else:
+                attentions.append(
+                    DualTransformer2DModel(
+                        attn_num_head_channels,
+                        in_channels // attn_num_head_channels,
+                        in_channels=in_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
                 )
-            )
             resnets.append(
                 ResnetBlock2D(
                     in_channels=in_channels,
@@ -534,7 +547,7 @@ class CrossAttnDownBlock2D(nn.Module):
                     pre_norm=resnet_pre_norm,
                 )
             )
-            if dual_cross_attention is False:
+            if not dual_cross_attention:
                 attentions.append(
                     Transformer2DModel(
                         attn_num_head_channels,
@@ -1106,6 +1119,7 @@ class CrossAttnUpBlock2D(nn.Module):
         attention_type="default",
         output_scale_factor=1.0,
         add_upsample=True,
+        dual_cross_attention=False,
     ):
         super().__init__()
         resnets = []
@@ -1132,16 +1146,28 @@ class CrossAttnUpBlock2D(nn.Module):
                     pre_norm=resnet_pre_norm,
                 )
             )
-            attentions.append(
-                Transformer2DModel(
-                    attn_num_head_channels,
-                    out_channels // attn_num_head_channels,
-                    in_channels=out_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
+            if not dual_cross_attention:
+                attentions.append(
+                    Transformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
+                )
+            else:
+                attentions.append(
+                    DualTransformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
                 )
-            )
         self.attentions = nn.ModuleList(attentions)
         self.resnets = nn.ModuleList(resnets)
 

src/diffusers/models/unet_2d_condition.py

@@ -56,7 +56,8 @@ class UNet2DConditionModel(ModelMixin, ConfigMixin):
     implements for all the models (such as downloading or saving, etc.)
 
     Parameters:
-        sample_size (`int`, *optional*): The size of the input sample.
+        sample_size (`int` or `Tuple[int, int]`, *optional*, defaults to `None`):
+            Height and width of input/output sample.
         in_channels (`int`, *optional*, defaults to 4): The number of channels in the input sample.
         out_channels (`int`, *optional*, defaults to 4): The number of channels in the output.
         center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.

src/diffusers/pipeline_flax_utils.py

@@ -411,13 +411,13 @@ class FlaxDiffusionPipeline(ConfigMixin):
                             f" {expected_class_obj}"
                         )
                 elif passed_class_obj[name] is None:
-                    logger.warn(
+                    logger.warning(
                         f"You have passed `None` for {name} to disable its functionality in {pipeline_class}. Note"
                         f" that this might lead to problems when using {pipeline_class} and is not recommended."
                     )
                     sub_model_should_be_defined = False
                 else:
-                    logger.warn(
+                    logger.warning(
                         f"You have passed a non-standard module {passed_class_obj[name]}. We cannot verify whether it"
                         " has the correct type"
                     )

src/diffusers/pipeline_utils.py

@@ -405,7 +405,7 @@ class DiffusionPipeline(ConfigMixin):
 
         if low_cpu_mem_usage and not is_accelerate_available():
             low_cpu_mem_usage = False
-            logger.warn(
+            logger.warning(
                 "Cannot initialize model with low cpu memory usage because `accelerate` was not found in the"
                 " environment. Defaulting to `low_cpu_mem_usage=False`. It is strongly recommended to install"
                 " `accelerate` for faster and less memory-intense model loading. You can do so with: \n```\npip"
@@ -571,13 +571,13 @@ class DiffusionPipeline(ConfigMixin):
                             f" {expected_class_obj}"
                         )
                 elif passed_class_obj[name] is None:
-                    logger.warn(
+                    logger.warning(
                         f"You have passed `None` for {name} to disable its functionality in {pipeline_class}. Note"
                         f" that this might lead to problems when using {pipeline_class} and is not recommended."
                     )
                     sub_model_should_be_defined = False
                 else:
-                    logger.warn(
+                    logger.warning(
                         f"You have passed a non-standard module {passed_class_obj[name]}. We cannot verify whether it"
                         " has the correct type"
                     )

src/diffusers/pipelines/__init__.py

@@ -19,14 +19,15 @@ if is_torch_available() and is_transformers_available():
     from .latent_diffusion import LDMTextToImagePipeline
     from .stable_diffusion import (
         CycleDiffusionPipeline,
+        StableDiffusionImageVariationPipeline,
         StableDiffusionImg2ImgPipeline,
         StableDiffusionInpaintPipeline,
         StableDiffusionInpaintPipelineLegacy,
         StableDiffusionPipeline,
     )
+    from .stable_diffusion_safe import StableDiffusionPipelineSafe
     from .versatile_diffusion import (
         VersatileDiffusionDualGuidedPipeline,
-        VersatileDiffusionImageToTextPipeline,
         VersatileDiffusionImageVariationPipeline,
         VersatileDiffusionPipeline,
         VersatileDiffusionTextToImagePipeline,

src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py

@@ -115,7 +115,7 @@ class AltDiffusionPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Alt Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py

@@ -128,7 +128,7 @@ class AltDiffusionImg2ImgPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Alt Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/ddim/pipeline_ddim.py

@@ -89,7 +89,11 @@ class DDIMPipeline(DiffusionPipeline):
             generator = None
 
         # Sample gaussian noise to begin loop
-        image_shape = (batch_size, self.unet.in_channels, self.unet.sample_size, self.unet.sample_size)
+        if isinstance(self.unet.sample_size, int):
+            image_shape = (batch_size, self.unet.in_channels, self.unet.sample_size, self.unet.sample_size)
+        else:
+            image_shape = (batch_size, self.unet.in_channels, *self.unet.sample_size)
+
         if self.device.type == "mps":
             # randn does not work reproducibly on mps
             image = torch.randn(image_shape, generator=generator)

src/diffusers/pipelines/ddpm/pipeline_ddpm.py

@@ -94,7 +94,11 @@ class DDPMPipeline(DiffusionPipeline):
             generator = None
 
         # Sample gaussian noise to begin loop
-        image_shape = (batch_size, self.unet.in_channels, self.unet.sample_size, self.unet.sample_size)
+        if isinstance(self.unet.sample_size, int):
+            image_shape = (batch_size, self.unet.in_channels, self.unet.sample_size, self.unet.sample_size)
+        else:
+            image_shape = (batch_size, self.unet.in_channels, *self.unet.sample_size)
+
         if self.device.type == "mps":
             # randn does not work reproducibly on mps
             image = torch.randn(image_shape, generator=generator)

src/diffusers/pipelines/stable_diffusion/__init__.py

@@ -6,7 +6,14 @@ import numpy as np
 import PIL
 from PIL import Image
 
-from ...utils import BaseOutput, is_flax_available, is_onnx_available, is_torch_available, is_transformers_available
+from ...utils import (
+    BaseOutput,
+    is_flax_available,
+    is_onnx_available,
+    is_torch_available,
+    is_transformers_available,
+    is_transformers_version,
+)
 
 
 @dataclass
@@ -35,6 +42,11 @@ if is_transformers_available() and is_torch_available():
     from .pipeline_stable_diffusion_inpaint_legacy import StableDiffusionInpaintPipelineLegacy
     from .safety_checker import StableDiffusionSafetyChecker
 
+if is_transformers_available() and is_torch_available() and is_transformers_version(">=", "4.25.0.dev0"):
+    from .pipeline_stable_diffusion_image_variation import StableDiffusionImageVariationPipeline
+else:
+    from ...utils.dummy_torch_and_transformers_objects import StableDiffusionImageVariationPipeline
+
 if is_transformers_available() and is_onnx_available():
     from .pipeline_onnx_stable_diffusion import OnnxStableDiffusionPipeline, StableDiffusionOnnxPipeline
     from .pipeline_onnx_stable_diffusion_img2img import OnnxStableDiffusionImg2ImgPipeline

src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py

@@ -160,7 +160,7 @@ class CycleDiffusionPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py

@@ -88,7 +88,7 @@ class FlaxStableDiffusionPipeline(FlaxDiffusionPipeline):
         self.dtype = dtype
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -114,7 +114,7 @@ class StableDiffusionPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py

@@ -13,100 +13,95 @@
 # limitations under the License.
 
 import inspect
-from dataclasses import dataclass
 from typing import Callable, List, Optional, Union
 
-import numpy as np
 import torch
-import torch.utils.checkpoint
 
 import PIL
-from transformers import CLIPFeatureExtractor, CLIPVisionModelWithProjection, GPT2Tokenizer
+from diffusers.utils import is_accelerate_available
+from transformers import CLIPFeatureExtractor, CLIPVisionModelWithProjection
 
 from ...models import AutoencoderKL, UNet2DConditionModel
-from ...models.attention import Transformer2DModel
-from ...pipeline_utils import BaseOutput, DiffusionPipeline
-from ...schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
-from ...utils import is_accelerate_available, logging
-from .modeling_gpt2_optimus import GPT2OptimusForLatentConnector
-from .modeling_text_unet import UNetFlatConditionModel
+from ...pipeline_utils import DiffusionPipeline
+from ...schedulers import (
+    DDIMScheduler,
+    DPMSolverMultistepScheduler,
+    EulerAncestralDiscreteScheduler,
+    EulerDiscreteScheduler,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+)
+from ...utils import logging
+from . import StableDiffusionPipelineOutput
+from .safety_checker import StableDiffusionSafetyChecker
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 
-@dataclass
-class TextPipelineOutput(BaseOutput):
-    """
-    Output class for text generation pipelines.
-
-    Args:
-        text (`List[str]` or `np.ndarray`)
-            List of generated text of length `batch_size` or a numpy array of tokens of shape `(batch_size,
-            num_tokens)`.
-    """
-
-    text: Union[List[str], np.ndarray]
-
-
-class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
+class StableDiffusionImageVariationPipeline(DiffusionPipeline):
     r"""
+    Pipeline to generate variations from an input image using Stable Diffusion.
+
     This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
     library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
 
-    Parameters:
-        vqvae ([`VQModel`]):
-            Vector-quantized (VQ) Model to encode and decode images to and from latent representations.
-        bert ([`LDMBertModel`]):
-            Text-encoder model based on [BERT](https://huggingface.co/docs/transformers/model_doc/bert) architecture.
-        tokenizer (`transformers.BertTokenizer`):
-            Tokenizer of class
-            [BertTokenizer](https://huggingface.co/docs/transformers/model_doc/bert#transformers.BertTokenizer).
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        image_encoder ([`CLIPVisionModelWithProjection`]):
+            Frozen CLIP image-encoder. Stable Diffusion Image Variation uses the vision portion of
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPVisionModelWithProjection),
+            specifically the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
         unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
         scheduler ([`SchedulerMixin`]):
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offensive or harmful.
+            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
+        feature_extractor ([`CLIPFeatureExtractor`]):
+            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
-    image_feature_extractor: CLIPFeatureExtractor
-    image_encoder: CLIPVisionModelWithProjection
-    image_unet: UNet2DConditionModel
-    text_unet: UNetFlatConditionModel
-    vae: AutoencoderKL
-    scheduler: Union[DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler]
 
     def __init__(
         self,
-        image_feature_extractor: CLIPFeatureExtractor,
-        image_encoder: CLIPVisionModelWithProjection,
-        image_unet: UNet2DConditionModel,
-        text_unet: UNetFlatConditionModel,
         vae: AutoencoderKL,
-        scheduler: Union[DDIMScheduler, PNDMScheduler, LMSDiscreteScheduler],
+        image_encoder: CLIPVisionModelWithProjection,
+        unet: UNet2DConditionModel,
+        scheduler: Union[
+            DDIMScheduler,
+            PNDMScheduler,
+            LMSDiscreteScheduler,
+            EulerDiscreteScheduler,
+            EulerAncestralDiscreteScheduler,
+            DPMSolverMultistepScheduler,
+        ],
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
     ):
         super().__init__()
+
+        if safety_checker is None:
+            logger.warn(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
         self.register_modules(
-            image_feature_extractor=image_feature_extractor,
-            image_encoder=image_encoder,
-            image_unet=image_unet,
-            text_unet=text_unet,
             vae=vae,
+            image_encoder=image_encoder,
+            unet=unet,
             scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
         )
 
-        self.text_vae_decoder = GPT2OptimusForLatentConnector.from_pretrained("fusing/gpt2_optimus")
-        self.text_vae_tokenizer = GPT2Tokenizer.from_pretrained("fusing/gpt2_optimus")
-
-    def swap_unet_attention_blocks(self):
-        for name, module in self.image_unet.named_modules():
-            if isinstance(module, Transformer2DModel):
-                parent_name, index = name.rsplit(".", 1)
-                index = int(index)
-                self.image_unet.get_submodule(parent_name)[index], self.text_unet.get_submodule(parent_name)[index] = (
-                    self.text_unet.get_submodule(parent_name)[index],
-                    self.image_unet.get_submodule(parent_name)[index],
-                )
-
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention with unet->image_unet
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention
     def enable_xformers_memory_efficient_attention(self):
         r"""
         Enable memory efficient attention as implemented in xformers.
@@ -117,16 +112,16 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
         Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
         is used.
         """
-        self.image_unet.set_use_memory_efficient_attention_xformers(True)
+        self.unet.set_use_memory_efficient_attention_xformers(True)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention with unet->image_unet
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_xformers_memory_efficient_attention
     def disable_xformers_memory_efficient_attention(self):
         r"""
         Disable memory efficient attention as implemented in xformers.
         """
-        self.image_unet.set_use_memory_efficient_attention_xformers(False)
+        self.unet.set_use_memory_efficient_attention_xformers(False)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing with unet->image_unet
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_attention_slicing
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -143,8 +138,8 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
         if slice_size == "auto":
             # half the attention head size is usually a good trade-off between
             # speed and memory
-            slice_size = self.image_unet.config.attention_head_dim // 2
-        self.image_unet.set_attention_slice(slice_size)
+            slice_size = self.unet.config.attention_head_dim // 2
+        self.unet.set_attention_slice(slice_size)
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_attention_slicing
     def disable_attention_slicing(self):
@@ -168,21 +163,21 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
 
         device = torch.device(f"cuda:{gpu_id}")
 
-        for cpu_offloaded_model in [self.image_unet, self.text_unet, self.text_encoder, self.vae]:
+        for cpu_offloaded_model in [self.unet, self.image_encoder, self.vae, self.safety_checker]:
             if cpu_offloaded_model is not None:
                 cpu_offload(cpu_offloaded_model, device)
 
     @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device with unet->image_unet
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
     def _execution_device(self):
         r"""
         Returns the device on which the pipeline's models will be executed. After calling
         `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
         hooks.
         """
-        if self.device != torch.device("meta") or not hasattr(self.image_unet, "_hf_hook"):
+        if self.device != torch.device("meta") or not hasattr(self.unet, "_hf_hook"):
             return self.device
-        for module in self.image_unet.modules():
+        for module in self.unet.modules():
             if (
                 hasattr(module, "_hf_hook")
                 and hasattr(module._hf_hook, "execution_device")
@@ -191,87 +186,50 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
                 return torch.device(module._hf_hook.execution_device)
         return self.device
 
-    def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
-        r"""
-        Encodes the prompt into text encoder hidden states.
+    def _encode_image(self, image, device, num_images_per_prompt, do_classifier_free_guidance):
+        dtype = next(self.image_encoder.parameters()).dtype
 
-        Args:
-            prompt (`str` or `list(int)`):
-                prompt to be encoded
-            device: (`torch.device`):
-                torch device
-            num_images_per_prompt (`int`):
-                number of images that should be generated per prompt
-            do_classifier_free_guidance (`bool`):
-                whether to use classifier free guidance or not
-            negative_prompt (`str` or `List[str]`):
-                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
-                if `guidance_scale` is less than `1`).
-        """
+        if not isinstance(image, torch.Tensor):
+            image = self.feature_extractor(images=image, return_tensors="pt").pixel_values
 
-        def normalize_embeddings(encoder_output):
-            embeds = self.image_encoder.vision_model.post_layernorm(encoder_output.last_hidden_state)
-            embeds = self.image_encoder.visual_projection(embeds)
-            embeds_pooled = embeds[:, 0:1]
-            embeds = embeds / torch.norm(embeds_pooled, dim=-1, keepdim=True)
-            return embeds
-
-        batch_size = len(prompt) if isinstance(prompt, list) else 1
-
-        # get prompt text embeddings
-        # prompt = [(np.asarray(prompt) / 255)]
-        image_input = self.image_feature_extractor(images=prompt, return_tensors="pt")
-        image_embeddings = self.image_encoder(image_input.pixel_values.to(self.device))
-        image_embeddings = normalize_embeddings(image_embeddings)
+        image = image.to(device=device, dtype=dtype)
+        image_embeddings = self.image_encoder(image).image_embeds
+        image_embeddings = image_embeddings.unsqueeze(1)
 
         # duplicate image embeddings for each generation per prompt, using mps friendly method
         bs_embed, seq_len, _ = image_embeddings.shape
         image_embeddings = image_embeddings.repeat(1, num_images_per_prompt, 1)
         image_embeddings = image_embeddings.view(bs_embed * num_images_per_prompt, seq_len, -1)
 
-        # get unconditional embeddings for classifier free guidance
         if do_classifier_free_guidance:
-            uncond_images: List[str]
-            if negative_prompt is None:
-                uncond_images = [np.zeros((512, 512, 3)) + 0.5] * batch_size
-            elif type(prompt) is not type(negative_prompt):
-                raise TypeError(
-                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
-                    f" {type(prompt)}."
-                )
-            elif isinstance(negative_prompt, PIL.Image.Image):
-                uncond_images = [negative_prompt]
-            elif batch_size != len(negative_prompt):
-                raise ValueError(
-                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
-                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
-                    " the batch size of `prompt`."
-                )
-            else:
-                uncond_images = negative_prompt
-
-            uncond_images = self.image_feature_extractor(images=uncond_images, return_tensors="pt")
-            uncond_embeddings = self.image_encoder(uncond_images.pixel_values.to(self.device))
-            uncond_embeddings = normalize_embeddings(uncond_embeddings)
-
-            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
-            seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
-            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
+            uncond_embeddings = torch.zeros_like(image_embeddings)
 
             # For classifier free guidance, we need to do two forward passes.
-            # Here we concatenate the unconditional and conditional embeddings into a single batch
+            # Here we concatenate the unconditional and text embeddings into a single batch
             # to avoid doing two forward passes
             image_embeddings = torch.cat([uncond_embeddings, image_embeddings])
 
         return image_embeddings
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        else:
+            has_nsfw_concept = None
+        return image, has_nsfw_concept
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
     def decode_latents(self, latents):
-        latents = latents.reshape(latents.shape[:-2])
-        self.text_vae_decoder = self.text_vae_decoder.to(self._execution_device)
-        bos_token = self.text_vae_tokenizer.bos_token_id
-        output = self.text_vae_decoder.generate(bos_token_id=bos_token, past=latents)
-        return output
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
     def prepare_extra_step_kwargs(self, generator, eta):
@@ -291,9 +249,18 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
             extra_step_kwargs["generator"] = generator
         return extra_step_kwargs
 
-    def check_inputs(self, image, callback_steps):
-        if not isinstance(image, PIL.Image.Image) and not isinstance(image, torch.Tensor):
-            raise ValueError(f"`image` has to be of type `PIL.Image.Image` or `torch.Tensor` but is {type(image)}")
+    def check_inputs(self, image, height, width, callback_steps):
+        if (
+            not isinstance(image, torch.Tensor)
+            and not isinstance(image, PIL.Image.Image)
+            and not isinstance(image, list)
+        ):
+            raise ValueError(
+                f"`image` has to be of type `torch.FloatTensor` or `PIL.Image.Image` or `list` but is {type(image)}"
+            )
+
+        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 (callback_steps is None) or (
             callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
@@ -303,8 +270,9 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
                 f" {type(callback_steps)}."
             )
 
-    def prepare_latents(self, batch_size, num_channels_latents, dtype, device, generator, latents=None):
-        shape = (batch_size, num_channels_latents, 1, 1)
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // 8, width // 8)
         if latents is None:
             if device.type == "mps":
                 # randn does not work reproducibly on mps
@@ -323,15 +291,16 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
     @torch.no_grad()
     def __call__(
         self,
-        image: Union[PIL.Image.Image, List[PIL.Image.Image], torch.Tensor],
+        image: Union[PIL.Image.Image, List[PIL.Image.Image], torch.FloatTensor],
+        height: int = 512,
+        width: int = 512,
         num_inference_steps: int = 50,
         guidance_scale: float = 7.5,
-        negative_prompt: Optional[Union[str, List[str]]] = None,
         num_images_per_prompt: Optional[int] = 1,
         eta: float = 0.0,
         generator: Optional[torch.Generator] = None,
         latents: Optional[torch.FloatTensor] = None,
-        output_type: Optional[str] = "str",
+        output_type: Optional[str] = "pil",
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: Optional[int] = 1,
@@ -341,8 +310,15 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
         Function invoked when calling the pipeline for generation.
 
         Args:
-            image (`PIL.Image.Image`, `List[PIL.Image.Image]` or `torch.Tensor`):
-                The image prompt or prompts to guide the image generation.
+            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `torch.FloatTensor`):
+                The image or images to guide the image generation. If you provide a tensor, it needs to comply with the
+                configuration of
+                [this](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json)
+                `CLIPFeatureExtractor`
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
@@ -352,9 +328,6 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
                 Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
                 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
                 usually at the expense of lower image quality.
-            negative_prompt (`str` or `List[str]`, *optional*):
-                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
-                if `guidance_scale` is less than `1`).
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             eta (`float`, *optional*, defaults to 0.0):
@@ -389,80 +362,76 @@ class VersatileDiffusionImageToTextPipeline(DiffusionPipeline):
         """
 
         # 1. Check inputs. Raise error if not correct
-        self.check_inputs(image, callback_steps)
+        self.check_inputs(image, height, width, callback_steps)
 
         # 2. Define call parameters
-        batch_size = 1 if isinstance(image, PIL.Image.Image) else len(image)
+        if isinstance(image, PIL.Image.Image):
+            batch_size = 1
+        elif isinstance(image, list):
+            batch_size = len(image)
+        else:
+            batch_size = image.shape[0]
         device = self._execution_device
         # 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
 
-        # 3. Encode input prompt
-        image_embeddings = self._encode_prompt(
-            image, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt
-        )
+        # 3. Encode input image
+        image_embeddings = self._encode_image(image, device, num_images_per_prompt, do_classifier_free_guidance)
 
         # 4. Prepare timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)
         timesteps = self.scheduler.timesteps
 
         # 5. Prepare latent variables
-        num_channels_latents = self.text_unet.in_channels[0]
+        num_channels_latents = self.unet.in_channels
         latents = self.prepare_latents(
             batch_size * num_images_per_prompt,
             num_channels_latents,
+            height,
+            width,
             image_embeddings.dtype,
             device,
             generator,
             latents,
         )
 
-        # 6. Prepare extra step kwargs.
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
-        # 7. Swap the attention blocks between the image and text UNets
-        self.swap_unet_attention_blocks()
-
-        # 8. Denoising loop
+        # 7. Denoising loop
         for i, t in enumerate(self.progress_bar(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)
 
-            print("latent_model_input", latent_model_input.abs().sum())
-            print("timestep", t)
-
             # predict the noise residual
-            noise_pred = self.text_unet(latent_model_input, t, encoder_hidden_states=image_embeddings).sample
+            noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=image_embeddings).sample
+
             # perform 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)
 
-            print("e_t", noise_pred.abs().sum())
-            print("e_t[3,3]", noise_pred[0, :5, 0, 0])
-
             # compute the previous noisy sample x_t -> x_t-1
             latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
-            print("latents", latents.abs().sum())
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        # 9. Swap the attention blocks backs in case the UNets are reused in another pipeline
-        self.swap_unet_attention_blocks()
+        # 8. Post-processing
+        image = self.decode_latents(latents)
 
-        # 10. Post-processing
-        text = self.decode_latents(latents)
+        # 9. Run safety checker
+        image, has_nsfw_concept = self.run_safety_checker(image, device, image_embeddings.dtype)
 
-        # 11. Convert to strings
-        if output_type == "str":
-            text = self.text_vae_tokenizer.batch_decode(text)
+        # 10. Convert to PIL
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
 
         if not return_dict:
-            return (text,)
+            return (image, has_nsfw_concept)
 
-        return TextPipelineOutput(text=text)
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -127,7 +127,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -192,7 +192,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py

@@ -140,7 +140,7 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             scheduler._internal_dict = FrozenDict(new_config)
 
         if safety_checker is None:
-            logger.warn(
+            logger.warning(
                 f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                 " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                 " results in services or applications open to the public. Both the diffusers team and Hugging Face"

src/diffusers/pipelines/stable_diffusion_safe/__init__.py

@@ -0,0 +1,72 @@
+from dataclasses import dataclass
+from enum import Enum
+from typing import List, Optional, Union
+
+import numpy as np
+
+import PIL
+from PIL import Image
+
+from ...utils import BaseOutput, is_torch_available, is_transformers_available
+
+
+@dataclass
+class SafetyConfig(object):
+    WEAK = {
+        "sld_warmup_steps": 15,
+        "sld_guidance_scale": 20,
+        "sld_threshold": 0.0,
+        "sld_momentum_scale": 0.0,
+        "sld_mom_beta": 0.0,
+    }
+    MEDIUM = {
+        "sld_warmup_steps": 10,
+        "sld_guidance_scale": 1000,
+        "sld_threshold": 0.01,
+        "sld_momentum_scale": 0.3,
+        "sld_mom_beta": 0.4,
+    }
+    STRONG = {
+        "sld_warmup_steps": 7,
+        "sld_guidance_scale": 2000,
+        "sld_threshold": 0.025,
+        "sld_momentum_scale": 0.5,
+        "sld_mom_beta": 0.7,
+    }
+    MAX = {
+        "sld_warmup_steps": 0,
+        "sld_guidance_scale": 5000,
+        "sld_threshold": 1.0,
+        "sld_momentum_scale": 0.5,
+        "sld_mom_beta": 0.7,
+    }
+
+
+@dataclass
+class StableDiffusionSafePipelineOutput(BaseOutput):
+    """
+    Output class for Safe Stable Diffusion pipelines.
+
+    Args:
+        images (`List[PIL.Image.Image]` or `np.ndarray`)
+            List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
+            num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
+        nsfw_content_detected (`List[bool]`)
+            List of flags denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, or `None` if safety checking could not be performed.
+        images (`List[PIL.Image.Image]` or `np.ndarray`)
+            List of denoised PIL images that were flagged by the safety checker any may contain "not-safe-for-work"
+            (nsfw) content, or `None` if no safety check was performed or no images were flagged.
+        applied_safety_concept (`str`)
+            The safety concept that was applied for safety guidance, or `None` if safety guidance was disabled
+    """
+
+    images: Union[List[PIL.Image.Image], np.ndarray]
+    nsfw_content_detected: Optional[List[bool]]
+    unsafe_images: Optional[Union[List[PIL.Image.Image], np.ndarray]]
+    applied_safety_concept: Optional[str]
+
+
+if is_transformers_available() and is_torch_available():
+    from .pipeline_stable_diffusion_safe import StableDiffusionPipelineSafe
+    from .safety_checker import SafeStableDiffusionSafetyChecker

src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py

@@ -0,0 +1,721 @@
+import inspect
+import warnings
+from typing import Callable, List, Optional, Union
+
+import numpy as np
+import torch
+
+from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
+
+from ...configuration_utils import FrozenDict
+from ...models import AutoencoderKL, UNet2DConditionModel
+from ...pipeline_utils import DiffusionPipeline
+from ...schedulers import (
+    DDIMScheduler,
+    DPMSolverMultistepScheduler,
+    EulerAncestralDiscreteScheduler,
+    EulerDiscreteScheduler,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+)
+from ...utils import deprecate, is_accelerate_available, logging
+from . import StableDiffusionSafePipelineOutput
+from .safety_checker import SafeStableDiffusionSafetyChecker
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+class StableDiffusionPipelineSafe(DiffusionPipeline):
+    r"""
+    Pipeline for text-to-image generation using Safe Latent Diffusion.
+
+    The implementation is based on the [`StableDiffusionPipeline`]
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+    library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`CLIPTextModel`]):
+            Frozen text-encoder. Stable Diffusion uses the text portion of
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        tokenizer (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offensive or harmful.
+            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
+        feature_extractor ([`CLIPFeatureExtractor`]):
+            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
+    """
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: Union[
+            DDIMScheduler,
+            DPMSolverMultistepScheduler,
+            EulerAncestralDiscreteScheduler,
+            EulerDiscreteScheduler,
+            LMSDiscreteScheduler,
+            PNDMScheduler,
+        ],
+        safety_checker: SafeStableDiffusionSafetyChecker,
+        feature_extractor: CLIPFeatureExtractor,
+    ):
+        super().__init__()
+        safety_concept: Optional[str] = (
+            "an image showing hate, harassment, violence, suffering, humiliation, harm, suicide, sexual, nudity,"
+            " bodily fluids, blood, obscene gestures, illegal activity, drug use, theft, vandalism, weapons, child"
+            " abuse, brutality, cruelty"
+        )
+
+        if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
+            deprecation_message = (
+                f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
+                f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
+                "to update the config accordingly as leaving `steps_offset` might led to incorrect results"
+                " in future versions. If you have downloaded this checkpoint from the Hugging Face Hub,"
+                " it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`"
+                " file"
+            )
+            deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(scheduler.config)
+            new_config["steps_offset"] = 1
+            scheduler._internal_dict = FrozenDict(new_config)
+
+        if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
+            deprecation_message = (
+                f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
+                " `clip_sample` should be set to False in the configuration file. Please make sure to update the"
+                " config accordingly as not setting `clip_sample` in the config might lead to incorrect results in"
+                " future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very"
+                " nice if you could open a Pull request for the `scheduler/scheduler_config.json` file"
+            )
+            deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(scheduler.config)
+            new_config["clip_sample"] = False
+            scheduler._internal_dict = FrozenDict(new_config)
+
+        if safety_checker is None:
+            logger.warning(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+        )
+        self._safety_text_concept = safety_concept
+
+    @property
+    def safety_concept(self):
+        r"""
+        Getter method for the safety concept used with SLD
+
+        Returns:
+            `str`: The text describing the safety concept
+        """
+        return self._safety_text_concept
+
+    @safety_concept.setter
+    def safety_concept(self, concept):
+        r"""
+        Setter method for the safety concept used with SLD
+
+        Args:
+            concept (`str`):
+                The text of the new safety concept
+        """
+        self._safety_text_concept = concept
+
+    def enable_xformers_memory_efficient_attention(self):
+        r"""
+        Enable memory efficient attention as implemented in xformers.
+
+        When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference
+        time. Speed up at training time is not guaranteed.
+
+        Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention
+        is used.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(True)
+
+    def disable_xformers_memory_efficient_attention(self):
+        r"""
+        Disable memory efficient attention as implemented in xformers.
+        """
+        self.unet.set_use_memory_efficient_attention_xformers(False)
+
+    def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
+        r"""
+        Enable sliced attention computation.
+
+        When this option is enabled, the attention module will split the input tensor in slices, to compute attention
+        in several steps. This is useful to save some memory in exchange for a small speed decrease.
+
+        Args:
+            slice_size (`str` or `int`, *optional*, defaults to `"auto"`):
+                When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
+                a number is provided, uses as many slices as `attention_head_dim // slice_size`. In this case,
+                `attention_head_dim` must be a multiple of `slice_size`.
+        """
+        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):
+        r"""
+        Disable sliced attention computation. If `enable_attention_slicing` was previously invoked, this method will go
+        back to computing attention in one step.
+        """
+        # set slice_size = `None` to disable `attention slicing`
+        self.enable_attention_slicing(None)
+
+    def enable_sequential_cpu_offload(self):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        """
+        if is_accelerate_available():
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("Please install accelerate via `pip install accelerate`")
+
+        device = torch.device("cuda")
+
+        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.safety_checker]:
+            if cpu_offloaded_model is not None:
+                cpu_offload(cpu_offloaded_model, device)
+
+    @property
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
+    def _execution_device(self):
+        r"""
+        Returns the device on which the pipeline's models will be executed. After calling
+        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
+        hooks.
+        """
+        if self.device != torch.device("meta") or not hasattr(self.unet, "_hf_hook"):
+            return self.device
+        for module in self.unet.modules():
+            if (
+                hasattr(module, "_hf_hook")
+                and hasattr(module._hf_hook, "execution_device")
+                and module._hf_hook.execution_device is not None
+            ):
+                return torch.device(module._hf_hook.execution_device)
+        return self.device
+
+    def _encode_prompt(
+        self,
+        prompt,
+        device,
+        num_images_per_prompt,
+        do_classifier_free_guidance,
+        negative_prompt,
+        enable_safety_guidance,
+    ):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+
+        Args:
+            prompt (`str` or `list(int)`):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            do_classifier_free_guidance (`bool`):
+                whether to use classifier free guidance or not
+            negative_prompt (`str` or `List[str]`):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+        """
+        batch_size = len(prompt) if isinstance(prompt, list) else 1
+
+        text_inputs = self.tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=self.tokenizer.model_max_length,
+            truncation=True,
+            return_tensors="pt",
+        )
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer(prompt, padding="max_length", return_tensors="pt").input_ids
+
+        if not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+            )
+
+        if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+            attention_mask = text_inputs.attention_mask.to(device)
+        else:
+            attention_mask = None
+
+        text_embeddings = self.text_encoder(
+            text_input_ids.to(device),
+            attention_mask=attention_mask,
+        )
+        text_embeddings = text_embeddings[0]
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        bs_embed, seq_len, _ = text_embeddings.shape
+        text_embeddings = text_embeddings.repeat(1, num_images_per_prompt, 1)
+        text_embeddings = text_embeddings.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance:
+            uncond_tokens: List[str]
+            if negative_prompt is None:
+                uncond_tokens = [""] * batch_size
+            elif type(prompt) is not type(negative_prompt):
+                raise TypeError(
+                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                    f" {type(prompt)}."
+                )
+            elif isinstance(negative_prompt, str):
+                uncond_tokens = [negative_prompt]
+            elif batch_size != len(negative_prompt):
+                raise ValueError(
+                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                    " the batch size of `prompt`."
+                )
+            else:
+                uncond_tokens = negative_prompt
+
+            max_length = text_input_ids.shape[-1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = uncond_input.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            uncond_embeddings = self.text_encoder(
+                uncond_input.input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            uncond_embeddings = uncond_embeddings[0]
+
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = uncond_embeddings.shape[1]
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+            # Encode the safety concept text
+            if enable_safety_guidance:
+                safety_concept_input = self.tokenizer(
+                    [self._safety_text_concept],
+                    padding="max_length",
+                    max_length=max_length,
+                    truncation=True,
+                    return_tensors="pt",
+                )
+                safety_embeddings = self.text_encoder(safety_concept_input.input_ids.to(self.device))[0]
+
+                # duplicate safety embeddings for each generation per prompt, using mps friendly method
+                seq_len = safety_embeddings.shape[1]
+                safety_embeddings = safety_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+                safety_embeddings = safety_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+                # For classifier free guidance + sld, we need to do three forward passes.
+                # Here we concatenate the unconditional and text embeddings into a single batch
+                # to avoid doing three forward passes
+                text_embeddings = torch.cat([uncond_embeddings, text_embeddings, safety_embeddings])
+
+            else:
+                # 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])
+
+        return text_embeddings
+
+    def run_safety_checker(self, image, device, dtype, enable_safety_guidance):
+        if self.safety_checker is not None:
+            safety_checker_input = self.feature_extractor(self.numpy_to_pil(image), return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+            flagged_images = None
+            if any(has_nsfw_concept):
+                logger.warning(
+                    "Potential NSFW content was detected in one or more images. A black image will be returned"
+                    " instead."
+                    f" {'You may look at this images in the `unsafe_images` variable of the output at your own discretion.' if enable_safety_guidance else 'Try again with a different prompt and/or seed.'} "
+                )
+                flagged_images = np.zeros((2, *image.shape[1:]))
+                for idx, has_nsfw_concept in enumerate(has_nsfw_concept):
+                    if has_nsfw_concept:
+                        flagged_images[idx] = image[idx]
+                        image[idx] = np.zeros(image[idx].shape)  # black image
+        else:
+            has_nsfw_concept = None
+            flagged_images = None
+        return image, has_nsfw_concept, flagged_images
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+    def decode_latents(self, latents):
+        latents = 1 / 0.18215 * latents
+        image = self.vae.decode(latents).sample
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # 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
+        return extra_step_kwargs
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.check_inputs
+    def check_inputs(self, prompt, height, width, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        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 (callback_steps is None) or (
+            callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+        ):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (batch_size, num_channels_latents, height // 8, width // 8)
+        if latents is None:
+            if device.type == "mps":
+                # randn does not work reproducibly on mps
+                latents = torch.randn(shape, generator=generator, device="cpu", dtype=dtype).to(device)
+            else:
+                latents = torch.randn(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            if latents.shape != shape:
+                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
+    def perform_safety_guidance(
+        self,
+        enable_safety_guidance,
+        safety_momentum,
+        noise_guidance,
+        noise_pred_out,
+        i,
+        sld_guidance_scale,
+        sld_warmup_steps,
+        sld_threshold,
+        sld_momentum_scale,
+        sld_mom_beta,
+    ):
+        # Perform SLD guidance
+        if enable_safety_guidance:
+            if safety_momentum is None:
+                safety_momentum = torch.zeros_like(noise_guidance)
+            noise_pred_text, noise_pred_uncond = noise_pred_out[0], noise_pred_out[1]
+            noise_pred_safety_concept = noise_pred_out[2]
+
+            # Equation 6
+            scale = torch.clamp(torch.abs((noise_pred_text - noise_pred_safety_concept)) * sld_guidance_scale, max=1.0)
+
+            # Equation 6
+            safety_concept_scale = torch.where(
+                (noise_pred_text - noise_pred_safety_concept) >= sld_threshold, torch.zeros_like(scale), scale
+            )
+
+            # Equation 4
+            noise_guidance_safety = torch.mul((noise_pred_safety_concept - noise_pred_uncond), safety_concept_scale)
+
+            # Equation 7
+            noise_guidance_safety = noise_guidance_safety + sld_momentum_scale * safety_momentum
+
+            # Equation 8
+            safety_momentum = sld_mom_beta * safety_momentum + (1 - sld_mom_beta) * noise_guidance_safety
+
+            if i >= sld_warmup_steps:  # Warmup
+                # Equation 3
+                noise_guidance = noise_guidance - noise_guidance_safety
+        return noise_guidance, safety_momentum
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        prompt: Union[str, List[str]],
+        height: int = 512,
+        width: int = 512,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 7.5,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[torch.Generator] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: Optional[int] = 1,
+        sld_guidance_scale: Optional[float] = 1000,
+        sld_warmup_steps: Optional[int] = 10,
+        sld_threshold: Optional[float] = 0.01,
+        sld_momentum_scale: Optional[float] = 0.3,
+        sld_mom_beta: Optional[float] = 0.4,
+        **kwargs,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`):
+                The prompt or prompts to guide the image generation.
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+            sld_guidance_scale (`float`, *optional*, defaults to 1000):
+                Safe latent guidance as defined in [Safe Latent Diffusion](https://arxiv.org/abs/2211.05105).
+                `sld_guidance_scale` is defined as sS of Eq. 6. If set to be less than 1, safety guidance will be
+                disabled.
+            sld_warmup_steps (`int`, *optional*, defaults to 10):
+                Number of warmup steps for safety guidance. SLD will only be applied for diffusion steps greater than
+                `sld_warmup_steps`. `sld_warmup_steps` is defined as `delta` of [Safe Latent
+                Diffusion](https://arxiv.org/abs/2211.05105).
+            sld_threshold (`float`, *optional*, defaults to 0.01):
+                Threshold that separates the hyperplane between appropriate and inappropriate images. `sld_threshold`
+                is defined as `lamda` of Eq. 5 in [Safe Latent Diffusion](https://arxiv.org/abs/2211.05105).
+            sld_momentum_scale (`float`, *optional*, defaults to 0.3):
+                Scale of the SLD momentum to be added to the safety guidance at each diffusion step. If set to 0.0
+                momentum will be disabled. Momentum is already built up during warmup, i.e. for diffusion steps smaller
+                than `sld_warmup_steps`. `sld_momentum_scale` is defined as `sm` of Eq. 7 in [Safe Latent
+                Diffusion](https://arxiv.org/abs/2211.05105).
+            sld_mom_beta (`float`, *optional*, defaults to 0.4):
+                Defines how safety guidance momentum builds up. `sld_mom_beta` indicates how much of the previous
+                momentum will be kept. Momentum is already built up during warmup, i.e. for diffusion steps smaller
+                than `sld_warmup_steps`. `sld_mom_beta` is defined as `beta m` of Eq. 8 in [Safe Latent
+                Diffusion](https://arxiv.org/abs/2211.05105).
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(prompt, height, width, callback_steps)
+
+        # 2. Define call parameters
+        batch_size = 1 if isinstance(prompt, str) else len(prompt)
+        device = self._execution_device
+
+        # 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
+
+        enable_safety_guidance = sld_guidance_scale > 1.0 and do_classifier_free_guidance
+        if not enable_safety_guidance:
+            warnings.warn("Safety checker disabled!")
+
+        # 3. Encode input prompt
+        text_embeddings = self._encode_prompt(
+            prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt, enable_safety_guidance
+        )
+
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            text_embeddings.dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 6. Prepare extra step kwargs.
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        safety_momentum = None
+
+        for i, t in enumerate(self.progress_bar(timesteps)):
+            # expand the latents if we are doing classifier free guidance
+            latent_model_input = (
+                torch.cat([latents] * (3 if enable_safety_guidance else 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 guidance
+            if do_classifier_free_guidance:
+                noise_pred_out = noise_pred.chunk((3 if enable_safety_guidance else 2))
+                noise_pred_uncond, noise_pred_text = noise_pred_out[0], noise_pred_out[1]
+
+                # default classifier free guidance
+                noise_guidance = noise_pred_text - noise_pred_uncond
+
+                # Perform SLD guidance
+                if enable_safety_guidance:
+                    if safety_momentum is None:
+                        safety_momentum = torch.zeros_like(noise_guidance)
+                    noise_pred_safety_concept = noise_pred_out[2]
+
+                    # Equation 6
+                    scale = torch.clamp(
+                        torch.abs((noise_pred_text - noise_pred_safety_concept)) * sld_guidance_scale, max=1.0
+                    )
+
+                    # Equation 6
+                    safety_concept_scale = torch.where(
+                        (noise_pred_text - noise_pred_safety_concept) >= sld_threshold, torch.zeros_like(scale), scale
+                    )
+
+                    # Equation 4
+                    noise_guidance_safety = torch.mul(
+                        (noise_pred_safety_concept - noise_pred_uncond), safety_concept_scale
+                    )
+
+                    # Equation 7
+                    noise_guidance_safety = noise_guidance_safety + sld_momentum_scale * safety_momentum
+
+                    # Equation 8
+                    safety_momentum = sld_mom_beta * safety_momentum + (1 - sld_mom_beta) * noise_guidance_safety
+
+                    if i >= sld_warmup_steps:  # Warmup
+                        # Equation 3
+                        noise_guidance = noise_guidance - noise_guidance_safety
+
+                noise_pred = noise_pred_uncond + guidance_scale * noise_guidance
+
+                # compute the previous noisy sample x_t -> x_t-1
+            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+            # call the callback, if provided
+            if callback is not None and i % callback_steps == 0:
+                callback(i, t, latents)
+
+        # 8. Post-processing
+        image = self.decode_latents(latents)
+
+        # 9. Run safety checker
+        image, has_nsfw_concept, flagged_images = self.run_safety_checker(
+            image, device, text_embeddings.dtype, enable_safety_guidance
+        )
+
+        # 10. Convert to PIL
+        if output_type == "pil":
+            image = self.numpy_to_pil(image)
+            if flagged_images is not None:
+                flagged_images = self.numpy_to_pil(flagged_images)
+
+        if not return_dict:
+            return (
+                image,
+                has_nsfw_concept,
+                self._safety_text_concept if enable_safety_guidance else None,
+                flagged_images,
+            )
+
+        return StableDiffusionSafePipelineOutput(
+            images=image,
+            nsfw_content_detected=has_nsfw_concept,
+            applied_safety_concept=self._safety_text_concept if enable_safety_guidance else None,
+            unsafe_images=flagged_images,
+        )

src/diffusers/pipelines/stable_diffusion_safe/safety_checker.py

@@ -0,0 +1,110 @@
+# Copyright 2022 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.
+
+import torch
+import torch.nn as nn
+
+from transformers import CLIPConfig, CLIPVisionModel, PreTrainedModel
+
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+def cosine_distance(image_embeds, text_embeds):
+    normalized_image_embeds = nn.functional.normalize(image_embeds)
+    normalized_text_embeds = nn.functional.normalize(text_embeds)
+    return torch.mm(normalized_image_embeds, normalized_text_embeds.t())
+
+
+class SafeStableDiffusionSafetyChecker(PreTrainedModel):
+    config_class = CLIPConfig
+
+    _no_split_modules = ["CLIPEncoderLayer"]
+
+    def __init__(self, config: CLIPConfig):
+        super().__init__(config)
+
+        self.vision_model = CLIPVisionModel(config.vision_config)
+        self.visual_projection = nn.Linear(config.vision_config.hidden_size, config.projection_dim, bias=False)
+
+        self.concept_embeds = nn.Parameter(torch.ones(17, config.projection_dim), requires_grad=False)
+        self.special_care_embeds = nn.Parameter(torch.ones(3, config.projection_dim), requires_grad=False)
+
+        self.concept_embeds_weights = nn.Parameter(torch.ones(17), requires_grad=False)
+        self.special_care_embeds_weights = nn.Parameter(torch.ones(3), requires_grad=False)
+
+    @torch.no_grad()
+    def forward(self, clip_input, images):
+        pooled_output = self.vision_model(clip_input)[1]  # pooled_output
+        image_embeds = self.visual_projection(pooled_output)
+
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
+        special_cos_dist = cosine_distance(image_embeds, self.special_care_embeds).cpu().float().numpy()
+        cos_dist = cosine_distance(image_embeds, self.concept_embeds).cpu().float().numpy()
+
+        result = []
+        batch_size = image_embeds.shape[0]
+        for i in range(batch_size):
+            result_img = {"special_scores": {}, "special_care": [], "concept_scores": {}, "bad_concepts": []}
+
+            # increase this value to create a stronger `nfsw` filter
+            # at the cost of increasing the possibility of filtering benign images
+            adjustment = 0.0
+
+            for concept_idx in range(len(special_cos_dist[0])):
+                concept_cos = special_cos_dist[i][concept_idx]
+                concept_threshold = self.special_care_embeds_weights[concept_idx].item()
+                result_img["special_scores"][concept_idx] = round(concept_cos - concept_threshold + adjustment, 3)
+                if result_img["special_scores"][concept_idx] > 0:
+                    result_img["special_care"].append({concept_idx, result_img["special_scores"][concept_idx]})
+                    adjustment = 0.01
+
+            for concept_idx in range(len(cos_dist[0])):
+                concept_cos = cos_dist[i][concept_idx]
+                concept_threshold = self.concept_embeds_weights[concept_idx].item()
+                result_img["concept_scores"][concept_idx] = round(concept_cos - concept_threshold + adjustment, 3)
+                if result_img["concept_scores"][concept_idx] > 0:
+                    result_img["bad_concepts"].append(concept_idx)
+
+            result.append(result_img)
+
+        has_nsfw_concepts = [len(res["bad_concepts"]) > 0 for res in result]
+
+        return images, has_nsfw_concepts
+
+    @torch.no_grad()
+    def forward_onnx(self, clip_input: torch.FloatTensor, images: torch.FloatTensor):
+        pooled_output = self.vision_model(clip_input)[1]  # pooled_output
+        image_embeds = self.visual_projection(pooled_output)
+
+        special_cos_dist = cosine_distance(image_embeds, self.special_care_embeds)
+        cos_dist = cosine_distance(image_embeds, self.concept_embeds)
+
+        # increase this value to create a stronger `nsfw` filter
+        # at the cost of increasing the possibility of filtering benign images
+        adjustment = 0.0
+
+        special_scores = special_cos_dist - self.special_care_embeds_weights + adjustment
+        # special_scores = special_scores.round(decimals=3)
+        special_care = torch.any(special_scores > 0, dim=1)
+        special_adjustment = special_care * 0.01
+        special_adjustment = special_adjustment.unsqueeze(1).expand(-1, cos_dist.shape[1])
+
+        concept_scores = (cos_dist - self.concept_embeds_weights) + special_adjustment
+        # concept_scores = concept_scores.round(decimals=3)
+        has_nsfw_concepts = torch.any(concept_scores > 0, dim=1)
+
+        return images, has_nsfw_concepts

src/diffusers/pipelines/versatile_diffusion/__init__.py

@@ -1,11 +1,16 @@
-from ...utils import is_torch_available, is_transformers_available
+from ...utils import is_torch_available, is_transformers_available, is_transformers_version
 
 
-if is_transformers_available() and is_torch_available():
-    from .modeling_gpt2_optimus import GPT2OptimusForLatentConnector
+if is_transformers_available() and is_torch_available() and is_transformers_version(">=", "4.25.0.dev0"):
     from .modeling_text_unet import UNetFlatConditionModel
     from .pipeline_versatile_diffusion import VersatileDiffusionPipeline
     from .pipeline_versatile_diffusion_dual_guided import VersatileDiffusionDualGuidedPipeline
-    from .pipeline_versatile_diffusion_image_to_text import VersatileDiffusionImageToTextPipeline
     from .pipeline_versatile_diffusion_image_variation import VersatileDiffusionImageVariationPipeline
     from .pipeline_versatile_diffusion_text_to_image import VersatileDiffusionTextToImagePipeline
+else:
+    from ...utils.dummy_torch_and_transformers_objects import (
+        VersatileDiffusionDualGuidedPipeline,
+        VersatileDiffusionImageVariationPipeline,
+        VersatileDiffusionPipeline,
+        VersatileDiffusionTextToImagePipeline,
+    )

src/diffusers/pipelines/versatile_diffusion/modeling_gpt2_optimus.py

@@ -1,345 +0,0 @@
-import math
-
-import torch
-from torch import nn
-
-from transformers.modeling_outputs import CausalLMOutputWithCrossAttentions
-from transformers.models.gpt2.modeling_gpt2 import GPT2MLP, GPT2PreTrainedModel
-from transformers.pytorch_utils import Conv1D
-
-
-class GPT2OptimusAttention(nn.Module):
-    def __init__(self, nx, n_ctx, config, scale=False):
-        super().__init__()
-        self.output_attentions = config.output_attentions
-
-        n_state = nx  # in Attention: n_state=768 (nx=n_embd)
-        # [switch nx => n_state from Block to Attention to keep identical to TF implem]
-        assert n_state % config.n_head == 0
-        self.register_buffer("bias", torch.tril(torch.ones(n_ctx, n_ctx)).view(1, 1, n_ctx, n_ctx))
-        self.n_head = config.n_head
-        self.split_size = n_state
-        self.scale = scale
-
-        self.c_attn = Conv1D(n_state * 3, nx)
-        self.c_proj = Conv1D(n_state, nx)
-        self.attn_dropout = nn.Dropout(config.attn_pdrop)
-        self.resid_dropout = nn.Dropout(config.resid_pdrop)
-        self.pruned_heads = set()
-
-    def _attn(self, q, k, v, attention_mask=None, head_mask=None):
-        w = torch.matmul(q, k)
-        if self.scale:
-            w = w / math.sqrt(v.size(-1))
-        nd, ns = w.size(-2), w.size(-1)
-        b = self.bias[:, :, ns - nd : ns, :ns]
-        w = w * b - 1e4 * (1 - b)
-
-        if attention_mask is not None:
-            # Apply the attention mask
-            w = w + attention_mask
-
-        w = nn.Softmax(dim=-1)(w)
-        w = self.attn_dropout(w)
-
-        # Mask heads if we want to
-        if head_mask is not None:
-            w = w * head_mask
-
-        outputs = [torch.matmul(w, v)]
-        if self.output_attentions:
-            outputs.append(w)
-        return outputs
-
-    def merge_heads(self, x):
-        x = x.permute(0, 2, 1, 3).contiguous()
-        new_x_shape = x.size()[:-2] + (x.size(-2) * x.size(-1),)
-        return x.view(*new_x_shape)  # in Tensorflow implem: fct merge_states
-
-    def split_heads(self, x, k=False):
-        new_x_shape = x.size()[:-1] + (self.n_head, x.size(-1) // self.n_head)
-        x = x.view(*new_x_shape)  # in Tensorflow implem: fct split_states
-        if k:
-            return x.permute(0, 2, 3, 1)  # (batch, head, head_features, seq_length)
-        else:
-            return x.permute(0, 2, 1, 3)  # (batch, head, seq_length, head_features)
-
-    def forward(self, x, layer_past=None, attention_mask=None, head_mask=None):
-        x = self.c_attn(x)
-        query, key, value = x.split(self.split_size, dim=2)
-        query = self.split_heads(query)
-        key = self.split_heads(key, k=True)
-        value = self.split_heads(value)
-
-        if layer_past is not None:
-            past_key, past_value = layer_past[0], layer_past[1]  # transpose back cf below
-
-            past_key = self.split_heads(past_key, k=True)
-            past_value = self.split_heads(past_value)
-            # pdb.set_trace()
-            key = torch.cat((past_key, key), dim=-1)
-            value = torch.cat((past_value, value), dim=-2)
-        present = torch.stack((key.transpose(-2, -1), value))  # transpose to have same shapes for stacking
-
-        attn_outputs = self._attn(query, key, value, attention_mask, head_mask)
-        a = attn_outputs[0]
-
-        a = self.merge_heads(a)
-        a = self.c_proj(a)
-        a = self.resid_dropout(a)
-
-        outputs = [a, present] + attn_outputs[1:]
-        return outputs  # a, present, (attentions)
-
-
-class GPT2OptimusBlock(nn.Module):
-    def __init__(self, config):
-        super().__init__()
-        nx = config.n_embd
-        self.ln_1 = nn.LayerNorm(nx, eps=config.layer_norm_epsilon)
-        self.attn = GPT2OptimusAttention(nx, config.n_ctx, config, scale=True)
-        self.ln_2 = nn.LayerNorm(nx, eps=config.layer_norm_epsilon)
-        self.mlp = GPT2MLP(4 * nx, config)
-
-    def forward(self, x, layer_past=None, attention_mask=None, head_mask=None):
-        output_attn = self.attn(
-            self.ln_1(x), layer_past=layer_past, attention_mask=attention_mask, head_mask=head_mask
-        )
-        a = output_attn[0]  # output_attn: a, present, (attentions)
-
-        x = x + a
-        m = self.mlp(self.ln_2(x))
-        x = x + m
-
-        outputs = [x] + output_attn[1:]
-        return outputs  # x, present, (attentions)
-
-
-class GPT2OptimusModel(GPT2PreTrainedModel):
-    def __init__(self, config, latent_as_gpt_emb, latent_as_gpt_memory, latent_size):
-        super().__init__(config)
-        self.latent_as_gpt_emb = latent_as_gpt_emb
-        self.latent_as_gpt_memory = latent_as_gpt_memory
-        self.latent_size = latent_size
-        self.output_hidden_states = config.output_hidden_states
-        self.output_attentions = config.output_attentions
-
-        self.wte = nn.Embedding(config.vocab_size, config.n_embd)
-        self.wpe = nn.Embedding(config.n_positions, config.n_embd)
-        self.drop = nn.Dropout(config.embd_pdrop)
-        self.h = nn.ModuleList([GPT2OptimusBlock(config) for i in range(config.n_layer)])
-        self.ln_f = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
-
-        self.linear = nn.Linear(
-            self.latent_size, config.hidden_size * config.n_layer, bias=False
-        )  # different latent vector for each layer
-        self.linear_emb = nn.Linear(
-            self.latent_size, config.hidden_size, bias=False
-        )  # share the same latent vector as the embeddings
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    def forward(
-        self,
-        input_ids,
-        past=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-    ):
-        if past is None:
-            past_length = 0
-            past = [None] * len(self.h)
-        else:
-            if self.latent_as_gpt_emb:
-                past_emb = self.linear_emb(past)  # used as embeddings to add on other three embeddings
-
-            if self.latent_as_gpt_memory:
-                past = self.linear(past)
-
-                # different latent vectors for each layer
-                past_split = torch.split(past.unsqueeze(1), self.config.hidden_size, dim=2)
-                past = list(zip(past_split, past_split))
-                past_length = 1  # past[0][0].size(-2)
-            else:
-                past_length = 0
-                past = [None] * len(self.h)
-
-        if position_ids is None:
-            position_ids = torch.arange(
-                past_length, input_ids.size(-1) + past_length, dtype=torch.long, device=input_ids.device
-            )
-            position_ids = position_ids.unsqueeze(0).expand_as(input_ids)
-
-        # Attention mask.
-        if attention_mask is not None:
-            # We create a 3D attention mask from a 2D tensor mask.
-            # Sizes are [batch_size, 1, 1, to_seq_length]
-            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
-            # this attention mask is more simple than the triangular masking of causal attention
-            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
-            attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
-
-            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
-            # masked positions, this operation will create a tensor which is 0.0 for
-            # positions we want to attend and -10000.0 for masked positions.
-            # Since we are adding it to the raw scores before the softmax, this is
-            # effectively the same as removing these entirely.
-            attention_mask = attention_mask.to(dtype=next(self.parameters()).dtype)  # fp16 compatibility
-            attention_mask = (1.0 - attention_mask) * -10000.0
-
-        # Prepare head mask if needed
-        # 1.0 in head_mask indicate we keep the head
-        # attention_probs has shape bsz x n_heads x N x N
-        # head_mask has shape n_layer x batch x n_heads x N x N
-        if head_mask is not None:
-            if head_mask.dim() == 1:
-                head_mask = head_mask.unsqueeze(0).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)
-                head_mask = head_mask.expand(self.config.n_layer, -1, -1, -1, -1)
-            elif head_mask.dim() == 2:
-                head_mask = (
-                    head_mask.unsqueeze(1).unsqueeze(-1).unsqueeze(-1)
-                )  # We can specify head_mask for each layer
-            head_mask = head_mask.to(
-                dtype=next(self.parameters()).dtype
-            )  # switch to fload if need + fp16 compatibility
-        else:
-            head_mask = [None] * self.config.n_layer
-
-        input_shape = input_ids.size()
-        input_ids = input_ids.view(-1, input_ids.size(-1))
-        position_ids = position_ids.view(-1, position_ids.size(-1))
-
-        inputs_embeds = self.wte(input_ids)
-        position_embeds = self.wpe(position_ids)
-        if token_type_ids is not None:
-            token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1))
-            token_type_embeds = self.wte(token_type_ids)
-        else:
-            token_type_embeds = 0
-
-        hidden_states = inputs_embeds + position_embeds + token_type_embeds
-        if self.latent_as_gpt_emb:
-            hidden_states = hidden_states + past_emb.unsqueeze(1)
-
-        hidden_states = self.drop(hidden_states)
-
-        output_shape = input_shape + (hidden_states.size(-1),)
-
-        presents = ()
-        all_attentions = []
-        all_hidden_states = ()
-        for i, (block, layer_past) in enumerate(zip(self.h, past)):
-            if self.output_hidden_states:
-                all_hidden_states = all_hidden_states + (hidden_states.view(*output_shape),)
-
-            outputs = block(
-                hidden_states, layer_past=layer_past, attention_mask=attention_mask, head_mask=head_mask[i]
-            )
-
-            hidden_states, present = outputs[:2]
-            presents = presents + (present,)
-
-            if self.output_attentions:
-                all_attentions.append(outputs[2])
-
-        hidden_states = self.ln_f(hidden_states)
-
-        hidden_states = hidden_states.view(*output_shape)
-        # Add last hidden state
-        if self.output_hidden_states:
-            all_hidden_states = all_hidden_states + (hidden_states,)
-
-        outputs = (hidden_states, presents)
-        if self.output_hidden_states:
-            outputs = outputs + (all_hidden_states,)
-        if self.output_attentions:
-            # let the number of heads free (-1) so we can extract attention even after head pruning
-            attention_output_shape = input_shape[:-1] + (-1,) + all_attentions[0].shape[-2:]
-            all_attentions = tuple(t.view(*attention_output_shape) for t in all_attentions)
-            outputs = outputs + (all_attentions,)
-
-        return outputs  # last hidden state, presents, (all hidden_states), (attentions)
-
-
-class GPT2OptimusForLatentConnector(GPT2PreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.latent_as_gpt_emb = True
-        self.latent_as_gpt_memory = True
-        self.latent_size = getattr(config, "latent_size", 32)
-        self.transformer = GPT2OptimusModel(
-            config,
-            latent_as_gpt_emb=self.latent_as_gpt_emb,
-            latent_as_gpt_memory=self.latent_as_gpt_memory,
-            latent_size=self.latent_size,
-        )
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
-        self.init_weights()
-        self.tie_weights()
-
-        # Initialize weights and apply final processing
-        self.post_init()
-        self.tie_weights()
-
-    def _tie_or_clone_weights(self, first_module, second_module):
-        """Tie or clone module weights depending of weither we are using TorchScript or not"""
-        if self.config.torchscript:
-            first_module.weight = nn.Parameter(second_module.weight.clone())
-        else:
-            first_module.weight = second_module.weight
-
-        if hasattr(first_module, "bias") and first_module.bias is not None:
-            first_module.bias.data = torch.nn.functional.pad(
-                first_module.bias.data,
-                (0, first_module.weight.shape[0] - first_module.bias.shape[0]),
-                "constant",
-                0,
-            )
-
-    def tie_weights(self):
-        """Make sure we are sharing the input and output embeddings.
-        Export to TorchScript can't handle parameter sharing so we are cloning them instead.
-        """
-        self._tie_or_clone_weights(self.lm_head, self.transformer.wte)
-
-    def forward(
-        self,
-        input_ids,
-        past_key_values=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        output_attentions=None,
-        output_hidden_states=None,
-        return_dict=True,
-    ):
-        transformer_outputs = self.transformer(
-            input_ids,
-            past=past_key_values,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-        )
-        hidden_states = transformer_outputs[0]
-
-        lm_logits = self.lm_head(hidden_states)
-
-        return CausalLMOutputWithCrossAttentions(
-            loss=None,
-            logits=lm_logits,
-            past_key_values=past_key_values,
-            hidden_states=None,
-            attentions=None,
-            cross_attentions=None,
-        )
-
-    def prepare_inputs_for_generation(self, input_ids, past, **kwargs):
-        return {
-            "input_ids": input_ids,
-            "past_key_values": past,
-        }

src/diffusers/pipelines/versatile_diffusion/modeling_text_unet.py

@@ -6,7 +6,7 @@ import torch.nn as nn
 
 from ...configuration_utils import ConfigMixin, register_to_config
 from ...modeling_utils import ModelMixin
-from ...models.attention import Transformer2DModel
+from ...models.attention import DualTransformer2DModel, Transformer2DModel
 from ...models.embeddings import TimestepEmbedding, Timesteps
 from ...models.unet_2d_condition import UNet2DConditionOutput
 from ...utils import logging
@@ -28,6 +28,7 @@ def get_down_block(
     resnet_groups=None,
     cross_attention_dim=None,
     downsample_padding=None,
+    dual_cross_attention=None,
 ):
     down_block_type = down_block_type[7:] if down_block_type.startswith("UNetRes") else down_block_type
     if down_block_type == "DownBlockFlat":
@@ -74,6 +75,7 @@ def get_up_block(
     attn_num_head_channels,
     resnet_groups=None,
     cross_attention_dim=None,
+    dual_cross_attention=None,
 ):
     up_block_type = up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type
     if up_block_type == "UpBlockFlat":
@@ -117,7 +119,8 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
     implements for all the models (such as downloading or saving, etc.)
 
     Parameters:
-        sample_size (`int`, *optional*): The size of the input sample.
+        sample_size (`int` or `Tuple[int, int]`, *optional*, defaults to `None`):
+            Height and width of input/output sample.
         in_channels (`int`, *optional*, defaults to 4): The number of channels in the input sample.
         out_channels (`int`, *optional*, defaults to 4): The number of channels in the output.
         center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.
@@ -172,6 +175,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
         norm_eps: float = 1e-5,
         cross_attention_dim: int = 1280,
         attention_head_dim: int = 8,
+        dual_cross_attention: bool = False,
     ):
         super().__init__()
 
@@ -211,6 +215,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
                 cross_attention_dim=cross_attention_dim,
                 attn_num_head_channels=attention_head_dim,
                 downsample_padding=downsample_padding,
+                dual_cross_attention=dual_cross_attention,
             )
             self.down_blocks.append(down_block)
 
@@ -225,6 +230,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
             cross_attention_dim=cross_attention_dim,
             attn_num_head_channels=attention_head_dim,
             resnet_groups=norm_num_groups,
+            dual_cross_attention=dual_cross_attention,
         )
 
         # count how many layers upsample the images
@@ -260,6 +266,7 @@ class UNetFlatConditionModel(ModelMixin, ConfigMixin):
                 resnet_groups=norm_num_groups,
                 cross_attention_dim=cross_attention_dim,
                 attn_num_head_channels=attention_head_dim,
+                dual_cross_attention=dual_cross_attention,
             )
             self.up_blocks.append(up_block)
             prev_output_channel = output_channel
@@ -632,6 +639,7 @@ class CrossAttnDownBlockFlat(nn.Module):
         output_scale_factor=1.0,
         downsample_padding=1,
         add_downsample=True,
+        dual_cross_attention=False,
     ):
         super().__init__()
         resnets = []
@@ -656,16 +664,28 @@ class CrossAttnDownBlockFlat(nn.Module):
                     pre_norm=resnet_pre_norm,
                 )
             )
-            attentions.append(
-                Transformer2DModel(
-                    attn_num_head_channels,
-                    out_channels // attn_num_head_channels,
-                    in_channels=out_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
+            if not dual_cross_attention:
+                attentions.append(
+                    Transformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
+                )
+            else:
+                attentions.append(
+                    DualTransformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
                 )
-            )
         self.attentions = nn.ModuleList(attentions)
         self.resnets = nn.ModuleList(resnets)
 
@@ -830,6 +850,7 @@ class CrossAttnUpBlockFlat(nn.Module):
         attention_type="default",
         output_scale_factor=1.0,
         add_upsample=True,
+        dual_cross_attention=False,
     ):
         super().__init__()
         resnets = []
@@ -856,16 +877,28 @@ class CrossAttnUpBlockFlat(nn.Module):
                     pre_norm=resnet_pre_norm,
                 )
             )
-            attentions.append(
-                Transformer2DModel(
-                    attn_num_head_channels,
-                    out_channels // attn_num_head_channels,
-                    in_channels=out_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
+            if not dual_cross_attention:
+                attentions.append(
+                    Transformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
+                )
+            else:
+                attentions.append(
+                    DualTransformer2DModel(
+                        attn_num_head_channels,
+                        out_channels // attn_num_head_channels,
+                        in_channels=out_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
                 )
-            )
         self.attentions = nn.ModuleList(attentions)
         self.resnets = nn.ModuleList(resnets)
 
@@ -954,6 +987,7 @@ class UNetMidBlockFlatCrossAttn(nn.Module):
         attention_type="default",
         output_scale_factor=1.0,
         cross_attention_dim=1280,
+        dual_cross_attention=False,
         **kwargs,
     ):
         super().__init__()
@@ -980,16 +1014,28 @@ class UNetMidBlockFlatCrossAttn(nn.Module):
         attentions = []
 
         for _ in range(num_layers):
-            attentions.append(
-                Transformer2DModel(
-                    attn_num_head_channels,
-                    in_channels // attn_num_head_channels,
-                    in_channels=in_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
+            if not dual_cross_attention:
+                attentions.append(
+                    Transformer2DModel(
+                        attn_num_head_channels,
+                        in_channels // attn_num_head_channels,
+                        in_channels=in_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
+                )
+            else:
+                attentions.append(
+                    DualTransformer2DModel(
+                        attn_num_head_channels,
+                        in_channels // attn_num_head_channels,
+                        in_channels=in_channels,
+                        num_layers=1,
+                        cross_attention_dim=cross_attention_dim,
+                        norm_num_groups=resnet_groups,
+                    )
                 )
-            )
             resnets.append(
                 ResnetBlockFlat(
                     in_channels=in_channels,

src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion.py

@@ -1,4 +1,5 @@
-from typing import Any, Callable, Dict, List, Optional, Union
+import inspect
+from typing import Callable, List, Optional, Union
 
 import torch
 
@@ -9,6 +10,7 @@ from ...models import AutoencoderKL, UNet2DConditionModel
 from ...pipeline_utils import DiffusionPipeline
 from ...schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
 from ...utils import logging
+from .pipeline_versatile_diffusion_dual_guided import VersatileDiffusionDualGuidedPipeline
 from .pipeline_versatile_diffusion_image_variation import VersatileDiffusionImageVariationPipeline
 from .pipeline_versatile_diffusion_text_to_image import VersatileDiffusionTextToImagePipeline
 
@@ -77,10 +79,6 @@ class VersatileDiffusionPipeline(DiffusionPipeline):
             scheduler=scheduler,
         )
 
-    @property
-    def components(self) -> Dict[str, Any]:
-        return {k: getattr(self, k) for k in self.config.keys() if not k.startswith("_")}
-
     def enable_attention_slicing(self, slice_size: Optional[Union[str, int]] = "auto"):
         r"""
         Enable sliced attention computation.
@@ -127,7 +125,88 @@ class VersatileDiffusionPipeline(DiffusionPipeline):
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: Optional[int] = 1,
     ):
-        return VersatileDiffusionImageVariationPipeline(**self.components)(
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            image (`PIL.Image.Image`, `List[PIL.Image.Image]` or `torch.Tensor`):
+                The image prompt or prompts to guide the image generation.
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionPipeline
+        >>> import torch
+        >>> import requests
+        >>> from io import BytesIO
+        >>> from PIL import Image
+
+        >>> # let's download an initial image
+        >>> url = "https://huggingface.co/datasets/diffusers/images/resolve/main/benz.jpg"
+
+        >>> response = requests.get(url)
+        >>> image = Image.open(BytesIO(response.content)).convert("RGB")
+
+        >>> pipe = VersatileDiffusionPipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> image = pipe(image, generator=generator).images[0]
+        >>> image.save("./car_variation.png")
+        ```
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+        expected_components = inspect.signature(VersatileDiffusionImageVariationPipeline.__init__).parameters.keys()
+        components = {name: component for name, component in self.components.items() if name in expected_components}
+        return VersatileDiffusionImageVariationPipeline(**components)(
             image=image,
             height=height,
             width=width,
@@ -162,7 +241,80 @@ class VersatileDiffusionPipeline(DiffusionPipeline):
         callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
         callback_steps: Optional[int] = 1,
     ):
-        return VersatileDiffusionTextToImagePipeline(**self.components)(
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`):
+                The prompt or prompts to guide the image generation.
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionPipeline
+        >>> import torch
+
+        >>> pipe = VersatileDiffusionPipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> image = pipe.text_to_image("an astronaut riding on a horse on mars", generator=generator).images[0]
+        >>> image.save("./astronaut.png")
+        ```
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
+            When returning a tuple, the first element is a list with the generated images, and the second element is a
+            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
+            (nsfw) content, according to the `safety_checker`.
+        """
+        expected_components = inspect.signature(VersatileDiffusionTextToImagePipeline.__init__).parameters.keys()
+        components = {name: component for name, component in self.components.items() if name in expected_components}
+        temp_pipeline = VersatileDiffusionTextToImagePipeline(**components)
+        output = temp_pipeline(
             prompt=prompt,
             height=height,
             width=width,
@@ -178,3 +330,133 @@ class VersatileDiffusionPipeline(DiffusionPipeline):
             callback=callback,
             callback_steps=callback_steps,
         )
+        # swap the attention blocks back to the original state
+        temp_pipeline._swap_unet_attention_blocks()
+
+        return output
+
+    @torch.no_grad()
+    def dual_guided(
+        self,
+        prompt: Union[PIL.Image.Image, List[PIL.Image.Image]],
+        image: Union[str, List[str]],
+        text_to_image_strength: float = 0.5,
+        height: int = 512,
+        width: int = 512,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 7.5,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[torch.Generator] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: Optional[int] = 1,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`):
+                The prompt or prompts to guide the image generation.
+            height (`int`, *optional*, defaults to 512):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to 512):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+                if `guidance_scale` is less than `1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator`, *optional*):
+                A [torch generator](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation
+                deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionPipeline
+        >>> import torch
+        >>> import requests
+        >>> from io import BytesIO
+        >>> from PIL import Image
+
+        >>> # let's download an initial image
+        >>> url = "https://huggingface.co/datasets/diffusers/images/resolve/main/benz.jpg"
+
+        >>> response = requests.get(url)
+        >>> image = Image.open(BytesIO(response.content)).convert("RGB")
+        >>> text = "a red car in the sun"
+
+        >>> pipe = VersatileDiffusionPipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> text_to_image_strength = 0.75
+
+        >>> image = pipe.dual_guided(
+        ...     prompt=text, image=image, text_to_image_strength=text_to_image_strength, generator=generator
+        ... ).images[0]
+        >>> image.save("./car_variation.png")
+        ```
+
+        Returns:
+            [`~pipelines.stable_diffusion.ImagePipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.ImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple. When
+            returning a tuple, the first element is a list with the generated images.
+        """
+
+        expected_components = inspect.signature(VersatileDiffusionDualGuidedPipeline.__init__).parameters.keys()
+        components = {name: component for name, component in self.components.items() if name in expected_components}
+        temp_pipeline = VersatileDiffusionDualGuidedPipeline(**components)
+        output = temp_pipeline(
+            prompt=prompt,
+            image=image,
+            text_to_image_strength=text_to_image_strength,
+            height=height,
+            width=width,
+            num_inference_steps=num_inference_steps,
+            guidance_scale=guidance_scale,
+            num_images_per_prompt=num_images_per_prompt,
+            eta=eta,
+            generator=generator,
+            latents=latents,
+            output_type=output_type,
+            return_dict=return_dict,
+            callback=callback,
+            callback_steps=callback_steps,
+        )
+        temp_pipeline._revert_dual_attention()
+
+        return output

src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import inspect
-from typing import Callable, List, Optional, Union
+from typing import Callable, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
@@ -88,7 +88,20 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
             scheduler=scheduler,
         )
 
-    def convert_to_dual_attention(self, mix_ratio=0.5, condition_types=("image", "text")):
+        if self.text_unet is not None and (
+            "dual_cross_attention" not in self.image_unet.config or not self.image_unet.config.dual_cross_attention
+        ):
+            # if loading from a universal checkpoint rather than a saved dual-guided pipeline
+            self._convert_to_dual_attention()
+
+    def remove_unused_weights(self):
+        self.register_modules(text_unet=None)
+
+    def _convert_to_dual_attention(self):
+        """
+        Replace image_unet's `Transformer2DModel` blocks with `DualTransformer2DModel` that contains transformer blocks
+        from both `image_unet` and `text_unet`
+        """
         for name, module in self.image_unet.named_modules():
             if isinstance(module, Transformer2DModel):
                 parent_name, index = name.rsplit(".", 1)
@@ -112,16 +125,17 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
                     activation_fn=config.activation_fn,
                     num_embeds_ada_norm=config.num_embeds_ada_norm,
                 )
-                for i, type in enumerate(condition_types):
-                    if type == "image":
-                        dual_transformer.transformers[i] = image_transformer
-                    else:
-                        dual_transformer.transformers[i] = text_transformer
+                dual_transformer.transformers[0] = image_transformer
+                dual_transformer.transformers[1] = text_transformer
 
-                dual_transformer.mix_ratio = mix_ratio
                 self.image_unet.get_submodule(parent_name)[index] = dual_transformer
+                self.image_unet.register_to_config(dual_cross_attention=True)
 
-    def remove_dual_attention(self):
+    def _revert_dual_attention(self):
+        """
+        Revert the image_unet `DualTransformer2DModel` blocks back to `Transformer2DModel` with image_unet weights Call
+        this function if you reuse `image_unet` in another pipeline, e.g. `VersatileDiffusionPipeline`
+        """
         for name, module in self.image_unet.named_modules():
             if isinstance(module, DualTransformer2DModel):
                 parent_name, index = name.rsplit(".", 1)
@@ -330,7 +344,8 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
 
         # get prompt text embeddings
         image_input = self.image_feature_extractor(images=prompt, return_tensors="pt")
-        image_embeddings = self.image_encoder(image_input.pixel_values.to(device))
+        pixel_values = image_input.pixel_values.to(device).to(self.image_encoder.dtype)
+        image_embeddings = self.image_encoder(pixel_values)
         image_embeddings = normalize_embeddings(image_embeddings)
 
         # duplicate image embeddings for each generation per prompt, using mps friendly method
@@ -340,9 +355,10 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
 
         # get unconditional embeddings for classifier free guidance
         if do_classifier_free_guidance:
-            uncond_images = [np.zeros((512, 512, 3))] * batch_size
+            uncond_images = [np.zeros((512, 512, 3)) + 0.5] * batch_size
             uncond_images = self.image_feature_extractor(images=uncond_images, return_tensors="pt")
-            uncond_embeddings = self.image_encoder(uncond_images.pixel_values.to(device))
+            pixel_values = uncond_images.pixel_values.to(device).to(self.image_encoder.dtype)
+            uncond_embeddings = self.image_encoder(pixel_values)
             uncond_embeddings = normalize_embeddings(uncond_embeddings)
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
@@ -384,23 +400,11 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
             extra_step_kwargs["generator"] = generator
         return extra_step_kwargs
 
-    def check_inputs(self, first_prompt, second_prompt, height, width, callback_steps):
-        if (
-            not isinstance(first_prompt, str)
-            and not isinstance(first_prompt, PIL.Image.Image)
-            and not isinstance(first_prompt, list)
-        ):
-            raise ValueError(
-                f"`first_prompt` has to be of type `str` `PIL.Image` or `list` but is {type(first_prompt)}"
-            )
-        if (
-            not isinstance(second_prompt, str)
-            and not isinstance(second_prompt, PIL.Image.Image)
-            and not isinstance(second_prompt, list)
-        ):
-            raise ValueError(
-                f"`second_prompt` has to be of type `str` `PIL.Image` or `list` but is {type(second_prompt)}"
-            )
+    def check_inputs(self, prompt, image, height, width, callback_steps):
+        if not isinstance(prompt, str) and not isinstance(prompt, PIL.Image.Image) and not isinstance(prompt, list):
+            raise ValueError(f"`prompt` has to be of type `str` `PIL.Image` or `list` but is {type(prompt)}")
+        if not isinstance(image, str) and not isinstance(image, PIL.Image.Image) and not isinstance(image, list):
+            raise ValueError(f"`image` has to be of type `str` `PIL.Image` or `list` but is {type(image)}")
 
         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}.")
@@ -431,17 +435,25 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
         latents = latents * self.scheduler.init_noise_sigma
         return latents
 
-    def set_mix_ratio(self, mix_ratio):
+    def set_transformer_params(self, mix_ratio: float = 0.5, condition_types: Tuple = ("text", "image")):
         for name, module in self.image_unet.named_modules():
             if isinstance(module, DualTransformer2DModel):
                 module.mix_ratio = mix_ratio
 
+                for i, type in enumerate(condition_types):
+                    if type == "text":
+                        module.condition_lengths[i] = self.text_encoder.config.max_position_embeddings
+                        module.transformer_index_for_condition[i] = 1  # use the second (text) transformer
+                    else:
+                        module.condition_lengths[i] = 257
+                        module.transformer_index_for_condition[i] = 0  # use the first (image) transformer
+
     @torch.no_grad()
     def __call__(
         self,
-        first_prompt: Union[str, List[str], PIL.Image.Image, List[PIL.Image.Image]],
-        second_prompt: Union[str, List[str], PIL.Image.Image, List[PIL.Image.Image]],
-        prompt_mix_ratio: float = 0.5,
+        prompt: Union[PIL.Image.Image, List[PIL.Image.Image]],
+        image: Union[str, List[str]],
+        text_to_image_strength: float = 0.5,
         height: int = 512,
         width: int = 512,
         num_inference_steps: int = 50,
@@ -503,21 +515,50 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.
 
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionDualGuidedPipeline
+        >>> import torch
+        >>> import requests
+        >>> from io import BytesIO
+        >>> from PIL import Image
+
+        >>> # let's download an initial image
+        >>> url = "https://huggingface.co/datasets/diffusers/images/resolve/main/benz.jpg"
+
+        >>> response = requests.get(url)
+        >>> image = Image.open(BytesIO(response.content)).convert("RGB")
+        >>> text = "a red car in the sun"
+
+        >>> pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe.remove_unused_weights()
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> text_to_image_strength = 0.75
+
+        >>> image = pipe(
+        ...     prompt=text, image=image, text_to_image_strength=text_to_image_strength, generator=generator
+        ... ).images[0]
+        >>> image.save("./car_variation.png")
+        ```
+
         Returns:
-            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
-            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
-            When returning a tuple, the first element is a list with the generated images, and the second element is a
-            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
-            (nsfw) content, according to the `safety_checker`.
+            [`~pipelines.stable_diffusion.ImagePipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion.ImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple. When
+            returning a tuple, the first element is a list with the generated images.
         """
 
         # 1. Check inputs. Raise error if not correct
-        self.check_inputs(first_prompt, second_prompt, height, width, callback_steps)
+        self.check_inputs(prompt, image, height, width, callback_steps)
 
         # 2. Define call parameters
-        first_prompt = [first_prompt] if not isinstance(first_prompt, list) else first_prompt
-        second_prompt = [second_prompt] if not isinstance(second_prompt, list) else second_prompt
-        batch_size = len(first_prompt)
+        prompt = [prompt] if not isinstance(prompt, list) else prompt
+        image = [image] if not isinstance(image, list) else image
+        batch_size = len(prompt)
         device = self._execution_device
         # 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`
@@ -525,21 +566,10 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
         do_classifier_free_guidance = guidance_scale > 1.0
 
         # 3. Encode input prompts
-        dual_prompt_embeddings = []
-        prompt_types = []
-        for prompt in [first_prompt, second_prompt]:
-            if isinstance(prompt[0], str):
-                embeddings = self._encode_text_prompt(
-                    prompt, device, num_images_per_prompt, do_classifier_free_guidance
-                )
-                prompt_types.append("text")
-            else:
-                embeddings = self._encode_image_prompt(
-                    prompt, device, num_images_per_prompt, do_classifier_free_guidance
-                )
-                prompt_types.append("image")
-            dual_prompt_embeddings.append(embeddings)
-        dual_prompt_embeddings = torch.cat(dual_prompt_embeddings, dim=1)
+        text_embeddings = self._encode_text_prompt(prompt, device, num_images_per_prompt, do_classifier_free_guidance)
+        image_embeddings = self._encode_image_prompt(image, device, num_images_per_prompt, do_classifier_free_guidance)
+        dual_prompt_embeddings = torch.cat([text_embeddings, image_embeddings], dim=1)
+        prompt_types = ("text", "image")
 
         # 4. Prepare timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)
@@ -562,8 +592,7 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
         # 7. Combine the attention blocks of the image and text UNets
-        self.convert_to_dual_attention(prompt_mix_ratio, prompt_types)
-        self.set_mix_ratio(prompt_mix_ratio)
+        self.set_transformer_params(text_to_image_strength, prompt_types)
 
         # 8. Denoising loop
         for i, t in enumerate(self.progress_bar(timesteps)):
@@ -586,13 +615,10 @@ class VersatileDiffusionDualGuidedPipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        # 9. Return the image unet to its original state
-        self.remove_dual_attention()
-
-        # 10. Post-processing
+        # 9. Post-processing
         image = self.decode_latents(latents)
 
-        # 11. Convert to PIL
+        # 10. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py

@@ -186,7 +186,8 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
 
         # get prompt text embeddings
         image_input = self.image_feature_extractor(images=prompt, return_tensors="pt")
-        image_embeddings = self.image_encoder(image_input.pixel_values.to(device))
+        pixel_values = image_input.pixel_values.to(device).to(self.image_encoder.dtype)
+        image_embeddings = self.image_encoder(pixel_values)
         image_embeddings = normalize_embeddings(image_embeddings)
 
         # duplicate image embeddings for each generation per prompt, using mps friendly method
@@ -198,7 +199,7 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_images: List[str]
             if negative_prompt is None:
-                uncond_images = [np.zeros((512, 512, 3))] * batch_size
+                uncond_images = [np.zeros((512, 512, 3)) + 0.5] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -216,7 +217,8 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
                 uncond_images = negative_prompt
 
             uncond_images = self.image_feature_extractor(images=uncond_images, return_tensors="pt")
-            uncond_embeddings = self.image_encoder(uncond_images.pixel_values.to(device))
+            pixel_values = uncond_images.pixel_values.to(device).to(self.image_encoder.dtype)
+            uncond_embeddings = self.image_encoder(pixel_values)
             uncond_embeddings = normalize_embeddings(uncond_embeddings)
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
@@ -357,6 +359,31 @@ class VersatileDiffusionImageVariationPipeline(DiffusionPipeline):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.
 
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionImageVariationPipeline
+        >>> import torch
+        >>> import requests
+        >>> from io import BytesIO
+        >>> from PIL import Image
+
+        >>> # let's download an initial image
+        >>> url = "https://huggingface.co/datasets/diffusers/images/resolve/main/benz.jpg"
+
+        >>> response = requests.get(url)
+        >>> image = Image.open(BytesIO(response.content)).convert("RGB")
+
+        >>> pipe = VersatileDiffusionImageVariationPipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> image = pipe(image, generator=generator).images[0]
+        >>> image.save("./car_variation.png")
+        ```
+
         Returns:
             [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
             [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.

src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py

@@ -76,7 +76,13 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
             scheduler=scheduler,
         )
 
-    def swap_unet_attention_blocks(self):
+        if self.text_unet is not None:
+            self._swap_unet_attention_blocks()
+
+    def _swap_unet_attention_blocks(self):
+        """
+        Swap the `Transformer2DModel` blocks between the image and text UNets
+        """
         for name, module in self.image_unet.named_modules():
             if isinstance(module, Transformer2DModel):
                 parent_name, index = name.rsplit(".", 1)
@@ -86,6 +92,9 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
                     self.image_unet.get_submodule(parent_name)[index],
                 )
 
+    def remove_unused_weights(self):
+        self.register_modules(text_unet=None)
+
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_xformers_memory_efficient_attention with unet->image_unet
     def enable_xformers_memory_efficient_attention(self):
         r"""
@@ -410,6 +419,23 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
                 The frequency at which the `callback` function will be called. If not specified, the callback will be
                 called at every step.
 
+        Examples:
+
+        ```py
+        >>> from diffusers import VersatileDiffusionTextToImagePipeline
+        >>> import torch
+
+        >>> pipe = VersatileDiffusionTextToImagePipeline.from_pretrained(
+        ...     "shi-labs/versatile-diffusion", torch_dtype=torch.float16
+        ... )
+        >>> pipe.remove_unused_weights()
+        >>> pipe = pipe.to("cuda")
+
+        >>> generator = torch.Generator(device="cuda").manual_seed(0)
+        >>> image = pipe("an astronaut riding on a horse on mars", generator=generator).images[0]
+        >>> image.save("./astronaut.png")
+        ```
+
         Returns:
             [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
             [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
@@ -454,10 +480,7 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
         # 6. Prepare extra step kwargs.
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
-        # 7. Swap the attention blocks between the image and text UNets
-        self.swap_unet_attention_blocks()
-
-        # 8. Denoising loop
+        # 7. Denoising loop
         for i, t in enumerate(self.progress_bar(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
@@ -478,13 +501,10 @@ class VersatileDiffusionTextToImagePipeline(DiffusionPipeline):
             if callback is not None and i % callback_steps == 0:
                 callback(i, t, latents)
 
-        # 9. Swap the attention blocks backs in case the UNets are reused in another pipeline
-        self.swap_unet_attention_blocks()
-
-        # 10. Post-processing
+        # 9. Post-processing
         image = self.decode_latents(latents)
 
-        # 11. Convert to PIL
+        # 10. Convert to PIL
         if output_type == "pil":
             image = self.numpy_to_pil(image)
 

src/diffusers/schedulers/scheduling_euler_ancestral_discrete.py

@@ -189,7 +189,7 @@ class EulerAncestralDiscreteScheduler(SchedulerMixin, ConfigMixin):
             )
 
         if not self.is_scale_input_called:
-            logger.warn(
+            logger.warning(
                 "The `scale_model_input` function should be called before `step` to ensure correct denoising. "
                 "See `StableDiffusionPipeline` for a usage example."
             )

src/diffusers/schedulers/scheduling_euler_discrete.py

@@ -198,7 +198,7 @@ class EulerDiscreteScheduler(SchedulerMixin, ConfigMixin):
             )
 
         if not self.is_scale_input_called:
-            logger.warn(
+            logger.warning(
                 "The `scale_model_input` function should be called before `step` to ensure correct denoising. "
                 "See `StableDiffusionPipeline` for a usage example."
             )

src/diffusers/utils/__init__.py

@@ -33,6 +33,7 @@ from .import_utils import (
     is_torch_available,
     is_torch_version,
     is_transformers_available,
+    is_transformers_version,
     is_unidecode_available,
     requires_backends,
 )

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -64,6 +64,21 @@ class LDMTextToImagePipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class StableDiffusionImageVariationPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class StableDiffusionImg2ImgPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 
@@ -124,6 +139,36 @@ class StableDiffusionPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class StableDiffusionPipelineSafe(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
+class VersatileDiffusionDualGuidedPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class VersatileDiffusionImageVariationPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

src/diffusers/utils/import_utils.py

@@ -303,6 +303,17 @@ def requires_backends(obj, backends):
     if failed:
         raise ImportError("".join(failed))
 
+    if name in [
+        "VersatileDiffusionTextToImagePipeline",
+        "VersatileDiffusionPipeline",
+        "VersatileDiffusionDualGuidedPipeline",
+        "StableDiffusionImageVariationPipeline",
+    ] and is_transformers_version("<", "4.25.0.dev0"):
+        raise ImportError(
+            f"You need to install `transformers` from 'main' in order to use {name}: \n```\n pip install"
+            " git+https://github.com/huggingface/transformers \n```"
+        )
+
 
 class DummyObject(type):
     """
@@ -347,3 +358,17 @@ def is_torch_version(operation: str, version: str):
             A string version of PyTorch
     """
     return compare_versions(parse(_torch_version), operation, version)
+
+
+def is_transformers_version(operation: str, version: str):
+    """
+    Args:
+    Compares the current Transformers version to a given reference with an operation.
+        operation (`str`):
+            A string representation of an operator, such as `">"` or `"<="`
+        version (`str`):
+            A string version of PyTorch
+    """
+    if not _transformers_available:
+        return False
+    return compare_versions(parse(_transformers_version), operation, version)

tests/pipelines/latent_diffusion/test_latent_diffusion_superresolution.py

@@ -87,6 +87,27 @@ class LDMSuperResolutionPipelineFastTests(PipelineTesterMixin, unittest.TestCase
         expected_slice = np.array([0.8678, 0.8245, 0.6381, 0.6830, 0.4385, 0.5599, 0.4641, 0.6201, 0.5150])
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
 
+    @unittest.skipIf(torch_device != "cuda", "This test requires a GPU")
+    def test_inference_superresolution_fp16(self):
+        unet = self.dummy_uncond_unet
+        scheduler = DDIMScheduler()
+        vqvae = self.dummy_vq_model
+
+        # put models in fp16
+        unet = unet.half()
+        vqvae = vqvae.half()
+
+        ldm = LDMSuperResolutionPipeline(unet=unet, vqvae=vqvae, scheduler=scheduler)
+        ldm.to(torch_device)
+        ldm.set_progress_bar_config(disable=None)
+
+        init_image = self.dummy_image.to(torch_device)
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = ldm(init_image, generator=generator, num_inference_steps=2, output_type="numpy").images
+
+        assert image.shape == (1, 64, 64, 3)
+
 
 @slow
 @require_torch

tests/pipelines/stable_diffusion/test_stable_diffusion_image_variation.py

@@ -0,0 +1,424 @@
+# coding=utf-8
+# Copyright 2022 HuggingFace Inc.
+#
+# 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.
+
+import gc
+import random
+import unittest
+
+import numpy as np
+import torch
+
+from diffusers import (
+    AutoencoderKL,
+    LMSDiscreteScheduler,
+    PNDMScheduler,
+    StableDiffusionImageVariationPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.utils import floats_tensor, load_image, load_numpy, slow, torch_device
+from diffusers.utils.testing_utils import require_torch_gpu
+from transformers import CLIPVisionConfig, CLIPVisionModelWithProjection
+
+from ...test_pipelines_common import PipelineTesterMixin
+
+
+torch.backends.cuda.matmul.allow_tf32 = False
+
+
+class StableDiffusionImageVariationPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    @property
+    def dummy_image(self):
+        batch_size = 1
+        num_channels = 3
+        sizes = (32, 32)
+
+        image = floats_tensor((batch_size, num_channels) + sizes, rng=random.Random(0)).to(torch_device)
+        return image
+
+    @property
+    def dummy_cond_unet(self):
+        torch.manual_seed(0)
+        model = UNet2DConditionModel(
+            block_out_channels=(32, 64),
+            layers_per_block=2,
+            sample_size=32,
+            in_channels=4,
+            out_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
+            cross_attention_dim=32,
+        )
+        return model
+
+    @property
+    def dummy_vae(self):
+        torch.manual_seed(0)
+        model = AutoencoderKL(
+            block_out_channels=[32, 64],
+            in_channels=3,
+            out_channels=3,
+            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+            latent_channels=4,
+        )
+        return model
+
+    @property
+    def dummy_image_encoder(self):
+        torch.manual_seed(0)
+        config = CLIPVisionConfig(
+            hidden_size=32,
+            projection_dim=32,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=5,
+            image_size=32,
+            patch_size=4,
+        )
+        return CLIPVisionModelWithProjection(config)
+
+    @property
+    def dummy_extractor(self):
+        def extract(*args, **kwargs):
+            class Out:
+                def __init__(self):
+                    self.pixel_values = torch.ones([0])
+
+                def to(self, device):
+                    self.pixel_values.to(device)
+                    return self
+
+            return Out()
+
+        return extract
+
+    def test_stable_diffusion_img_variation_default_case(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        image_encoder = self.dummy_image_encoder
+
+        init_image = self.dummy_image.to(device)
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionImageVariationPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            image_encoder=image_encoder,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        output = sd_pipe(
+            init_image,
+            generator=generator,
+            guidance_scale=6.0,
+            num_inference_steps=2,
+            output_type="np",
+        )
+
+        image = output.images
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        image_from_tuple = sd_pipe(
+            init_image,
+            generator=generator,
+            guidance_scale=6.0,
+            num_inference_steps=2,
+            output_type="np",
+            return_dict=False,
+        )[0]
+
+        image_slice = image[0, -3:, -3:, -1]
+        print(image_slice.flatten())
+        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
+
+        assert image.shape == (1, 128, 128, 3)
+        expected_slice = np.array([0.4935, 0.4784, 0.4802, 0.5027, 0.4805, 0.5149, 0.5143, 0.4879, 0.4731])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3
+        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-3
+
+    def test_stable_diffusion_img_variation_multiple_images(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        image_encoder = self.dummy_image_encoder
+
+        init_image = self.dummy_image.to(device).repeat(2, 1, 1, 1)
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionImageVariationPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            image_encoder=image_encoder,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        output = sd_pipe(
+            init_image,
+            generator=generator,
+            guidance_scale=6.0,
+            num_inference_steps=2,
+            output_type="np",
+        )
+
+        image = output.images
+
+        image_slice = image[-1, -3:, -3:, -1]
+
+        assert image.shape == (2, 128, 128, 3)
+        expected_slice = np.array([0.4939, 0.4627, 0.4831, 0.5710, 0.5387, 0.4428, 0.5230, 0.5545, 0.4586])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-3
+
+    def test_stable_diffusion_img_variation_num_images_per_prompt(self):
+        device = "cpu"
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        image_encoder = self.dummy_image_encoder
+
+        init_image = self.dummy_image.to(device)
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionImageVariationPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            image_encoder=image_encoder,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        # test num_images_per_prompt=1 (default)
+        images = sd_pipe(
+            init_image,
+            num_inference_steps=2,
+            output_type="np",
+        ).images
+
+        assert images.shape == (1, 128, 128, 3)
+
+        # test num_images_per_prompt=1 (default) for batch of images
+        batch_size = 2
+        images = sd_pipe(
+            init_image.repeat(batch_size, 1, 1, 1),
+            num_inference_steps=2,
+            output_type="np",
+        ).images
+
+        assert images.shape == (batch_size, 128, 128, 3)
+
+        # test num_images_per_prompt for single prompt
+        num_images_per_prompt = 2
+        images = sd_pipe(
+            init_image,
+            num_inference_steps=2,
+            output_type="np",
+            num_images_per_prompt=num_images_per_prompt,
+        ).images
+
+        assert images.shape == (num_images_per_prompt, 128, 128, 3)
+
+        # test num_images_per_prompt for batch of prompts
+        batch_size = 2
+        images = sd_pipe(
+            init_image.repeat(batch_size, 1, 1, 1),
+            num_inference_steps=2,
+            output_type="np",
+            num_images_per_prompt=num_images_per_prompt,
+        ).images
+
+        assert images.shape == (batch_size * num_images_per_prompt, 128, 128, 3)
+
+    @unittest.skipIf(torch_device != "cuda", "This test requires a GPU")
+    def test_stable_diffusion_img_variation_fp16(self):
+        """Test that stable diffusion img2img works with fp16"""
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        image_encoder = self.dummy_image_encoder
+
+        init_image = self.dummy_image.to(torch_device).float()
+
+        # put models in fp16
+        unet = unet.half()
+        vae = vae.half()
+        image_encoder = image_encoder.half()
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionImageVariationPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            image_encoder=image_encoder,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(torch_device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = sd_pipe(
+            init_image,
+            generator=generator,
+            num_inference_steps=2,
+            output_type="np",
+        ).images
+
+        assert image.shape == (1, 128, 128, 3)
+
+
+@slow
+@require_torch_gpu
+class StableDiffusionImageVariationPipelineIntegrationTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_stable_diffusion_img_variation_pipeline_default(self):
+        init_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/img2img/vermeer.jpg"
+        )
+        init_image = init_image.resize((512, 512))
+        expected_image = load_numpy(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/img2img/vermeer.npy"
+        )
+
+        model_id = "fusing/sd-image-variations-diffusers"
+        pipe = StableDiffusionImageVariationPipeline.from_pretrained(
+            model_id,
+            safety_checker=None,
+        )
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        pipe.enable_attention_slicing()
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        output = pipe(
+            init_image,
+            guidance_scale=7.5,
+            generator=generator,
+            output_type="np",
+        )
+        image = output.images[0]
+
+        assert image.shape == (512, 512, 3)
+        # img2img is flaky across GPUs even in fp32, so using MAE here
+        assert np.abs(expected_image - image).max() < 1e-3
+
+    def test_stable_diffusion_img_variation_intermediate_state(self):
+        number_of_steps = 0
+
+        def test_callback_fn(step: int, timestep: int, latents: torch.FloatTensor) -> None:
+            test_callback_fn.has_been_called = True
+            nonlocal number_of_steps
+            number_of_steps += 1
+            if step == 0:
+                latents = latents.detach().cpu().numpy()
+                assert latents.shape == (1, 4, 64, 64)
+                latents_slice = latents[0, -3:, -3:, -1]
+                expected_slice = np.array([1.83, 1.293, -0.09705, 1.256, -2.293, 1.091, -0.0809, -0.65, -2.953])
+                assert np.abs(latents_slice.flatten() - expected_slice).max() < 1e-3
+            elif step == 37:
+                latents = latents.detach().cpu().numpy()
+                assert latents.shape == (1, 4, 64, 64)
+                latents_slice = latents[0, -3:, -3:, -1]
+                expected_slice = np.array([2.285, 2.703, 1.969, 0.696, -1.323, 0.9253, -0.5464, -1.521, -2.537])
+                assert np.abs(latents_slice.flatten() - expected_slice).max() < 1e-2
+
+        test_callback_fn.has_been_called = False
+
+        init_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+            "/img2img/sketch-mountains-input.jpg"
+        )
+        init_image = init_image.resize((512, 512))
+
+        pipe = StableDiffusionImageVariationPipeline.from_pretrained(
+            "fusing/sd-image-variations-diffusers",
+            torch_dtype=torch.float16,
+        )
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        pipe.enable_attention_slicing()
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        with torch.autocast(torch_device):
+            pipe(
+                init_image,
+                num_inference_steps=50,
+                guidance_scale=7.5,
+                generator=generator,
+                callback=test_callback_fn,
+                callback_steps=1,
+            )
+        assert test_callback_fn.has_been_called
+        assert number_of_steps == 51
+
+    def test_stable_diffusion_pipeline_with_sequential_cpu_offloading(self):
+        torch.cuda.empty_cache()
+        torch.cuda.reset_max_memory_allocated()
+        torch.cuda.reset_peak_memory_stats()
+
+        init_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+            "/img2img/sketch-mountains-input.jpg"
+        )
+        init_image = init_image.resize((512, 512))
+
+        model_id = "fusing/sd-image-variations-diffusers"
+        lms = LMSDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
+        pipe = StableDiffusionImageVariationPipeline.from_pretrained(
+            model_id, scheduler=lms, safety_checker=None, torch_dtype=torch.float16
+        )
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+        pipe.enable_attention_slicing(1)
+        pipe.enable_sequential_cpu_offload()
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        _ = pipe(
+            init_image,
+            guidance_scale=7.5,
+            generator=generator,
+            output_type="np",
+            num_inference_steps=5,
+        )
+
+        mem_bytes = torch.cuda.max_memory_allocated()
+        # make sure that less than 2.6 GB is allocated
+        assert mem_bytes < 2.6 * 10**9

tests/pipelines/stable_diffusion_safe/test_safe_diffusion.py

@@ -0,0 +1,435 @@
+# coding=utf-8
+# Copyright 2022 HuggingFace Inc.
+#
+# 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.
+
+import gc
+import random
+import tempfile
+import unittest
+
+import numpy as np
+import torch
+
+from diffusers import AutoencoderKL, DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler, UNet2DConditionModel
+from diffusers.pipelines.stable_diffusion_safe import StableDiffusionPipelineSafe as StableDiffusionPipeline
+from diffusers.utils import floats_tensor, slow, torch_device
+from diffusers.utils.testing_utils import require_torch_gpu
+from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
+
+from ...test_pipelines_common import PipelineTesterMixin
+
+
+torch.backends.cuda.matmul.allow_tf32 = False
+
+
+class SafeDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    @property
+    def dummy_image(self):
+        batch_size = 1
+        num_channels = 3
+        sizes = (32, 32)
+
+        image = floats_tensor((batch_size, num_channels) + sizes, rng=random.Random(0)).to(torch_device)
+        return image
+
+    @property
+    def dummy_cond_unet(self):
+        torch.manual_seed(0)
+        model = UNet2DConditionModel(
+            block_out_channels=(32, 64),
+            layers_per_block=2,
+            sample_size=32,
+            in_channels=4,
+            out_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
+            cross_attention_dim=32,
+        )
+        return model
+
+    @property
+    def dummy_vae(self):
+        torch.manual_seed(0)
+        model = AutoencoderKL(
+            block_out_channels=[32, 64],
+            in_channels=3,
+            out_channels=3,
+            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+            latent_channels=4,
+        )
+        return model
+
+    @property
+    def dummy_text_encoder(self):
+        torch.manual_seed(0)
+        config = CLIPTextConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=32,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=5,
+            pad_token_id=1,
+            vocab_size=1000,
+        )
+        return CLIPTextModel(config)
+
+    @property
+    def dummy_extractor(self):
+        def extract(*args, **kwargs):
+            class Out:
+                def __init__(self):
+                    self.pixel_values = torch.ones([0])
+
+                def to(self, device):
+                    self.pixel_values.to(device)
+                    return self
+
+            return Out()
+
+        return extract
+
+    def test_safe_diffusion_ddim(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        unet = self.dummy_cond_unet
+        scheduler = DDIMScheduler(
+            beta_start=0.00085,
+            beta_end=0.012,
+            beta_schedule="scaled_linear",
+            clip_sample=False,
+            set_alpha_to_one=False,
+        )
+
+        vae = self.dummy_vae
+        bert = self.dummy_text_encoder
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=bert,
+            tokenizer=tokenizer,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = "A painting of a squirrel eating a burger"
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        output = sd_pipe([prompt], generator=generator, guidance_scale=6.0, num_inference_steps=2, output_type="np")
+        image = output.images
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        image_from_tuple = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=6.0,
+            num_inference_steps=2,
+            output_type="np",
+            return_dict=False,
+        )[0]
+
+        image_slice = image[0, -3:, -3:, -1]
+        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
+
+        assert image.shape == (1, 128, 128, 3)
+        expected_slice = np.array([0.5112, 0.4692, 0.4715, 0.5206, 0.4894, 0.5114, 0.5096, 0.4932, 0.4755])
+
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_stable_diffusion_pndm(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        bert = self.dummy_text_encoder
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=bert,
+            tokenizer=tokenizer,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = "A painting of a squirrel eating a burger"
+        generator = torch.Generator(device=device).manual_seed(0)
+        output = sd_pipe([prompt], generator=generator, guidance_scale=6.0, num_inference_steps=2, output_type="np")
+
+        image = output.images
+
+        generator = torch.Generator(device=device).manual_seed(0)
+        image_from_tuple = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=6.0,
+            num_inference_steps=2,
+            output_type="np",
+            return_dict=False,
+        )[0]
+
+        image_slice = image[0, -3:, -3:, -1]
+        image_from_tuple_slice = image_from_tuple[0, -3:, -3:, -1]
+
+        assert image.shape == (1, 128, 128, 3)
+        expected_slice = np.array([0.4937, 0.4649, 0.4716, 0.5145, 0.4889, 0.513, 0.513, 0.4905, 0.4738])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+        assert np.abs(image_from_tuple_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_stable_diffusion_no_safety_checker(self):
+        pipe = StableDiffusionPipeline.from_pretrained(
+            "hf-internal-testing/tiny-stable-diffusion-lms-pipe", safety_checker=None
+        )
+        assert isinstance(pipe, StableDiffusionPipeline)
+        assert isinstance(pipe.scheduler, LMSDiscreteScheduler)
+        assert pipe.safety_checker is None
+
+        image = pipe("example prompt", num_inference_steps=2).images[0]
+        assert image is not None
+
+        # check that there's no error when saving a pipeline with one of the models being None
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            pipe.save_pretrained(tmpdirname)
+            pipe = StableDiffusionPipeline.from_pretrained(tmpdirname)
+
+        # sanity check that the pipeline still works
+        assert pipe.safety_checker is None
+        image = pipe("example prompt", num_inference_steps=2).images[0]
+        assert image is not None
+
+    @unittest.skipIf(torch_device != "cuda", "This test requires a GPU")
+    def test_stable_diffusion_fp16(self):
+        """Test that stable diffusion works with fp16"""
+        unet = self.dummy_cond_unet
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        bert = self.dummy_text_encoder
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        # put models in fp16
+        unet = unet.half()
+        vae = vae.half()
+        bert = bert.half()
+
+        # make sure here that pndm scheduler skips prk
+        sd_pipe = StableDiffusionPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=bert,
+            tokenizer=tokenizer,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+        sd_pipe = sd_pipe.to(torch_device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = "A painting of a squirrel eating a burger"
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = sd_pipe([prompt], generator=generator, num_inference_steps=2, output_type="np").images
+
+        assert image.shape == (1, 128, 128, 3)
+
+
+@slow
+@require_torch_gpu
+class SafeDiffusionPipelineIntegrationTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_harm_safe_stable_diffusion(self):
+        sd_pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", safety_checker=None)
+        sd_pipe.scheduler = LMSDiscreteScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe = sd_pipe.to(torch_device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = (
+            "portrait of girl with smokey eyes makeup in abandoned hotel, grange clothes, redshift, wide high angle"
+            " coloured polaroid photograph with flash, kodak film, hyper real, stunning moody cinematography, with"
+            " anamorphic lenses, by maripol, fallen angels by wong kar - wai, style of suspiria and neon demon and"
+            " children from bahnhof zoo, detailed "
+        )
+        seed = 4003660346
+        guidance_scale = 7
+
+        # without safety guidance (sld_guidance_scale = 0)
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=0,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = [0.2278, 0.2231, 0.2249, 0.2333, 0.2303, 0.1885, 0.2273, 0.2144, 0.2176]
+
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+        # without safety guidance (strong configuration)
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=2000,
+            sld_warmup_steps=7,
+            sld_threshold=0.025,
+            sld_momentum_scale=0.5,
+            sld_mom_beta=0.7,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = [0.2383, 0.2276, 0.236, 0.2192, 0.2186, 0.2053, 0.1971, 0.1901, 0.1719]
+
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_nudity_safe_stable_diffusion(self):
+        sd_pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", safety_checker=None)
+        sd_pipe.scheduler = LMSDiscreteScheduler.from_config(sd_pipe.scheduler.config)
+        sd_pipe = sd_pipe.to(torch_device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = "padme amidala taking a bath artwork, safe for work, no nudity"
+        seed = 2734971755
+        guidance_scale = 7
+
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=0,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = [0.3502, 0.3622, 0.3396, 0.3642, 0.3478, 0.3318, 0.35, 0.3348, 0.3297]
+
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=2000,
+            sld_warmup_steps=7,
+            sld_threshold=0.025,
+            sld_momentum_scale=0.5,
+            sld_mom_beta=0.7,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = [0.5531, 0.5206, 0.4895, 0.5156, 0.5182, 0.4751, 0.4802, 0.4803, 0.4443]
+
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_nudity_safetychecker_safe_stable_diffusion(self):
+        sd_pipe = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
+        sd_pipe = sd_pipe.to(torch_device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        prompt = (
+            "the four horsewomen of the apocalypse, painting by tom of finland, gaston bussiere, craig mullins, j. c."
+            " leyendecker"
+        )
+        seed = 1044355234
+        guidance_scale = 12
+
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=0,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = np.array([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0])
+
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-7
+
+        generator = torch.Generator(device=torch_device).manual_seed(seed)
+        output = sd_pipe(
+            [prompt],
+            generator=generator,
+            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            output_type="np",
+            width=512,
+            height=512,
+            sld_guidance_scale=2000,
+            sld_warmup_steps=7,
+            sld_threshold=0.025,
+            sld_momentum_scale=0.5,
+            sld_mom_beta=0.7,
+        )
+
+        image = output.images
+        image_slice = image[0, -3:, -3:, -1]
+        expected_slice = np.array([0.5818, 0.6285, 0.6835, 0.6019, 0.625, 0.6754, 0.6096, 0.6334, 0.6561])
+        assert image.shape == (1, 512, 512, 3)
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

tests/pipelines/versatile_diffusion/test_versatile_diffusion_dual_guided.py

@@ -42,16 +42,22 @@ class VersatileDiffusionDualGuidedPipelineIntegrationTests(unittest.TestCase):
         gc.collect()
         torch.cuda.empty_cache()
 
-    def test_from_pretrained_save_pretrained(self):
-        pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained("diffusers/vd-official-test")
+    def test_remove_unused_weights_save_load(self):
+        pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained("shi-labs/versatile-diffusion")
+        # remove text_unet
+        pipe.remove_unused_weights()
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
+        second_prompt = load_image(
+            "https://raw.githubusercontent.com/SHI-Labs/Versatile-Diffusion/master/assets/benz.jpg"
+        )
+
         generator = torch.Generator(device=torch_device).manual_seed(0)
         image = pipe(
-            first_prompt="first prompt",
-            second_prompt="second prompt",
-            prompt_mix_ratio=0.75,
+            prompt="first prompt",
+            image=second_prompt,
+            text_to_image_strength=0.75,
             generator=generator,
             guidance_scale=7.5,
             num_inference_steps=2,
@@ -61,14 +67,15 @@ class VersatileDiffusionDualGuidedPipelineIntegrationTests(unittest.TestCase):
         with tempfile.TemporaryDirectory() as tmpdirname:
             pipe.save_pretrained(tmpdirname)
             pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained(tmpdirname)
+
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
         generator = generator.manual_seed(0)
         new_image = pipe(
-            first_prompt="first prompt",
-            second_prompt="second prompt",
-            prompt_mix_ratio=0.75,
+            prompt="first prompt",
+            image=second_prompt,
+            text_to_image_strength=0.75,
             generator=generator,
             guidance_scale=7.5,
             num_inference_steps=2,
@@ -77,8 +84,9 @@ class VersatileDiffusionDualGuidedPipelineIntegrationTests(unittest.TestCase):
 
         assert np.abs(image - new_image).sum() < 1e-5, "Models don't have the same forward pass"
 
-    def test_inference_image_variations(self):
-        pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained("diffusers/vd-official-test")
+    def test_inference_dual_guided(self):
+        pipe = VersatileDiffusionDualGuidedPipeline.from_pretrained("shi-labs/versatile-diffusion")
+        pipe.remove_unused_weights()
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
@@ -88,9 +96,9 @@ class VersatileDiffusionDualGuidedPipelineIntegrationTests(unittest.TestCase):
         )
         generator = torch.Generator(device=torch_device).manual_seed(0)
         image = pipe(
-            first_prompt=first_prompt,
-            second_prompt=second_prompt,
-            prompt_mix_ratio=0.75,
+            prompt=first_prompt,
+            image=second_prompt,
+            text_to_image_strength=0.75,
             generator=generator,
             guidance_scale=7.5,
             num_inference_steps=50,
@@ -100,5 +108,5 @@ class VersatileDiffusionDualGuidedPipelineIntegrationTests(unittest.TestCase):
         image_slice = image[0, 253:256, 253:256, -1]
 
         assert image.shape == (1, 512, 512, 3)
-        expected_slice = np.array([0.1811, 0.0430, 0.0433, 0.1082, 0.0144, 0.0306, 0.0683, 0.0248, 0.0876])
+        expected_slice = np.array([0.014, 0.0112, 0.0136, 0.0145, 0.0107, 0.0113, 0.0272, 0.0215, 0.0216])
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

tests/pipelines/versatile_diffusion/test_versatile_diffusion_image_to_text.py

@@ -1,57 +0,0 @@
-# coding=utf-8
-# Copyright 2022 HuggingFace Inc.
-#
-# 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.
-
-import unittest
-
-import numpy as np
-import torch
-
-from diffusers import VersatileDiffusionImageToTextPipeline, DDIMScheduler
-from diffusers.utils.testing_utils import load_image, require_torch_gpu, slow, torch_device
-
-from ...test_pipelines_common import PipelineTesterMixin
-
-
-torch.backends.cuda.matmul.allow_tf32 = False
-
-
-class VersatileDiffusionImageToTextPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
-    pass
-
-
-@slow
-@require_torch_gpu
-class VersatileDiffusionImageToTextPipelineIntegrationTests(unittest.TestCase):
-    def test_inference_image_to_text(self):
-        pipe = VersatileDiffusionImageToTextPipeline.from_pretrained("diffusers/vd-official-test")
-        pipe.to(torch_device)
-        pipe.set_progress_bar_config(disable=None)
-
-        image_prompt = load_image(
-            "https://raw.githubusercontent.com/SHI-Labs/Versatile-Diffusion/master/assets/boy_and_girl.jpg"
-        )
-        # generator = torch.Generator(device=torch_device).manual_seed(0)
-        np.random.seed(8)
-        torch.manual_seed(108)
-        pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
-        text = pipe(
-            image=image_prompt,
-            # generator=generator,
-            guidance_scale=7.5,
-            num_inference_steps=50,
-            output_type="str",
-        ).text
-
-        assert text == "Corret me"

tests/pipelines/versatile_diffusion/test_versatile_diffusion_image_variation.py

@@ -35,7 +35,7 @@ class VersatileDiffusionImageVariationPipelineFastTests(PipelineTesterMixin, uni
 @require_torch_gpu
 class VersatileDiffusionImageVariationPipelineIntegrationTests(unittest.TestCase):
     def test_inference_image_variations(self):
-        pipe = VersatileDiffusionImageVariationPipeline.from_pretrained("diffusers/vd-official-test")
+        pipe = VersatileDiffusionImageVariationPipeline.from_pretrained("shi-labs/versatile-diffusion")
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
@@ -54,5 +54,5 @@ class VersatileDiffusionImageVariationPipelineIntegrationTests(unittest.TestCase
         image_slice = image[0, 253:256, 253:256, -1]
 
         assert image.shape == (1, 512, 512, 3)
-        expected_slice = np.array([0.1811, 0.0430, 0.0433, 0.1082, 0.0144, 0.0306, 0.0683, 0.0248, 0.0876])
+        expected_slice = np.array([0.0113, 0.2241, 0.4024, 0.0839, 0.0871, 0.2725, 0.2581, 0.0, 0.1096])
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

tests/pipelines/versatile_diffusion/test_versatile_diffusion_mega.py

@@ -0,0 +1,129 @@
+# coding=utf-8
+# Copyright 2022 HuggingFace Inc.
+#
+# 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.
+
+import gc
+import tempfile
+import unittest
+
+import numpy as np
+import torch
+
+from diffusers import VersatileDiffusionPipeline
+from diffusers.utils.testing_utils import load_image, require_torch_gpu, slow, torch_device
+
+from ...test_pipelines_common import PipelineTesterMixin
+
+
+torch.backends.cuda.matmul.allow_tf32 = False
+
+
+class VersatileDiffusionMegaPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pass
+
+
+@slow
+@require_torch_gpu
+class VersatileDiffusionMegaPipelineIntegrationTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_from_pretrained_save_pretrained(self):
+        pipe = VersatileDiffusionPipeline.from_pretrained("shi-labs/versatile-diffusion", torch_dtype=torch.float16)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        prompt_image = load_image(
+            "https://raw.githubusercontent.com/SHI-Labs/Versatile-Diffusion/master/assets/benz.jpg"
+        )
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = pipe.dual_guided(
+            prompt="first prompt",
+            image=prompt_image,
+            text_to_image_strength=0.75,
+            generator=generator,
+            guidance_scale=7.5,
+            num_inference_steps=2,
+            output_type="numpy",
+        ).images
+
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            pipe.save_pretrained(tmpdirname)
+            pipe = VersatileDiffusionPipeline.from_pretrained(tmpdirname, torch_dtype=torch.float16)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator = generator.manual_seed(0)
+        new_image = pipe.dual_guided(
+            prompt="first prompt",
+            image=prompt_image,
+            text_to_image_strength=0.75,
+            generator=generator,
+            guidance_scale=7.5,
+            num_inference_steps=2,
+            output_type="numpy",
+        ).images
+
+        assert np.abs(image - new_image).sum() < 1e-5, "Models don't have the same forward pass"
+
+    def test_inference_dual_guided_then_text_to_image(self):
+        pipe = VersatileDiffusionPipeline.from_pretrained("shi-labs/versatile-diffusion", torch_dtype=torch.float16)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        prompt = "cyberpunk 2077"
+        init_image = load_image(
+            "https://raw.githubusercontent.com/SHI-Labs/Versatile-Diffusion/master/assets/benz.jpg"
+        )
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = pipe.dual_guided(
+            prompt=prompt,
+            image=init_image,
+            text_to_image_strength=0.75,
+            generator=generator,
+            guidance_scale=7.5,
+            num_inference_steps=50,
+            output_type="numpy",
+        ).images
+
+        image_slice = image[0, 253:256, 253:256, -1]
+
+        assert image.shape == (1, 512, 512, 3)
+        expected_slice = np.array([0.014, 0.0112, 0.0136, 0.0145, 0.0107, 0.0113, 0.0272, 0.0215, 0.0216])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+        prompt = "A painting of a squirrel eating a burger "
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = pipe.text_to_image(
+            prompt=prompt, generator=generator, guidance_scale=7.5, num_inference_steps=50, output_type="numpy"
+        ).images
+
+        image_slice = image[0, 253:256, 253:256, -1]
+
+        assert image.shape == (1, 512, 512, 3)
+        expected_slice = np.array([0.0408, 0.0181, 0.0, 0.0388, 0.0046, 0.0461, 0.0411, 0.0, 0.0222])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+        pipe = VersatileDiffusionPipeline.from_pretrained("shi-labs/versatile-diffusion", torch_dtype=torch.float16)
+        image = pipe.image_variation(init_image, generator=generator, output_type="numpy").images[0]
+
+        image_slice = image[0, 253:256, 253:256, -1]
+
+        assert image.shape == (1, 512, 512, 3)
+        expected_slice = np.array([0.0657, 0.0529, 0.0455, 0.0802, 0.0570, 0.0179, 0.0267, 0.0483, 0.0769])
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

tests/pipelines/versatile_diffusion/test_versatile_diffusion_text_to_image.py

@@ -13,6 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import gc
+import tempfile
 import unittest
 
 import numpy as np
@@ -34,8 +36,40 @@ class VersatileDiffusionTextToImagePipelineFastTests(PipelineTesterMixin, unitte
 @slow
 @require_torch_gpu
 class VersatileDiffusionTextToImagePipelineIntegrationTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def test_remove_unused_weights_save_load(self):
+        pipe = VersatileDiffusionTextToImagePipeline.from_pretrained("shi-labs/versatile-diffusion")
+        # remove text_unet
+        pipe.remove_unused_weights()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        prompt = "A painting of a squirrel eating a burger "
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        image = pipe(
+            prompt=prompt, generator=generator, guidance_scale=7.5, num_inference_steps=2, output_type="numpy"
+        ).images
+
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            pipe.save_pretrained(tmpdirname)
+            pipe = VersatileDiffusionTextToImagePipeline.from_pretrained(tmpdirname)
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator = generator.manual_seed(0)
+        new_image = pipe(
+            prompt=prompt, generator=generator, guidance_scale=7.5, num_inference_steps=2, output_type="numpy"
+        ).images
+
+        assert np.abs(image - new_image).sum() < 1e-5, "Models don't have the same forward pass"
+
     def test_inference_text2img(self):
-        pipe = VersatileDiffusionTextToImagePipeline.from_pretrained("diffusers/vd-official-test")
+        pipe = VersatileDiffusionTextToImagePipeline.from_pretrained("shi-labs/versatile-diffusion")
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 
@@ -48,5 +82,5 @@ class VersatileDiffusionTextToImagePipelineIntegrationTests(unittest.TestCase):
         image_slice = image[0, 253:256, 253:256, -1]
 
         assert image.shape == (1, 512, 512, 3)
-        expected_slice = np.array([0.0657, 0.0529, 0.0455, 0.0802, 0.0570, 0.0179, 0.0267, 0.0483, 0.0769])
+        expected_slice = np.array([0.0408, 0.0181, 0.0, 0.0388, 0.0046, 0.0461, 0.0411, 0.0, 0.0222])
         assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2

tests/test_pipelines.py

@@ -18,6 +18,7 @@ import os
 import random
 import tempfile
 import unittest
+from functools import partial
 
 import numpy as np
 import torch
@@ -46,6 +47,7 @@ from diffusers.pipeline_utils import DiffusionPipeline
 from diffusers.schedulers.scheduling_utils import SCHEDULER_CONFIG_NAME
 from diffusers.utils import CONFIG_NAME, WEIGHTS_NAME, floats_tensor, slow, torch_device
 from diffusers.utils.testing_utils import CaptureLogger, get_tests_dir, require_torch_gpu
+from parameterized import parameterized
 from PIL import Image
 from transformers import CLIPFeatureExtractor, CLIPModel, CLIPTextConfig, CLIPTextModel, CLIPTokenizer
 
@@ -247,7 +249,6 @@ class CustomPipelineTests(unittest.TestCase):
 
 
 class PipelineFastTests(unittest.TestCase):
-    @property
     def dummy_image(self):
         batch_size = 1
         num_channels = 3
@@ -256,13 +257,12 @@ class PipelineFastTests(unittest.TestCase):
         image = floats_tensor((batch_size, num_channels) + sizes, rng=random.Random(0)).to(torch_device)
         return image
 
-    @property
-    def dummy_uncond_unet(self):
+    def dummy_uncond_unet(self, sample_size=32):
         torch.manual_seed(0)
         model = UNet2DModel(
             block_out_channels=(32, 64),
             layers_per_block=2,
-            sample_size=32,
+            sample_size=sample_size,
             in_channels=3,
             out_channels=3,
             down_block_types=("DownBlock2D", "AttnDownBlock2D"),
@@ -270,13 +270,12 @@ class PipelineFastTests(unittest.TestCase):
         )
         return model
 
-    @property
-    def dummy_cond_unet(self):
+    def dummy_cond_unet(self, sample_size=32):
         torch.manual_seed(0)
         model = UNet2DConditionModel(
             block_out_channels=(32, 64),
             layers_per_block=2,
-            sample_size=32,
+            sample_size=sample_size,
             in_channels=4,
             out_channels=4,
             down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
@@ -285,13 +284,12 @@ class PipelineFastTests(unittest.TestCase):
         )
         return model
 
-    @property
-    def dummy_cond_unet_inpaint(self):
+    def dummy_cond_unet_inpaint(self, sample_size=32):
         torch.manual_seed(0)
         model = UNet2DConditionModel(
             block_out_channels=(32, 64),
             layers_per_block=2,
-            sample_size=32,
+            sample_size=sample_size,
             in_channels=9,
             out_channels=4,
             down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
@@ -300,7 +298,6 @@ class PipelineFastTests(unittest.TestCase):
         )
         return model
 
-    @property
     def dummy_vq_model(self):
         torch.manual_seed(0)
         model = VQModel(
@@ -313,7 +310,6 @@ class PipelineFastTests(unittest.TestCase):
         )
         return model
 
-    @property
     def dummy_vae(self):
         torch.manual_seed(0)
         model = AutoencoderKL(
@@ -326,7 +322,6 @@ class PipelineFastTests(unittest.TestCase):
         )
         return model
 
-    @property
     def dummy_text_encoder(self):
         torch.manual_seed(0)
         config = CLIPTextConfig(
@@ -342,7 +337,6 @@ class PipelineFastTests(unittest.TestCase):
         )
         return CLIPTextModel(config)
 
-    @property
     def dummy_extractor(self):
         def extract(*args, **kwargs):
             class Out:
@@ -357,15 +351,43 @@ class PipelineFastTests(unittest.TestCase):
 
         return extract
 
-    def test_components(self):
+    @parameterized.expand(
+        [
+            [DDIMScheduler, DDIMPipeline, 32],
+            [partial(DDPMScheduler, predict_epsilon=True), DDPMPipeline, 32],
+            [DDIMScheduler, DDIMPipeline, (32, 64)],
+            [partial(DDPMScheduler, predict_epsilon=True), DDPMPipeline, (64, 32)],
+        ]
+    )
+    def test_uncond_unet_components(self, scheduler_fn=DDPMScheduler, pipeline_fn=DDPMPipeline, sample_size=32):
+        unet = self.dummy_uncond_unet(sample_size)
+        # DDIM doesn't take `predict_epsilon`, and DDPM requires it -- so using partial in parameterized decorator
+        scheduler = scheduler_fn()
+        pipeline = pipeline_fn(unet, scheduler).to(torch_device)
+
+        # Device type MPS is not supported for torch.Generator() api.
+        if torch_device == "mps":
+            generator = torch.manual_seed(0)
+        else:
+            generator = torch.Generator(device=torch_device).manual_seed(0)
+
+        out_image = pipeline(
+            generator=generator,
+            num_inference_steps=2,
+            output_type="np",
+        ).images
+        sample_size = (sample_size, sample_size) if isinstance(sample_size, int) else sample_size
+        assert out_image.shape == (1, *sample_size, 3)
+
+    def test_stable_diffusion_components(self):
         """Test that components property works correctly"""
-        unet = self.dummy_cond_unet
+        unet = self.dummy_cond_unet()
         scheduler = PNDMScheduler(skip_prk_steps=True)
-        vae = self.dummy_vae
-        bert = self.dummy_text_encoder
+        vae = self.dummy_vae()
+        bert = self.dummy_text_encoder()
         tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
 
-        image = self.dummy_image.cpu().permute(0, 2, 3, 1)[0]
+        image = self.dummy_image().cpu().permute(0, 2, 3, 1)[0]
         init_image = Image.fromarray(np.uint8(image)).convert("RGB")
         mask_image = Image.fromarray(np.uint8(image + 4)).convert("RGB").resize((128, 128))
 
@@ -377,7 +399,7 @@ class PipelineFastTests(unittest.TestCase):
             text_encoder=bert,
             tokenizer=tokenizer,
             safety_checker=None,
-            feature_extractor=self.dummy_extractor,
+            feature_extractor=self.dummy_extractor(),
         ).to(torch_device)
         img2img = StableDiffusionImg2ImgPipeline(**inpaint.components).to(torch_device)
         text2img = StableDiffusionPipeline(**inpaint.components).to(torch_device)