update

* fix: animate diff lora stuff.

* fix scaling function for UNetMotionModel

* emoty

.github/workflows/build_docker_images.yml

@@ -20,7 +20,8 @@ env:
 
 jobs:
   test-build-docker-images:
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     if: github.event_name == 'pull_request'
     steps:
       - name: Set up Docker Buildx
@@ -50,7 +51,8 @@ jobs:
         if: steps.file_changes.outputs.all != ''
 
   build-and-push-docker-images:
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     if: github.event_name != 'pull_request'
 
     permissions:
@@ -98,4 +100,4 @@ jobs:
           slack_channel: ${{ env.CI_SLACK_CHANNEL }}
           title: "🤗 Results of the ${{ matrix.image-name }} Docker Image build"
           status: ${{ job.status }}
-          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
+          slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

.github/workflows/nightly_tests.yml

@@ -19,7 +19,8 @@ env:
 jobs:
   setup_torch_cuda_pipeline_matrix:
     name: Setup Torch Pipelines CUDA Slow Tests Matrix
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     container:
       image: diffusers/diffusers-pytorch-cpu
     outputs:
@@ -55,7 +56,8 @@ jobs:
       max-parallel: 8
       matrix:
         module: ${{ fromJson(needs.setup_torch_cuda_pipeline_matrix.outputs.pipeline_test_matrix) }}
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
       options: --shm-size "16gb" --ipc host --gpus 0
@@ -105,7 +107,8 @@ jobs:
 
   run_nightly_tests_for_other_torch_modules:
     name: Nightly Torch CUDA Tests
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
       options: --shm-size "16gb" --ipc host --gpus 0
@@ -234,7 +237,8 @@ jobs:
 
   run_nightly_onnx_tests:
     name: Nightly ONNXRuntime CUDA tests on Ubuntu
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-onnxruntime-cuda
       options: --gpus 0 --shm-size "16gb" --ipc host

.github/workflows/pr_test_fetcher.yml

@@ -15,7 +15,8 @@ concurrency:
 jobs:
   setup_pr_tests:
     name: Setup PR Tests
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     container:
       image: diffusers/diffusers-pytorch-cpu
       options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
@@ -73,7 +74,8 @@ jobs:
       max-parallel: 2
       matrix:
         modules: ${{ fromJson(needs.setup_pr_tests.outputs.matrix) }}
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     container:
       image: diffusers/diffusers-pytorch-cpu
       options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
@@ -123,12 +125,13 @@ jobs:
         config:
           - name: Hub tests for models, schedulers, and pipelines
             framework: hub_tests_pytorch
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_hub
 
     name: ${{ matrix.config.name }}
-    runs-on: ${{ matrix.config.runner }}
+    runs-on:
+      group: ${{ matrix.config.runner }}
     container:
       image: ${{ matrix.config.image }}
       options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/

.github/workflows/pr_test_peft_backend.yml

@@ -71,7 +71,8 @@ jobs:
 
     name: LoRA - ${{ matrix.lib-versions }}
 
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
 
     container:
       image: diffusers/diffusers-pytorch-cpu
@@ -128,4 +129,4 @@ jobs:
       uses: actions/upload-artifact@v2
       with:
         name: pr_${{ matrix.config.report }}_test_reports
-        path: reports
+        path: reports

.github/workflows/pr_tests.yml

@@ -77,28 +77,29 @@ jobs:
         config:
           - name: Fast PyTorch Pipeline CPU tests
             framework: pytorch_pipelines
-            runner: [ self-hosted, intel-cpu, 32-cpu, 256-ram, ci ]
+            runner: aws-highmemory-32-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_cpu_pipelines
           - name: Fast PyTorch Models & Schedulers CPU tests
             framework: pytorch_models
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_cpu_models_schedulers
           - name: Fast Flax CPU tests
             framework: flax
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-flax-cpu
             report: flax_cpu
           - name: PyTorch Example CPU tests
             framework: pytorch_examples
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_example_cpu
 
     name: ${{ matrix.config.name }}
 
-    runs-on: ${{ matrix.config.runner }}
+    runs-on:
+      group: ${{ matrix.config.runner }}
 
     container:
       image: ${{ matrix.config.image }}
@@ -180,7 +181,8 @@ jobs:
         config:
           - name: Hub tests for models, schedulers, and pipelines
             framework: hub_tests_pytorch
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner:
+              group: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_hub
 

.github/workflows/push_tests.yml

@@ -19,7 +19,8 @@ env:
 jobs:
   setup_torch_cuda_pipeline_matrix:
     name: Setup Torch Pipelines CUDA Slow Tests Matrix
-    runs-on: [ self-hosted, intel-cpu, 8-cpu, ci ]
+    runs-on:
+      group: aws-general-8-plus
     container:
       image: diffusers/diffusers-pytorch-cpu
     outputs:
@@ -57,7 +58,8 @@ jobs:
       max-parallel: 8
       matrix:
         module: ${{ fromJson(needs.setup_torch_cuda_pipeline_matrix.outputs.pipeline_test_matrix) }}
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
       options: --shm-size "16gb" --ipc host --gpus 0
@@ -101,7 +103,8 @@ jobs:
 
   torch_cuda_tests:
     name: Torch CUDA Tests
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
       options: --shm-size "16gb" --ipc host --gpus 0
@@ -201,7 +204,8 @@ jobs:
 
   onnx_cuda_tests:
     name: ONNX CUDA Tests
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-onnxruntime-cuda
       options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/ --gpus 0
@@ -249,7 +253,8 @@ jobs:
   run_torch_compile_tests:
     name: PyTorch Compile CUDA tests
 
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
 
     container:
       image: diffusers/diffusers-pytorch-compile-cuda
@@ -291,7 +296,8 @@ jobs:
   run_xformers_tests:
     name: PyTorch xformers CUDA tests
 
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
 
     container:
       image: diffusers/diffusers-pytorch-xformers-cuda
@@ -332,7 +338,8 @@ jobs:
   run_examples_tests:
     name: Examples PyTorch CUDA tests on Ubuntu
 
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
 
     container:
       image: diffusers/diffusers-pytorch-cuda

.github/workflows/push_tests_fast.yml

@@ -29,28 +29,29 @@ jobs:
         config:
           - name: Fast PyTorch CPU tests on Ubuntu
             framework: pytorch
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_cpu
           - name: Fast Flax CPU tests on Ubuntu
             framework: flax
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-flax-cpu
             report: flax_cpu
           - name: Fast ONNXRuntime CPU tests on Ubuntu
             framework: onnxruntime
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-onnxruntime-cpu
             report: onnx_cpu
           - name: PyTorch Example CPU tests on Ubuntu
             framework: pytorch_examples
-            runner: [ self-hosted, intel-cpu, 8-cpu, ci ]
+            runner: aws-general-8-plus
             image: diffusers/diffusers-pytorch-cpu
             report: torch_example_cpu
 
     name: ${{ matrix.config.name }}
 
-    runs-on: ${{ matrix.config.runner }}
+    runs-on:
+      group: ${{ matrix.config.runner }}
 
     container:
       image: ${{ matrix.config.image }}

.github/workflows/run_tests_from_a_pr.yml

@@ -26,7 +26,8 @@ env:
 jobs:
   run_tests:
     name: "Run a test on our runner from a PR"
-    runs-on: [single-gpu, nvidia-gpu, t4, ci]
+    runs-on:
+      group: aws-g4dn-2xlarge
     container:
       image: ${{ github.event.inputs.docker_image }}
       options: --gpus 0 --privileged --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
@@ -70,4 +71,4 @@ jobs:
         env:
             PY_TEST: ${{ github.event.inputs.test }}
         run: |
-          pytest "$PY_TEST"
+          pytest "$PY_TEST"

.github/workflows/ssh-pr-runner.yml

@@ -19,7 +19,8 @@ env:
 jobs:
   ssh_runner:
     name: "SSH"
-    runs-on: [self-hosted, intel-cpu, 32-cpu, 256-ram, ci]
+    runs-on:
+      group: aws-highmemory-32-plus
     container:
       image: ${{ github.event.inputs.docker_image }}
       options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface/diffusers:/mnt/cache/ --privileged

.github/workflows/ssh-runner.yml

@@ -22,7 +22,8 @@ env:
 jobs:
   ssh_runner:
     name: "SSH"
-    runs-on: [single-gpu, nvidia-gpu, "${{ github.event.inputs.runner_type }}", ci]
+    runs-on:
+      group: "${{ github.event.inputs.runner_type }}"
     container:
       image: ${{ github.event.inputs.docker_image }}
       options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface/diffusers:/mnt/cache/ --gpus 0 --privileged

docs/source/en/_toctree.yml

@@ -190,6 +190,10 @@
   - local: conceptual/evaluation
     title: Evaluating Diffusion Models
   title: Conceptual Guides
+- sections:
+  - local: community_projects
+    title: Projects built with Diffusers
+  title: Community Projects
 - sections:
   - isExpanded: false
     sections:

docs/source/en/api/pipelines/latte.md

@@ -24,6 +24,8 @@ The abstract from the paper is:
 
 **Highlights**: Latte is a latent diffusion transformer proposed as a backbone for modeling different modalities (trained for text-to-video generation here). It achieves state-of-the-art performance across four standard video benchmarks - [FaceForensics](https://arxiv.org/abs/1803.09179), [SkyTimelapse](https://arxiv.org/abs/1709.07592), [UCF101](https://arxiv.org/abs/1212.0402) and [Taichi-HD](https://arxiv.org/abs/2003.00196). To prepare and download the datasets for evaluation, please refer to [this https URL](https://github.com/Vchitect/Latte/blob/main/docs/datasets_evaluation.md).
 
+This pipeline was contributed by [maxin-cn](https://github.com/maxin-cn). The original codebase can be found [here](https://github.com/Vchitect/Latte). The original weights can be found under [hf.co/maxin-cn](https://huggingface.co/maxin-cn).
+
 <Tip>
 
 Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.md) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading.md#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.

docs/source/en/api/pipelines/lumina.md

@@ -43,6 +43,8 @@ Lumina-T2X has the following components:
 * It uses a Flow-based Large Diffusion Transformer as the backbone
 * It supports different any modalities with one backbone and corresponding encoder, decoder.
 
+This pipeline was contributed by [PommesPeter](https://github.com/PommesPeter). The original codebase can be found [here](https://github.com/Alpha-VLLM/Lumina-T2X). The original weights can be found under [hf.co/Alpha-VLLM](https://huggingface.co/Alpha-VLLM).
+
 <Tip>
 
 Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.md) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading.md#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.

docs/source/en/community_projects.md

@@ -0,0 +1,78 @@
+<!--Copyright 2024 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.
+-->
+
+# Community Projects
+
+Welcome to Community Projects. This space is dedicated to showcasing the incredible work and innovative applications created by our vibrant community using the `diffusers` library.
+
+This section aims to:
+
+- Highlight diverse and inspiring projects built with `diffusers`
+- Foster knowledge sharing within our community
+- Provide real-world examples of how `diffusers` can be leveraged
+
+Happy exploring, and thank you for being part of the Diffusers community!
+
+<table>
+    <tr>
+        <th>Project Name</th>
+        <th>Description</th>
+    </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/carson-katri/dream-textures"> dream-textures </a></td>
+    <td>Stable Diffusion built-in to Blender</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/megvii-research/HiDiffusion"> HiDiffusion </a></td>
+    <td>Increases the resolution and speed of your diffusion model by only adding a single line of code</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/lllyasviel/IC-Light"> IC-Light </a></td>
+    <td>IC-Light is a project to manipulate the illumination of images</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/InstantID/InstantID"> InstantID </a></td>
+    <td>InstantID : Zero-shot Identity-Preserving Generation in Seconds</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/Sanster/IOPaint"> IOPaint </a></td>
+    <td>Image inpainting tool powered by SOTA AI Model. Remove any unwanted object, defect, people from your pictures or erase and replace(powered by stable diffusion) any thing on your pictures.</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/bmaltais/kohya_ss"> Kohya </a></td>
+    <td>Gradio GUI for Kohya's Stable Diffusion trainers</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/magic-research/magic-animate"> MagicAnimate </a></td>
+    <td>MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/levihsu/OOTDiffusion"> OOTDiffusion </a></td>
+    <td>Outfitting Fusion based Latent Diffusion for Controllable Virtual Try-on</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/vladmandic/automatic"> SD.Next </a></td>
+    <td>SD.Next: Advanced Implementation of Stable Diffusion and other Diffusion-based generative image models</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/ashawkey/stable-dreamfusion"> stable-dreamfusion </a></td>
+    <td>Text-to-3D & Image-to-3D & Mesh Exportation with NeRF + Diffusion</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/HVision-NKU/StoryDiffusion"> StoryDiffusion </a></td>
+    <td>StoryDiffusion can create a magic story by generating consistent images and videos.</td>
+  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/cumulo-autumn/StreamDiffusion"> StreamDiffusion </a></td>
+    <td>A Pipeline-Level Solution for Real-Time Interactive Generation</td>
+  </tr>
+</table>

examples/community/lpw_stable_diffusion.py

@@ -13,13 +13,17 @@ from diffusers.configuration_utils import FrozenDict
 from diffusers.image_processor import VaeImageProcessor
 from diffusers.loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from diffusers.models import AutoencoderKL, UNet2DConditionModel
+from diffusers.models.lora import adjust_lora_scale_text_encoder
 from diffusers.pipelines.pipeline_utils import StableDiffusionMixin
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput, StableDiffusionSafetyChecker
 from diffusers.schedulers import KarrasDiffusionSchedulers
 from diffusers.utils import (
     PIL_INTERPOLATION,
+    USE_PEFT_BACKEND,
     deprecate,
     logging,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from diffusers.utils.torch_utils import randn_tensor
 
@@ -199,6 +203,7 @@ def get_unweighted_text_embeddings(
     text_input: torch.Tensor,
     chunk_length: int,
     no_boseos_middle: Optional[bool] = True,
+    clip_skip: Optional[int] = None,
 ):
     """
     When the length of tokens is a multiple of the capacity of the text encoder,
@@ -214,7 +219,20 @@ def get_unweighted_text_embeddings(
             # cover the head and the tail by the starting and the ending tokens
             text_input_chunk[:, 0] = text_input[0, 0]
             text_input_chunk[:, -1] = text_input[0, -1]
-            text_embedding = pipe.text_encoder(text_input_chunk)[0]
+            if clip_skip is None:
+                prompt_embeds = pipe.text_encoder(text_input_chunk.to(pipe.device))
+                text_embedding = prompt_embeds[0]
+            else:
+                prompt_embeds = pipe.text_encoder(text_input_chunk.to(pipe.device), output_hidden_states=True)
+                # Access the `hidden_states` first, that contains a tuple of
+                # all the hidden states from the encoder layers. Then index into
+                # the tuple to access the hidden states from the desired layer.
+                prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
+                # We also need to apply the final LayerNorm here to not mess with the
+                # representations. The `last_hidden_states` that we typically use for
+                # obtaining the final prompt representations passes through the LayerNorm
+                # layer.
+                text_embedding = pipe.text_encoder.text_model.final_layer_norm(prompt_embeds)
 
             if no_boseos_middle:
                 if i == 0:
@@ -230,7 +248,10 @@ def get_unweighted_text_embeddings(
             text_embeddings.append(text_embedding)
         text_embeddings = torch.concat(text_embeddings, axis=1)
     else:
-        text_embeddings = pipe.text_encoder(text_input)[0]
+        if clip_skip is None:
+            clip_skip = 0
+        prompt_embeds = pipe.text_encoder(text_input, output_hidden_states=True)[-1][-(clip_skip + 1)]
+        text_embeddings = pipe.text_encoder.text_model.final_layer_norm(prompt_embeds)
     return text_embeddings
 
 
@@ -242,6 +263,8 @@ def get_weighted_text_embeddings(
     no_boseos_middle: Optional[bool] = False,
     skip_parsing: Optional[bool] = False,
     skip_weighting: Optional[bool] = False,
+    clip_skip=None,
+    lora_scale=None,
 ):
     r"""
     Prompts can be assigned with local weights using brackets. For example,
@@ -268,6 +291,16 @@ def get_weighted_text_embeddings(
         skip_weighting (`bool`, *optional*, defaults to `False`):
             Skip the weighting. When the parsing is skipped, it is forced True.
     """
+    # set lora scale so that monkey patched LoRA
+    # function of text encoder can correctly access it
+    if lora_scale is not None and isinstance(pipe, StableDiffusionLoraLoaderMixin):
+        pipe._lora_scale = lora_scale
+
+        # dynamically adjust the LoRA scale
+        if not USE_PEFT_BACKEND:
+            adjust_lora_scale_text_encoder(pipe.text_encoder, lora_scale)
+        else:
+            scale_lora_layers(pipe.text_encoder, lora_scale)
     max_length = (pipe.tokenizer.model_max_length - 2) * max_embeddings_multiples + 2
     if isinstance(prompt, str):
         prompt = [prompt]
@@ -334,10 +367,7 @@ def get_weighted_text_embeddings(
 
     # get the embeddings
     text_embeddings = get_unweighted_text_embeddings(
-        pipe,
-        prompt_tokens,
-        pipe.tokenizer.model_max_length,
-        no_boseos_middle=no_boseos_middle,
+        pipe, prompt_tokens, pipe.tokenizer.model_max_length, no_boseos_middle=no_boseos_middle, clip_skip=clip_skip
     )
     prompt_weights = torch.tensor(prompt_weights, dtype=text_embeddings.dtype, device=text_embeddings.device)
     if uncond_prompt is not None:
@@ -346,6 +376,7 @@ def get_weighted_text_embeddings(
             uncond_tokens,
             pipe.tokenizer.model_max_length,
             no_boseos_middle=no_boseos_middle,
+            clip_skip=clip_skip,
         )
         uncond_weights = torch.tensor(uncond_weights, dtype=uncond_embeddings.dtype, device=uncond_embeddings.device)
 
@@ -362,6 +393,11 @@ def get_weighted_text_embeddings(
             current_mean = uncond_embeddings.float().mean(axis=[-2, -1]).to(uncond_embeddings.dtype)
             uncond_embeddings *= (previous_mean / current_mean).unsqueeze(-1).unsqueeze(-1)
 
+    if pipe.text_encoder is not None:
+        if isinstance(pipe, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
+            # Retrieve the original scale by scaling back the LoRA layers
+            unscale_lora_layers(pipe.text_encoder, lora_scale)
+
     if uncond_prompt is not None:
         return text_embeddings, uncond_embeddings
     return text_embeddings, None
@@ -549,6 +585,8 @@ class StableDiffusionLongPromptWeightingPipeline(
         max_embeddings_multiples=3,
         prompt_embeds: Optional[torch.Tensor] = None,
         negative_prompt_embeds: Optional[torch.Tensor] = None,
+        clip_skip: Optional[int] = None,
+        lora_scale: Optional[float] = None,
     ):
         r"""
         Encodes the prompt into text encoder hidden states.
@@ -597,6 +635,8 @@ class StableDiffusionLongPromptWeightingPipeline(
                 prompt=prompt,
                 uncond_prompt=negative_prompt if do_classifier_free_guidance else None,
                 max_embeddings_multiples=max_embeddings_multiples,
+                clip_skip=clip_skip,
+                lora_scale=lora_scale,
             )
             if prompt_embeds is None:
                 prompt_embeds = prompt_embeds1
@@ -790,6 +830,7 @@ class StableDiffusionLongPromptWeightingPipeline(
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         is_cancelled_callback: Optional[Callable[[], bool]] = None,
+        clip_skip: Optional[int] = None,
         callback_steps: int = 1,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
@@ -865,6 +906,9 @@ class StableDiffusionLongPromptWeightingPipeline(
             is_cancelled_callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. If the function returns
                 `True`, the inference will be cancelled.
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
             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.
@@ -903,6 +947,7 @@ class StableDiffusionLongPromptWeightingPipeline(
         # 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
+        lora_scale = cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
 
         # 3. Encode input prompt
         prompt_embeds = self._encode_prompt(
@@ -914,6 +959,8 @@ class StableDiffusionLongPromptWeightingPipeline(
             max_embeddings_multiples,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
+            clip_skip=clip_skip,
+            lora_scale=lora_scale,
         )
         dtype = prompt_embeds.dtype
 
@@ -1044,6 +1091,7 @@ class StableDiffusionLongPromptWeightingPipeline(
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
         is_cancelled_callback: Optional[Callable[[], bool]] = None,
+        clip_skip=None,
         callback_steps: int = 1,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
@@ -1101,6 +1149,9 @@ class StableDiffusionLongPromptWeightingPipeline(
             is_cancelled_callback (`Callable`, *optional*):
                 A function that will be called every `callback_steps` steps during inference. If the function returns
                 `True`, the inference will be cancelled.
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
             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.
@@ -1135,6 +1186,7 @@ class StableDiffusionLongPromptWeightingPipeline(
             return_dict=return_dict,
             callback=callback,
             is_cancelled_callback=is_cancelled_callback,
+            clip_skip=clip_skip,
             callback_steps=callback_steps,
             cross_attention_kwargs=cross_attention_kwargs,
         )

examples/community/lpw_stable_diffusion_xl.py

@@ -25,21 +25,25 @@ from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
 from diffusers.loaders import (
     FromSingleFileMixin,
     IPAdapterMixin,
-    StableDiffusionLoraLoaderMixin,
+    StableDiffusionXLLoraLoaderMixin,
     TextualInversionLoaderMixin,
 )
 from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from diffusers.models.attention_processor import AttnProcessor2_0, XFormersAttnProcessor
+from diffusers.models.lora import adjust_lora_scale_text_encoder
 from diffusers.pipelines.pipeline_utils import StableDiffusionMixin
 from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
 from diffusers.schedulers import KarrasDiffusionSchedulers
 from diffusers.utils import (
+    USE_PEFT_BACKEND,
     deprecate,
     is_accelerate_available,
     is_accelerate_version,
     is_invisible_watermark_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from diffusers.utils.torch_utils import randn_tensor
 
@@ -261,6 +265,7 @@ def get_weighted_text_embeddings_sdxl(
     num_images_per_prompt: int = 1,
     device: Optional[torch.device] = None,
     clip_skip: Optional[int] = None,
+    lora_scale: Optional[int] = None,
 ):
     """
     This function can process long prompt with weights, no length limitation
@@ -281,6 +286,24 @@ def get_weighted_text_embeddings_sdxl(
     """
     device = device or pipe._execution_device
 
+    # set lora scale so that monkey patched LoRA
+    # function of text encoder can correctly access it
+    if lora_scale is not None and isinstance(pipe, StableDiffusionXLLoraLoaderMixin):
+        pipe._lora_scale = lora_scale
+
+        # dynamically adjust the LoRA scale
+        if pipe.text_encoder is not None:
+            if not USE_PEFT_BACKEND:
+                adjust_lora_scale_text_encoder(pipe.text_encoder, lora_scale)
+            else:
+                scale_lora_layers(pipe.text_encoder, lora_scale)
+
+        if pipe.text_encoder_2 is not None:
+            if not USE_PEFT_BACKEND:
+                adjust_lora_scale_text_encoder(pipe.text_encoder_2, lora_scale)
+            else:
+                scale_lora_layers(pipe.text_encoder_2, lora_scale)
+
     if prompt_2:
         prompt = f"{prompt} {prompt_2}"
 
@@ -429,6 +452,16 @@ def get_weighted_text_embeddings_sdxl(
         bs_embed * num_images_per_prompt, -1
     )
 
+    if pipe.text_encoder is not None:
+        if isinstance(pipe, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+            # Retrieve the original scale by scaling back the LoRA layers
+            unscale_lora_layers(pipe.text_encoder, lora_scale)
+
+    if pipe.text_encoder_2 is not None:
+        if isinstance(pipe, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+            # Retrieve the original scale by scaling back the LoRA layers
+            unscale_lora_layers(pipe.text_encoder_2, lora_scale)
+
     return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
 
@@ -549,7 +582,7 @@ class SDXLLongPromptWeightingPipeline(
     StableDiffusionMixin,
     FromSingleFileMixin,
     IPAdapterMixin,
-    StableDiffusionLoraLoaderMixin,
+    StableDiffusionXLLoraLoaderMixin,
     TextualInversionLoaderMixin,
 ):
     r"""
@@ -561,8 +594,8 @@ class SDXLLongPromptWeightingPipeline(
     The pipeline also inherits the following loading methods:
         - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
-        - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
 
     Args:
@@ -743,7 +776,7 @@ class SDXLLongPromptWeightingPipeline(
 
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
             self._lora_scale = lora_scale
 
         if prompt is not None and isinstance(prompt, str):
@@ -1612,7 +1645,9 @@ class SDXLLongPromptWeightingPipeline(
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 
         # 3. Encode input prompt
-        (self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None)
+        lora_scale = (
+            self._cross_attention_kwargs.get("scale", None) if self._cross_attention_kwargs is not None else None
+        )
 
         negative_prompt = negative_prompt if negative_prompt is not None else ""
 
@@ -1627,6 +1662,7 @@ class SDXLLongPromptWeightingPipeline(
             neg_prompt=negative_prompt,
             num_images_per_prompt=num_images_per_prompt,
             clip_skip=clip_skip,
+            lora_scale=lora_scale,
         )
         dtype = prompt_embeds.dtype
 

src/diffusers/loaders/peft.py

@@ -30,6 +30,7 @@ from .unet_loader_utils import _maybe_expand_lora_scales
 
 _SET_ADAPTER_SCALE_FN_MAPPING = {
     "UNet2DConditionModel": _maybe_expand_lora_scales,
+    "UNetMotionModel": _maybe_expand_lora_scales,
     "SD3Transformer2DModel": lambda model_cls, weights: weights,
 }
 

src/diffusers/models/unets/unet_motion_model.py

@@ -19,7 +19,7 @@ import torch.nn.functional as F
 import torch.utils.checkpoint
 
 from ...configuration_utils import ConfigMixin, FrozenDict, register_to_config
-from ...loaders import FromOriginalModelMixin, UNet2DConditionLoadersMixin
+from ...loaders import FromOriginalModelMixin, PeftAdapterMixin, UNet2DConditionLoadersMixin
 from ...utils import logging
 from ..attention_processor import (
     ADDED_KV_ATTENTION_PROCESSORS,
@@ -231,7 +231,7 @@ class MotionAdapter(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         pass
 
 
-class UNetMotionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
+class UNetMotionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, PeftAdapterMixin):
     r"""
     A modified conditional 2D UNet model that takes a noisy sample, conditional state, and a timestep and returns a
     sample shaped output.