update

init
Add correct number of channels when resuming from checkpoint for Flux Control LoRa training (#10422 )
2025-01-02 15:35:58 +01:00 · 2025-01-02 13:11:44 +01:00 · 2025-01-02 15:51:44 +05:30 · 2024-12-31 12:44:57 -08:00 · 2024-12-31 08:52:11 -08:00 · 2024-12-31 08:37:00 -08:00
261 changed files with 35957 additions and 1462 deletions
@@ -238,12 +238,13 @@ jobs:

  run_flax_tpu_tests:
    name: Nightly Flax TPU Tests
-    runs-on: docker-tpu
+    runs-on:
+      group: gcp-ct5lp-hightpu-8t
    if: github.event_name == 'schedule'

    container:
      image: diffusers/diffusers-flax-tpu
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --privileged
+      options: --shm-size "16gb" --ipc host --privileged ${{ vars.V5_LITEPOD_8_ENV}} -v /mnt/hf_cache:/mnt/hf_cache
    defaults:
      run:
        shell: bash
@@ -356,6 +357,10 @@ jobs:
        config:
          - backend: "bitsandbytes"
            test_location: "bnb"
+          - backend: "gguf"
+            test_location: "gguf"
+          - backend: "torchao"
+            test_location: "torchao"
    runs-on:
      group: aws-g6e-xlarge-plus
    container:
@@ -519,4 +524,4 @@ jobs:
 #        if: always()
 #        run: |
 #          pip install slack_sdk tabulate
-#          python utils/log_reports.py >> $GITHUB_STEP_SUMMARY
+#          python utils/log_reports.py >> $GITHUB_STEP_SUMMARY
@@ -1,134 +0,0 @@
-name: Fast tests for PRs - PEFT backend
-
-on:
-  pull_request:
-    branches:
-      - main
-    paths:
-      - "src/diffusers/**.py"
-      - "tests/**.py"
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  DIFFUSERS_IS_CI: yes
-  OMP_NUM_THREADS: 4
-  MKL_NUM_THREADS: 4
-  PYTEST_TIMEOUT: 60
-
-jobs:
-  check_code_quality:
-    runs-on: ubuntu-22.04
-    steps:
-      - uses: actions/checkout@v3
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: "3.8"
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install .[quality]
-      - name: Check quality
-        run: make quality
-      - name: Check if failure
-        if: ${{ failure() }}
-        run: |
-          echo "Quality check failed. Please ensure the right dependency versions are installed with 'pip install -e .[quality]' and run 'make style && make quality'" >> $GITHUB_STEP_SUMMARY
-
-  check_repository_consistency:
-    needs: check_code_quality
-    runs-on: ubuntu-22.04
-    steps:
-      - uses: actions/checkout@v3
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: "3.8"
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install .[quality]
-      - name: Check repo consistency
-        run: |
-          python utils/check_copies.py
-          python utils/check_dummies.py
-          make deps_table_check_updated
-      - name: Check if failure
-        if: ${{ failure() }}
-        run: |
-          echo "Repo consistency check failed. Please ensure the right dependency versions are installed with 'pip install -e .[quality]' and run 'make fix-copies'" >> $GITHUB_STEP_SUMMARY
-
-  run_fast_tests:
-    needs: [check_code_quality, check_repository_consistency]
-    strategy:
-      fail-fast: false
-      matrix:
-        lib-versions: ["main", "latest"]
-
-
-    name: LoRA - ${{ matrix.lib-versions }}
-
-    runs-on:
-      group: aws-general-8-plus
-
-    container:
-      image: diffusers/diffusers-pytorch-cpu
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
-
-    defaults:
-      run:
-        shell: bash
-
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: Install dependencies
-      run: |
-        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
-        python -m uv pip install -e [quality,test]
-        # TODO (sayakpaul, DN6): revisit `--no-deps`
-        if [ "${{ matrix.lib-versions }}" == "main" ]; then
-            python -m pip install -U peft@git+https://github.com/huggingface/peft.git --no-deps
-            python -m uv pip install -U transformers@git+https://github.com/huggingface/transformers.git --no-deps
-            pip uninstall accelerate -y && python -m uv pip install -U accelerate@git+https://github.com/huggingface/accelerate.git --no-deps
-        else
-            python -m uv pip install -U peft --no-deps
-            python -m uv pip install -U transformers accelerate --no-deps
-        fi
-
-    - name: Environment
-      run: |
-        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
-        python utils/print_env.py
-
-    - name: Run fast PyTorch LoRA CPU tests with PEFT backend
-      run: |
-        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
-        python -m pytest -n 4 --max-worker-restart=0 --dist=loadfile \
-          -s -v \
-          --make-reports=tests_${{ matrix.lib-versions }} \
-          tests/lora/
-        python -m pytest -n 4 --max-worker-restart=0 --dist=loadfile \
-          -s -v \
-          --make-reports=tests_models_lora_${{ matrix.lib-versions }} \
-          tests/models/ -k "lora"
-
-
-    - name: Failure short reports
-      if: ${{ failure() }}
-      run: |
-        cat reports/tests_${{ matrix.lib-versions }}_failures_short.txt
-        cat reports/tests_models_lora_${{ matrix.lib-versions }}_failures_short.txt
-
-    - name: Test suite reports artifacts
-      if: ${{ always() }}
-      uses: actions/upload-artifact@v4
-      with:
-        name: pr_${{ matrix.lib-versions }}_test_reports
-        path: reports
@@ -234,3 +234,67 @@ jobs:
      with:
        name: pr_${{ matrix.config.report }}_test_reports
        path: reports
+
+  run_lora_tests:
+    needs: [check_code_quality, check_repository_consistency]
+    strategy:
+      fail-fast: false
+
+    name: LoRA tests with PEFT main
+
+    runs-on:
+      group: aws-general-8-plus
+
+    container:
+      image: diffusers/diffusers-pytorch-cpu
+      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
+
+    defaults:
+      run:
+        shell: bash
+
+    steps:
+    - name: Checkout diffusers
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 2
+
+    - name: Install dependencies
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python -m uv pip install -e [quality,test]
+        # TODO (sayakpaul, DN6): revisit `--no-deps`
+        python -m pip install -U peft@git+https://github.com/huggingface/peft.git --no-deps
+        python -m uv pip install -U transformers@git+https://github.com/huggingface/transformers.git --no-deps
+        pip uninstall accelerate -y && python -m uv pip install -U accelerate@git+https://github.com/huggingface/accelerate.git --no-deps
+
+    - name: Environment
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python utils/print_env.py
+
+    - name: Run fast PyTorch LoRA tests with PEFT
+      run: |
+        python -m venv /opt/venv && export PATH="/opt/venv/bin:$PATH"
+        python -m pytest -n 4 --max-worker-restart=0 --dist=loadfile \
+          -s -v \
+          --make-reports=tests_peft_main \
+          tests/lora/
+        python -m pytest -n 4 --max-worker-restart=0 --dist=loadfile \
+          -s -v \
+          --make-reports=tests_models_lora_peft_main \
+          tests/models/ -k "lora"
+
+    - name: Failure short reports
+      if: ${{ failure() }}
+      run: |
+        cat reports/tests_lora_failures_short.txt
+        cat reports/tests_models_lora_failures_short.txt
+
+    - name: Test suite reports artifacts
+      if: ${{ always() }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: pr_main_test_reports
+        path: reports
+
@@ -161,10 +161,11 @@ jobs:

  flax_tpu_tests:
    name: Flax TPU Tests
-    runs-on: docker-tpu
+    runs-on:
+      group: gcp-ct5lp-hightpu-8t
    container:
      image: diffusers/diffusers-flax-tpu
-      options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/ --privileged
+      options: --shm-size "16gb" --ipc host --privileged ${{ vars.V5_LITEPOD_8_ENV}} -v /mnt/hf_cache:/mnt/hf_cache 
    defaults:
      run:
        shell: bash
@@ -46,7 +46,7 @@ jobs:
      shell: arch -arch arm64 bash {0}
      run: |
        ${CONDA_RUN} python -m pip install --upgrade pip uv
-        ${CONDA_RUN} python -m uv pip install -e [quality,test]
+        ${CONDA_RUN} python -m uv pip install -e ".[quality,test]"
        ${CONDA_RUN} python -m uv pip install torch torchvision torchaudio
        ${CONDA_RUN} python -m uv pip install accelerate@git+https://github.com/huggingface/accelerate.git
        ${CONDA_RUN} python -m uv pip install transformers --upgrade
@@ -68,7 +68,7 @@ jobs:
      - name: Test installing diffusers and importing
        run: |
          pip install diffusers && pip uninstall diffusers -y
-          pip install -i https://testpypi.python.org/pypi diffusers
+          pip install -i https://test.pypi.org/simple/ diffusers
          python -c "from diffusers import __version__; print(__version__)"
          python -c "from diffusers import DiffusionPipeline; pipe = DiffusionPipeline.from_pretrained('fusing/unet-ldm-dummy-update'); pipe()"
          python -c "from diffusers import DiffusionPipeline; pipe = DiffusionPipeline.from_pretrained('hf-internal-testing/tiny-stable-diffusion-pipe', safety_checker=None); pipe('ah suh du')"
@@ -48,7 +48,7 @@
  - local: using-diffusers/inpaint
    title: Inpainting
  - local: using-diffusers/text-img2vid
-    title: Text or image-to-video
+    title: Video generation
  - local: using-diffusers/depth2img
    title: Depth-to-image
  title: Generative tasks
@@ -157,6 +157,10 @@
    title: Getting Started
  - local: quantization/bitsandbytes
    title: bitsandbytes
+  - local: quantization/gguf
+    title: gguf
+  - local: quantization/torchao
+    title: torchao
  title: Quantization Methods
 - sections:
  - local: optimization/fp16
@@ -234,6 +238,8 @@
      title: Textual Inversion
    - local: api/loaders/unet
      title: UNet
+    - local: api/loaders/transformer_sd3
+      title: SD3Transformer2D
    - local: api/loaders/peft
      title: PEFT
    title: Loaders
@@ -252,6 +258,8 @@
        title: SD3ControlNetModel
      - local: api/models/controlnet_sparsectrl
        title: SparseControlNetModel
+      - local: api/models/controlnet_union
+        title: ControlNetUnionModel
      title: ControlNets
    - sections:
      - local: api/models/allegro_transformer3d
@@ -268,10 +276,14 @@
        title: FluxTransformer2DModel
      - local: api/models/hunyuan_transformer2d
        title: HunyuanDiT2DModel
+      - local: api/models/hunyuan_video_transformer_3d
+        title: HunyuanVideoTransformer3DModel
      - local: api/models/latte_transformer3d
        title: LatteTransformer3DModel
      - local: api/models/lumina_nextdit2d
        title: LuminaNextDiT2DModel
+      - local: api/models/ltx_video_transformer3d
+        title: LTXVideoTransformer3DModel
      - local: api/models/mochi_transformer3d
        title: MochiTransformer3DModel
      - local: api/models/pixart_transformer2d
@@ -280,6 +292,8 @@
        title: PriorTransformer
      - local: api/models/sd3_transformer2d
        title: SD3Transformer2DModel
+      - local: api/models/sana_transformer2d
+        title: SanaTransformer2DModel
      - local: api/models/stable_audio_transformer
        title: StableAudioDiTModel
      - local: api/models/transformer2d
@@ -310,6 +324,10 @@
        title: AutoencoderKLAllegro
      - local: api/models/autoencoderkl_cogvideox
        title: AutoencoderKLCogVideoX
+      - local: api/models/autoencoder_kl_hunyuan_video
+        title: AutoencoderKLHunyuanVideo
+      - local: api/models/autoencoderkl_ltx_video
+        title: AutoencoderKLLTXVideo
      - local: api/models/autoencoderkl_mochi
        title: AutoencoderKLMochi
      - local: api/models/asymmetricautoencoderkl
@@ -368,6 +386,8 @@
      title: ControlNet-XS
    - local: api/pipelines/controlnetxs_sdxl
      title: ControlNet-XS with Stable Diffusion XL
+    - local: api/pipelines/controlnet_union
+      title: ControlNetUnion
    - local: api/pipelines/dance_diffusion
      title: Dance Diffusion
    - local: api/pipelines/ddim
@@ -382,8 +402,12 @@
      title: DiT
    - local: api/pipelines/flux
      title: Flux
+    - local: api/pipelines/control_flux_inpaint
+      title: FluxControlInpaint
    - local: api/pipelines/hunyuandit
      title: Hunyuan-DiT
+    - local: api/pipelines/hunyuan_video
+      title: HunyuanVideo
    - local: api/pipelines/i2vgenxl
      title: I2VGen-XL
    - local: api/pipelines/pix2pix
@@ -404,6 +428,8 @@
      title: Latte
    - local: api/pipelines/ledits_pp
      title: LEDITS++
+    - local: api/pipelines/ltx_video
+      title: LTXVideo
    - local: api/pipelines/lumina
      title: Lumina-T2X
    - local: api/pipelines/marigold
@@ -424,6 +450,8 @@
      title: PixArt-α
    - local: api/pipelines/pixart_sigma
      title: PixArt-Σ
+    - local: api/pipelines/sana
+      title: Sana
    - local: api/pipelines/self_attention_guidance
      title: Self-Attention Guidance
    - local: api/pipelines/semantic_stable_diffusion
@@ -15,40 +15,135 @@ specific language governing permissions and limitations under the License.
 An attention processor is a class for applying different types of attention mechanisms.

 ## AttnProcessor
+
 [[autodoc]] models.attention_processor.AttnProcessor

-## AttnProcessor2_0
 [[autodoc]] models.attention_processor.AttnProcessor2_0

-## AttnAddedKVProcessor
 [[autodoc]] models.attention_processor.AttnAddedKVProcessor

-## AttnAddedKVProcessor2_0
 [[autodoc]] models.attention_processor.AttnAddedKVProcessor2_0

-## CrossFrameAttnProcessor
-[[autodoc]] pipelines.text_to_video_synthesis.pipeline_text_to_video_zero.CrossFrameAttnProcessor
+[[autodoc]] models.attention_processor.AttnProcessorNPU

-## CustomDiffusionAttnProcessor
-[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor
-
-## CustomDiffusionAttnProcessor2_0
-[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor2_0
-
-## CustomDiffusionXFormersAttnProcessor
-[[autodoc]] models.attention_processor.CustomDiffusionXFormersAttnProcessor
-
-## FusedAttnProcessor2_0
 [[autodoc]] models.attention_processor.FusedAttnProcessor2_0

+## Allegro
+
+[[autodoc]] models.attention_processor.AllegroAttnProcessor2_0
+
+## AuraFlow
+
+[[autodoc]] models.attention_processor.AuraFlowAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FusedAuraFlowAttnProcessor2_0
+
+## CogVideoX
+
+[[autodoc]] models.attention_processor.CogVideoXAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FusedCogVideoXAttnProcessor2_0
+
+## CrossFrameAttnProcessor
+
+[[autodoc]] pipelines.text_to_video_synthesis.pipeline_text_to_video_zero.CrossFrameAttnProcessor
+
+## Custom Diffusion
+
+[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor
+
+[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.CustomDiffusionXFormersAttnProcessor
+
+## Flux
+
+[[autodoc]] models.attention_processor.FluxAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FusedFluxAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FluxSingleAttnProcessor2_0
+
+## Hunyuan
+
+[[autodoc]] models.attention_processor.HunyuanAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FusedHunyuanAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGHunyuanAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGCFGHunyuanAttnProcessor2_0
+
+## IdentitySelfAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGIdentitySelfAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGCFGIdentitySelfAttnProcessor2_0
+
+## IP-Adapter
+
+[[autodoc]] models.attention_processor.IPAdapterAttnProcessor
+
+[[autodoc]] models.attention_processor.IPAdapterAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.SD3IPAdapterJointAttnProcessor2_0
+
+## JointAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.JointAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGJointAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGCFGJointAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.FusedJointAttnProcessor2_0
+
+## LoRA
+
+[[autodoc]] models.attention_processor.LoRAAttnProcessor
+
+[[autodoc]] models.attention_processor.LoRAAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.LoRAAttnAddedKVProcessor
+
+[[autodoc]] models.attention_processor.LoRAXFormersAttnProcessor
+
+## Lumina-T2X
+
+[[autodoc]] models.attention_processor.LuminaAttnProcessor2_0
+
+## Mochi
+
+[[autodoc]] models.attention_processor.MochiAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.MochiVaeAttnProcessor2_0
+
+## Sana
+
+[[autodoc]] models.attention_processor.SanaLinearAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.SanaMultiscaleAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGCFGSanaLinearAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.PAGIdentitySanaLinearAttnProcessor2_0
+
+## Stable Audio
+
+[[autodoc]] models.attention_processor.StableAudioAttnProcessor2_0
+
 ## SlicedAttnProcessor
+
 [[autodoc]] models.attention_processor.SlicedAttnProcessor

-## SlicedAttnAddedKVProcessor
 [[autodoc]] models.attention_processor.SlicedAttnAddedKVProcessor

 ## XFormersAttnProcessor
+
 [[autodoc]] models.attention_processor.XFormersAttnProcessor

-## AttnProcessorNPU
-[[autodoc]] models.attention_processor.AttnProcessorNPU
+[[autodoc]] models.attention_processor.XFormersAttnAddedKVProcessor
+
+## XLAFlashAttnProcessor2_0
+
+[[autodoc]] models.attention_processor.XLAFlashAttnProcessor2_0
@@ -24,6 +24,12 @@ Learn how to load an IP-Adapter checkpoint and image in the IP-Adapter [loading]

 [[autodoc]] loaders.ip_adapter.IPAdapterMixin

+## SD3IPAdapterMixin
+
+[[autodoc]] loaders.ip_adapter.SD3IPAdapterMixin
+    - all
+    - is_ip_adapter_active
+
 ## IPAdapterMaskProcessor

 [[autodoc]] image_processor.IPAdapterMaskProcessor
@@ -17,6 +17,9 @@ LoRA is a fast and lightweight training method that inserts and trains a signifi
 - [`StableDiffusionLoraLoaderMixin`] provides functions for loading and unloading, fusing and unfusing, enabling and disabling, and more functions for managing LoRA weights. This class can be used with any model.
 - [`StableDiffusionXLLoraLoaderMixin`] is a [Stable Diffusion (SDXL)](../../api/pipelines/stable_diffusion/stable_diffusion_xl) version of the [`StableDiffusionLoraLoaderMixin`] class for loading and saving LoRA weights. It can only be used with the SDXL model.
 - [`SD3LoraLoaderMixin`] provides similar functions for [Stable Diffusion 3](https://huggingface.co/blog/sd3).
+- [`FluxLoraLoaderMixin`] provides similar functions for [Flux](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux).
+- [`CogVideoXLoraLoaderMixin`] provides similar functions for [CogVideoX](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cogvideox).
+- [`Mochi1LoraLoaderMixin`] provides similar functions for [Mochi](https://huggingface.co/docs/diffusers/main/en/api/pipelines/mochi).
 - [`AmusedLoraLoaderMixin`] is for the [`AmusedPipeline`].
 - [`LoraBaseMixin`] provides a base class with several utility methods to fuse, unfuse, unload, LoRAs and more.

@@ -38,6 +41,18 @@ To learn more about how to load LoRA weights, see the [LoRA](../../using-diffuse

 [[autodoc]] loaders.lora_pipeline.SD3LoraLoaderMixin

+## FluxLoraLoaderMixin
+
+[[autodoc]] loaders.lora_pipeline.FluxLoraLoaderMixin
+
+## CogVideoXLoraLoaderMixin
+
+[[autodoc]] loaders.lora_pipeline.CogVideoXLoraLoaderMixin
+
+## Mochi1LoraLoaderMixin
+
+[[autodoc]] loaders.lora_pipeline.Mochi1LoraLoaderMixin
+
 ## AmusedLoraLoaderMixin

 [[autodoc]] loaders.lora_pipeline.AmusedLoraLoaderMixin
@@ -0,0 +1,29 @@
+<!--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.
+-->
+
+# SD3Transformer2D
+
+This class is useful when *only* loading weights into a [`SD3Transformer2DModel`]. If you need to load weights into the text encoder or a text encoder and SD3Transformer2DModel, check [`SD3LoraLoaderMixin`](lora#diffusers.loaders.SD3LoraLoaderMixin) class instead.
+
+The [`SD3Transformer2DLoadersMixin`] class currently only loads IP-Adapter weights, but will be used in the future to save weights and load LoRAs.
+
+<Tip>
+
+To learn more about how to load LoRA weights, see the [LoRA](../../using-diffusers/loading_adapters#lora) loading guide.
+
+</Tip>
+
+## SD3Transformer2DLoadersMixin
+
+[[autodoc]] loaders.transformer_sd3.SD3Transformer2DLoadersMixin
+    - all
+    - _load_ip_adapter_weights
@@ -29,6 +29,8 @@ The following DCAE models are released and supported in Diffusers.
 | [`mit-han-lab/dc-ae-f128c512-in-1.0-diffusers`](https://huggingface.co/mit-han-lab/dc-ae-f128c512-in-1.0-diffusers) | [`mit-han-lab/dc-ae-f128c512-in-1.0`](https://huggingface.co/mit-han-lab/dc-ae-f128c512-in-1.0)
 | [`mit-han-lab/dc-ae-f128c512-mix-1.0-diffusers`](https://huggingface.co/mit-han-lab/dc-ae-f128c512-mix-1.0-diffusers) | [`mit-han-lab/dc-ae-f128c512-mix-1.0`](https://huggingface.co/mit-han-lab/dc-ae-f128c512-mix-1.0)

+This model was contributed by [lawrence-cj](https://github.com/lawrence-cj).
+
 Load a model in Diffusers format with [`~ModelMixin.from_pretrained`].

 ```python
@@ -37,6 +39,26 @@ from diffusers import AutoencoderDC
 ae = AutoencoderDC.from_pretrained("mit-han-lab/dc-ae-f32c32-sana-1.0-diffusers", torch_dtype=torch.float32).to("cuda")
 ```

+## Load a model in Diffusers via `from_single_file`
+
+```python
+from difusers import AutoencoderDC
+
+ckpt_path = "https://huggingface.co/mit-han-lab/dc-ae-f32c32-sana-1.0/blob/main/model.safetensors"
+model = AutoencoderDC.from_single_file(ckpt_path) 
+
+```
+
+The `AutoencoderDC` model has `in` and `mix` single file checkpoint variants that have matching checkpoint keys, but use different scaling factors. It is not possible for Diffusers to automatically infer the correct config file to use with the model based on just the checkpoint and will default to configuring the model using the `mix` variant config file. To override the automatically determined config, please use the `config` argument when using single file loading with `in` variant checkpoints. 
+
+```python
+from diffusers import AutoencoderDC
+
+ckpt_path = "https://huggingface.co/mit-han-lab/dc-ae-f128c512-in-1.0/blob/main/model.safetensors"
+model = AutoencoderDC.from_single_file(ckpt_path, config="mit-han-lab/dc-ae-f128c512-in-1.0-diffusers")
+```
+
+
 ## AutoencoderDC

 [[autodoc]] AutoencoderDC
@@ -0,0 +1,32 @@
+<!-- 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. -->
+
+# AutoencoderKLHunyuanVideo
+
+The 3D variational autoencoder (VAE) model with KL loss used in [HunyuanVideo](https://github.com/Tencent/HunyuanVideo/), which was introduced in [HunyuanVideo: A Systematic Framework For Large Video Generative Models](https://huggingface.co/papers/2412.03603) by Tencent.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import AutoencoderKLHunyuanVideo
+
+vae = AutoencoderKLHunyuanVideo.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder="vae", torch_dtype=torch.float16)
+```
+
+## AutoencoderKLHunyuanVideo
+
+[[autodoc]] AutoencoderKLHunyuanVideo
+  - decode
+  - all
+
+## DecoderOutput
+
+[[autodoc]] models.autoencoders.vae.DecoderOutput
@@ -0,0 +1,37 @@
+<!-- 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. -->
+
+# AutoencoderKLLTXVideo
+
+The 3D variational autoencoder (VAE) model with KL loss used in [LTX](https://huggingface.co/Lightricks/LTX-Video) was introduced by Lightricks.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import AutoencoderKLLTXVideo
+
+vae = AutoencoderKLLTXVideo.from_pretrained("Lightricks/LTX-Video", subfolder="vae", torch_dtype=torch.float32).to("cuda")
+```
+
+## AutoencoderKLLTXVideo
+
+[[autodoc]] AutoencoderKLLTXVideo
+    - decode
+    - encode
+    - all
+
+## AutoencoderKLOutput
+
+[[autodoc]] models.autoencoders.autoencoder_kl.AutoencoderKLOutput
+
+## DecoderOutput
+
+[[autodoc]] models.autoencoders.vae.DecoderOutput
@@ -0,0 +1,35 @@
+<!--Copyright 2024 The HuggingFace Team and The InstantX 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.
+-->
+
+# ControlNetUnionModel
+
+ControlNetUnionModel is an implementation of ControlNet for Stable Diffusion XL.
+
+The ControlNet model was introduced in [ControlNetPlus](https://github.com/xinsir6/ControlNetPlus) by xinsir6. It supports multiple conditioning inputs without increasing computation.
+
+*We design a new architecture that can support 10+ control types in condition text-to-image generation and can generate high resolution images visually comparable with midjourney. The network is based on the original ControlNet architecture, we propose two new modules to: 1 Extend the original ControlNet to support different image conditions using the same network parameter. 2 Support multiple conditions input without increasing computation offload, which is especially important for designers who want to edit image in detail, different conditions use the same condition encoder, without adding extra computations or parameters.*
+
+## Loading
+
+By default the [`ControlNetUnionModel`] should be loaded with [`~ModelMixin.from_pretrained`].
+
+```py
+from diffusers import StableDiffusionXLControlNetUnionPipeline, ControlNetUnionModel
+
+controlnet = ControlNetUnionModel.from_pretrained("xinsir/controlnet-union-sdxl-1.0")
+pipe = StableDiffusionXLControlNetUnionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet)
+```
+
+## ControlNetUnionModel
+
+[[autodoc]] ControlNetUnionModel
+
@@ -0,0 +1,30 @@
+<!-- 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. -->
+
+# HunyuanVideoTransformer3DModel
+
+A Diffusion Transformer model for 3D video-like data was introduced in [HunyuanVideo: A Systematic Framework For Large Video Generative Models](https://huggingface.co/papers/2412.03603) by Tencent.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import HunyuanVideoTransformer3DModel
+
+transformer = HunyuanVideoTransformer3DModel.from_pretrained("hunyuanvideo-community/HunyuanVideo", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
+## HunyuanVideoTransformer3DModel
+
+[[autodoc]] HunyuanVideoTransformer3DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput
@@ -0,0 +1,30 @@
+<!-- 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. -->
+
+# LTXVideoTransformer3DModel
+
+A Diffusion Transformer model for 3D data from [LTX](https://huggingface.co/Lightricks/LTX-Video) was introduced by Lightricks.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import LTXVideoTransformer3DModel
+
+transformer = LTXVideoTransformer3DModel.from_pretrained("Lightricks/LTX-Video", subfolder="transformer", torch_dtype=torch.bfloat16).to("cuda")
+```
+
+## LTXVideoTransformer3DModel
+
+[[autodoc]] LTXVideoTransformer3DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput
@@ -0,0 +1,34 @@
+<!-- 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. -->
+
+# SanaTransformer2DModel
+
+A Diffusion Transformer model for 2D data from [SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers](https://huggingface.co/papers/2410.10629) was introduced from NVIDIA and MIT HAN Lab, by Enze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, Song Han.
+
+The abstract from the paper is:
+
+*We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096×4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8×, we trained an AE that can compress images 32×, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024×1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released.*
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import SanaTransformer2DModel
+
+transformer = SanaTransformer2DModel.from_pretrained("Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
+## SanaTransformer2DModel
+
+[[autodoc]] SanaTransformer2DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput
@@ -19,10 +19,55 @@ The abstract from the paper is:

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`AllegroPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, AllegroTransformer3DModel, AllegroPipeline
+from diffusers.utils import export_to_video
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "rhymes-ai/Allegro",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = AllegroTransformer3DModel.from_pretrained(
+    "rhymes-ai/Allegro",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = AllegroPipeline.from_pretrained(
+    "rhymes-ai/Allegro",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = (
+    "A seaside harbor with bright sunlight and sparkling seawater, with many boats in the water. From an aerial view, "
+    "the boats vary in size and color, some moving and some stationary. Fishing boats in the water suggest that this "
+    "location might be a popular spot for docking fishing boats."
+)
+video = pipeline(prompt, guidance_scale=7.5, max_sequence_length=512).frames[0]
+export_to_video(video, "harbor.mp4", fps=15)
+```
+
 ## AllegroPipeline

 [[autodoc]] AllegroPipeline
@@ -12,7 +12,7 @@ specific language governing permissions and limitations under the License.

 # AuraFlow

-AuraFlow is inspired by [Stable Diffusion 3](../pipelines/stable_diffusion/stable_diffusion_3.md) and is by far the largest text-to-image generation model that comes with an Apache 2.0 license. This model achieves state-of-the-art results on the [GenEval](https://github.com/djghosh13/geneval) benchmark.
+AuraFlow is inspired by [Stable Diffusion 3](../pipelines/stable_diffusion/stable_diffusion_3) and is by far the largest text-to-image generation model that comes with an Apache 2.0 license. This model achieves state-of-the-art results on the [GenEval](https://github.com/djghosh13/geneval) benchmark.

 It was developed by the Fal team and more details about it can be found in [this blog post](https://blog.fal.ai/auraflow/).

@@ -22,6 +22,46 @@ AuraFlow can be quite expensive to run on consumer hardware devices. However, yo

 </Tip>

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`AuraFlowPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, AuraFlowTransformer2DModel, AuraFlowPipeline
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "fal/AuraFlow",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = AuraFlowTransformer2DModel.from_pretrained(
+    "fal/AuraFlow",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = AuraFlowPipeline.from_pretrained(
+    "fal/AuraFlow",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "a tiny astronaut hatching from an egg on the moon"
+image = pipeline(prompt).images[0]
+image.save("auraflow.png")
+```
+
 ## AuraFlowPipeline

 [[autodoc]] AuraFlowPipeline
@@ -23,7 +23,7 @@ The abstract from the paper is:

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

@@ -112,13 +112,46 @@ CogVideoX-2b requires about 19 GB of GPU memory to decode 49 frames (6 seconds o
  - With enabling cpu offloading and tiling, memory usage is `11 GB`
 - `pipe.vae.enable_slicing()`

-### Quantized inference
+## Quantization

-[torchao](https://github.com/pytorch/ao) and [optimum-quanto](https://github.com/huggingface/optimum-quanto/) can be used to quantize the text encoder, transformer and VAE modules to lower the memory requirements. This makes it possible to run the model on a free-tier T4 Colab or lower VRAM GPUs!
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.

-It is also worth noting that torchao quantization is fully compatible with [torch.compile](/optimization/torch2.0#torchcompile), which allows for much faster inference speed. Additionally, models can be serialized and stored in a quantized datatype to save disk space with torchao. Find examples and benchmarks in the gists below.
- [torchao](https://gist.github.com/a-r-r-o-w/4d9732d17412888c885480c6521a9897)
- [quanto](https://gist.github.com/a-r-r-o-w/31be62828b00a9292821b85c1017effa)
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`CogVideoXPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, CogVideoXTransformer3DModel, CogVideoXPipeline
+from diffusers.utils import export_to_video
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "THUDM/CogVideoX-2b",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = CogVideoXTransformer3DModel.from_pretrained(
+    "THUDM/CogVideoX-2b",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = CogVideoXPipeline.from_pretrained(
+    "THUDM/CogVideoX-2b",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "A detailed wooden toy ship with intricately carved masts and sails is seen gliding smoothly over a plush, blue carpet that mimics the waves of the sea. The ship's hull is painted a rich brown, with tiny windows. The carpet, soft and textured, provides a perfect backdrop, resembling an oceanic expanse. Surrounding the ship are various other toys and children's items, hinting at a playful environment. The scene captures the innocence and imagination of childhood, with the toy ship's journey symbolizing endless adventures in a whimsical, indoor setting."
+video = pipeline(prompt=prompt, guidance_scale=6, num_inference_steps=50).frames[0]
+export_to_video(video, "ship.mp4", fps=8)
+```

 ## CogVideoXPipeline

@@ -23,7 +23,7 @@ The abstract from the paper is:

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

@@ -0,0 +1,89 @@
+<!--Copyright 2024 The HuggingFace Team, The Black Forest 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.
+-->
+
+# FluxControlInpaint
+
+FluxControlInpaintPipeline is an implementation of Inpainting for Flux.1 Depth/Canny models. It is a pipeline that allows you to inpaint images using the Flux.1 Depth/Canny models. The pipeline takes an image and a mask as input and returns the inpainted image.
+
+FLUX.1 Depth and Canny [dev] is a 12 billion parameter rectified flow transformer capable of generating an image based on a text description while following the structure of a given input image. **This is not a ControlNet model**.
+
+| Control type | Developer | Link |
+| -------- | ---------- | ---- |
+| Depth | [Black Forest Labs](https://huggingface.co/black-forest-labs) | [Link](https://huggingface.co/black-forest-labs/FLUX.1-Depth-dev) |
+| Canny | [Black Forest Labs](https://huggingface.co/black-forest-labs) | [Link](https://huggingface.co/black-forest-labs/FLUX.1-Canny-dev) |
+
+
+<Tip>
+
+Flux can be quite expensive to run on consumer hardware devices. However, you can perform a suite of optimizations to run it faster and in a more memory-friendly manner. Check out [this section](https://huggingface.co/blog/sd3#memory-optimizations-for-sd3) for more details. Additionally, Flux can benefit from quantization for memory efficiency with a trade-off in inference latency. Refer to [this blog post](https://huggingface.co/blog/quanto-diffusers) to learn more. For an exhaustive list of resources, check out [this gist](https://gist.github.com/sayakpaul/b664605caf0aa3bf8585ab109dd5ac9c).
+
+</Tip>
+
+```python
+import torch
+from diffusers import FluxControlInpaintPipeline
+from diffusers.models.transformers import FluxTransformer2DModel
+from transformers import T5EncoderModel
+from diffusers.utils import load_image, make_image_grid
+from image_gen_aux import DepthPreprocessor # https://github.com/huggingface/image_gen_aux
+from PIL import Image
+import numpy as np
+
+pipe = FluxControlInpaintPipeline.from_pretrained(
+    "black-forest-labs/FLUX.1-Depth-dev",
+    torch_dtype=torch.bfloat16,
+)
+# use following lines if you have GPU constraints
+# ---------------------------------------------------------------
+transformer = FluxTransformer2DModel.from_pretrained(
+    "sayakpaul/FLUX.1-Depth-dev-nf4", subfolder="transformer", torch_dtype=torch.bfloat16
+)
+text_encoder_2 = T5EncoderModel.from_pretrained(
+    "sayakpaul/FLUX.1-Depth-dev-nf4", subfolder="text_encoder_2", torch_dtype=torch.bfloat16
+)
+pipe.transformer = transformer
+pipe.text_encoder_2 = text_encoder_2
+pipe.enable_model_cpu_offload()
+# ---------------------------------------------------------------
+pipe.to("cuda")
+
+prompt = "a blue robot singing opera with human-like expressions"
+image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/robot.png")
+
+head_mask = np.zeros_like(image)
+head_mask[65:580,300:642] = 255
+mask_image = Image.fromarray(head_mask)
+
+processor = DepthPreprocessor.from_pretrained("LiheYoung/depth-anything-large-hf")
+control_image = processor(image)[0].convert("RGB")
+
+output = pipe(
+    prompt=prompt,
+    image=image,
+    control_image=control_image,
+    mask_image=mask_image,
+    num_inference_steps=30,
+    strength=0.9,
+    guidance_scale=10.0,
+    generator=torch.Generator().manual_seed(42),
+).images[0]
+make_image_grid([image, control_image, mask_image, output.resize(image.size)], rows=1, cols=4).save("output.png")
+```
+
+## FluxControlInpaintPipeline
+[[autodoc]] FluxControlInpaintPipeline
+	- all
+	- __call__
+
+
+## FluxPipelineOutput
+[[autodoc]] pipelines.flux.pipeline_output.FluxPipelineOutput
@@ -0,0 +1,35 @@
+<!--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.
+-->
+
+# ControlNetUnion
+
+ControlNetUnionModel is an implementation of ControlNet for Stable Diffusion XL.
+
+The ControlNet model was introduced in [ControlNetPlus](https://github.com/xinsir6/ControlNetPlus) by xinsir6. It supports multiple conditioning inputs without increasing computation.
+
+*We design a new architecture that can support 10+ control types in condition text-to-image generation and can generate high resolution images visually comparable with midjourney. The network is based on the original ControlNet architecture, we propose two new modules to: 1 Extend the original ControlNet to support different image conditions using the same network parameter. 2 Support multiple conditions input without increasing computation offload, which is especially important for designers who want to edit image in detail, different conditions use the same condition encoder, without adding extra computations or parameters.*
+
+
+## StableDiffusionXLControlNetUnionPipeline
+[[autodoc]] StableDiffusionXLControlNetUnionPipeline
+	- all
+	- __call__
+
+## StableDiffusionXLControlNetUnionImg2ImgPipeline
+[[autodoc]] StableDiffusionXLControlNetUnionImg2ImgPipeline
+	- all
+	- __call__
+
+## StableDiffusionXLControlNetUnionInpaintPipeline
+[[autodoc]] StableDiffusionXLControlNetUnionInpaintPipeline
+	- all
+	- __call__
@@ -268,6 +268,43 @@ images = pipe(
 images[0].save("flux-redux.png")
 ```

+## Combining Flux Turbo LoRAs with Flux Control, Fill, and Redux
+
+We can combine Flux Turbo LoRAs with Flux Control and other pipelines like Fill and Redux to enable few-steps' inference. The example below shows how to do that for Flux Control LoRA for depth and turbo LoRA from [`ByteDance/Hyper-SD`](https://hf.co/ByteDance/Hyper-SD).
+
+```py
+from diffusers import FluxControlPipeline
+from image_gen_aux import DepthPreprocessor
+from diffusers.utils import load_image
+from huggingface_hub import hf_hub_download
+import torch
+
+control_pipe = FluxControlPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
+control_pipe.load_lora_weights("black-forest-labs/FLUX.1-Depth-dev-lora", adapter_name="depth")
+control_pipe.load_lora_weights(
+    hf_hub_download("ByteDance/Hyper-SD", "Hyper-FLUX.1-dev-8steps-lora.safetensors"), adapter_name="hyper-sd"
+)
+control_pipe.set_adapters(["depth", "hyper-sd"], adapter_weights=[0.85, 0.125])
+control_pipe.enable_model_cpu_offload()
+
+prompt = "A robot made of exotic candies and chocolates of different kinds. The background is filled with confetti and celebratory gifts."
+control_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/robot.png")
+
+processor = DepthPreprocessor.from_pretrained("LiheYoung/depth-anything-large-hf")
+control_image = processor(control_image)[0].convert("RGB")
+
+image = control_pipe(
+    prompt=prompt,
+    control_image=control_image,
+    height=1024,
+    width=1024,
+    num_inference_steps=8,
+    guidance_scale=10.0,
+    generator=torch.Generator().manual_seed(42),
+).images[0]
+image.save("output.png")
+```
+
 ## Running FP16 inference

 Flux can generate high-quality images with FP16 (i.e. to accelerate inference on Turing/Volta GPUs) but produces different outputs compared to FP32/BF16. The issue is that some activations in the text encoders have to be clipped when running in FP16, which affects the overall image. Forcing text encoders to run with FP32 inference thus removes this output difference. See [here](https://github.com/huggingface/diffusers/pull/9097#issuecomment-2272292516) for details.
@@ -297,6 +334,46 @@ out = pipe(
 out.save("image.png")
 ```

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`FluxPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, FluxTransformer2DModel, FluxPipeline
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "black-forest-labs/FLUX.1-dev",
+    subfolder="text_encoder_2",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = FluxTransformer2DModel.from_pretrained(
+    "black-forest-labs/FLUX.1-dev",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = FluxPipeline.from_pretrained(
+    "black-forest-labs/FLUX.1-dev",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "a tiny astronaut hatching from an egg on the moon"
+image = pipeline(prompt, guidance_scale=3.5, height=768, width=1360, num_inference_steps=50).images[0]
+image.save("flux.png")
+```
+
 ## Single File Loading for the `FluxTransformer2DModel`

 The `FluxTransformer2DModel` supports loading checkpoints in the original format shipped by Black Forest Labs. This is also useful when trying to load finetunes or quantized versions of the models that have been published by the community.
@@ -0,0 +1,74 @@
+<!-- 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. -->
+
+# HunyuanVideo
+
+[HunyuanVideo](https://www.arxiv.org/abs/2412.03603) by Tencent.
+
+*Recent advancements in video generation have significantly impacted daily life for both individuals and industries. However, the leading video generation models remain closed-source, resulting in a notable performance gap between industry capabilities and those available to the public. In this report, we introduce HunyuanVideo, an innovative open-source video foundation model that demonstrates performance in video generation comparable to, or even surpassing, that of leading closed-source models. HunyuanVideo encompasses a comprehensive framework that integrates several key elements, including data curation, advanced architectural design, progressive model scaling and training, and an efficient infrastructure tailored for large-scale model training and inference. As a result, we successfully trained a video generative model with over 13 billion parameters, making it the largest among all open-source models. We conducted extensive experiments and implemented a series of targeted designs to ensure high visual quality, motion dynamics, text-video alignment, and advanced filming techniques. According to evaluations by professionals, HunyuanVideo outperforms previous state-of-the-art models, including Runway Gen-3, Luma 1.6, and three top-performing Chinese video generative models. By releasing the code for the foundation model and its applications, we aim to bridge the gap between closed-source and open-source communities. This initiative will empower individuals within the community to experiment with their ideas, fostering a more dynamic and vibrant video generation ecosystem. The code is publicly available at [this https URL](https://github.com/Tencent/HunyuanVideo).*
+
+<Tip>
+
+Make sure to check out the Schedulers [guide](https://huggingface.co/docs/diffusers/main/en/using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.
+
+</Tip>
+
+Recommendations for inference:
+- Both text encoders should be in `torch.float16`.
+- Transformer should be in `torch.bfloat16`.
+- VAE should be in `torch.float16`.
+- `num_frames` should be of the form `4 * k + 1`, for example `49` or `129`.
+- For smaller resolution videos, try lower values of `shift` (between `2.0` to `5.0`) in the [Scheduler](https://huggingface.co/docs/diffusers/main/en/api/schedulers/flow_match_euler_discrete#diffusers.FlowMatchEulerDiscreteScheduler.shift). For larger resolution images, try higher values (between `7.0` and `12.0`). The default value is `7.0` for HunyuanVideo.
+- For more information about supported resolutions and other details, please refer to the original repository [here](https://github.com/Tencent/HunyuanVideo/).
+
+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`HunyuanVideoPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, HunyuanVideoTransformer3DModel, HunyuanVideoPipeline
+from diffusers.utils import export_to_video
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = HunyuanVideoTransformer3DModel.from_pretrained(
+    "tencent/HunyuanVideo",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = HunyuanVideoPipeline.from_pretrained(
+    "tencent/HunyuanVideo",
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "A cat walks on the grass, realistic style."
+video = pipeline(prompt=prompt, num_frames=61, num_inference_steps=30).frames[0]
+export_to_video(video, "cat.mp4", fps=15)
+```
+
+## HunyuanVideoPipeline
+
+[[autodoc]] HunyuanVideoPipeline
+  - all
+  - __call__
+
+## HunyuanVideoPipelineOutput
+
+[[autodoc]] pipelines.hunyuan_video.pipeline_output.HunyuanVideoPipelineOutput
@@ -30,7 +30,7 @@ HunyuanDiT has the following components:

 <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.
+Make sure to check out the Schedulers [guide](https://huggingface.co/docs/diffusers/main/en/using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](https://huggingface.co/docs/diffusers/main/en/using-diffusers/loading#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.

 </Tip>

@@ -28,7 +28,7 @@ This pipeline was contributed by [maxin-cn](https://github.com/maxin-cn). The or

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

@@ -70,6 +70,47 @@ Without torch.compile(): Average inference time: 16.246 seconds.
 With torch.compile(): Average inference time: 14.573 seconds.
 ```

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`LattePipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, LatteTransformer3DModel, LattePipeline
+from diffusers.utils import export_to_gif
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "maxin-cn/Latte-1",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = LatteTransformer3DModel.from_pretrained(
+    "maxin-cn/Latte-1",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = LattePipeline.from_pretrained(
+    "maxin-cn/Latte-1",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "A small cactus with a happy face in the Sahara desert."
+video = pipeline(prompt).frames[0]
+export_to_gif(video, "latte.gif")
+```
+
 ## LattePipeline

 [[autodoc]] LattePipeline
@@ -0,0 +1,197 @@
+<!-- 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. -->
+
+# LTX Video
+
+[LTX Video](https://huggingface.co/Lightricks/LTX-Video) is the first DiT-based video generation model capable of generating high-quality videos in real-time. It produces 24 FPS videos at a 768x512 resolution faster than they can be watched. Trained on a large-scale dataset of diverse videos, the model generates high-resolution videos with realistic and varied content. We provide a model for both text-to-video as well as image + text-to-video usecases.
+
+<Tip>
+
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.
+
+</Tip>
+
+Available models:
+
+|  Model name   | Recommended dtype |
+|:-------------:|:-----------------:|
+| [`LTX Video 0.9.0`](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltx-video-2b-v0.9.safetensors) | `torch.bfloat16` |
+| [`LTX Video 0.9.1`](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltx-video-2b-v0.9.1.safetensors) | `torch.bfloat16` |
+
+Note: The recommended dtype is for the transformer component. The VAE and text encoders can be either `torch.float32`, `torch.bfloat16` or `torch.float16` but the recommended dtype is `torch.bfloat16` as used in the original repository.
+
+## Loading Single Files
+
+Loading the original LTX Video checkpoints is also possible with [`~ModelMixin.from_single_file`]. We recommend using `from_single_file` for the Lightricks series of models, as they plan to release multiple models in the future in the single file format.
+
+```python
+import torch
+from diffusers import AutoencoderKLLTXVideo, LTXImageToVideoPipeline, LTXVideoTransformer3DModel
+
+# `single_file_url` could also be https://huggingface.co/Lightricks/LTX-Video/ltx-video-2b-v0.9.1.safetensors
+single_file_url = "https://huggingface.co/Lightricks/LTX-Video/ltx-video-2b-v0.9.safetensors"
+transformer = LTXVideoTransformer3DModel.from_single_file(
+  single_file_url, torch_dtype=torch.bfloat16
+)
+vae = AutoencoderKLLTXVideo.from_single_file(single_file_url, torch_dtype=torch.bfloat16)
+pipe = LTXImageToVideoPipeline.from_pretrained(
+  "Lightricks/LTX-Video", transformer=transformer, vae=vae, torch_dtype=torch.bfloat16
+)
+
+# ... inference code ...
+```
+
+Alternatively, the pipeline can be used to load the weights with [`~FromSingleFileMixin.from_single_file`].
+
+```python
+import torch
+from diffusers import LTXImageToVideoPipeline
+from transformers import T5EncoderModel, T5Tokenizer
+
+single_file_url = "https://huggingface.co/Lightricks/LTX-Video/ltx-video-2b-v0.9.safetensors"
+text_encoder = T5EncoderModel.from_pretrained(
+  "Lightricks/LTX-Video", subfolder="text_encoder", torch_dtype=torch.bfloat16
+)
+tokenizer = T5Tokenizer.from_pretrained(
+  "Lightricks/LTX-Video", subfolder="tokenizer", torch_dtype=torch.bfloat16
+)
+pipe = LTXImageToVideoPipeline.from_single_file(
+  single_file_url, text_encoder=text_encoder, tokenizer=tokenizer, torch_dtype=torch.bfloat16
+)
+```
+
+Loading [LTX GGUF checkpoints](https://huggingface.co/city96/LTX-Video-gguf) are also supported:
+
+```py
+import torch
+from diffusers.utils import export_to_video
+from diffusers import LTXPipeline, LTXVideoTransformer3DModel, GGUFQuantizationConfig
+
+ckpt_path = (
+    "https://huggingface.co/city96/LTX-Video-gguf/blob/main/ltx-video-2b-v0.9-Q3_K_S.gguf"
+)
+transformer = LTXVideoTransformer3DModel.from_single_file(
+    ckpt_path,
+    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
+    torch_dtype=torch.bfloat16,
+)
+pipe = LTXPipeline.from_pretrained(
+    "Lightricks/LTX-Video",
+    transformer=transformer,
+    torch_dtype=torch.bfloat16,
+)
+pipe.enable_model_cpu_offload()
+
+prompt = "A woman with long brown hair and light skin smiles at another woman with long blonde hair. The woman with brown hair wears a black jacket and has a small, barely noticeable mole on her right cheek. The camera angle is a close-up, focused on the woman with brown hair's face. The lighting is warm and natural, likely from the setting sun, casting a soft glow on the scene. The scene appears to be real-life footage"
+negative_prompt = "worst quality, inconsistent motion, blurry, jittery, distorted"
+
+video = pipe(
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    width=704,
+    height=480,
+    num_frames=161,
+    num_inference_steps=50,
+).frames[0]
+export_to_video(video, "output_gguf_ltx.mp4", fps=24)
+```
+
+Make sure to read the [documentation on GGUF](../../quantization/gguf) to learn more about our GGUF support.
+
+<!-- TODO(aryan): Update this when official weights are supported -->
+
+Loading and running inference with [LTX Video 0.9.1](https://huggingface.co/Lightricks/LTX-Video/blob/main/ltx-video-2b-v0.9.1.safetensors) weights.
+
+```python
+import torch
+from diffusers import LTXPipeline
+from diffusers.utils import export_to_video
+
+pipe = LTXPipeline.from_pretrained("a-r-r-o-w/LTX-Video-0.9.1-diffusers", torch_dtype=torch.bfloat16)
+pipe.to("cuda")
+
+prompt = "A woman with long brown hair and light skin smiles at another woman with long blonde hair. The woman with brown hair wears a black jacket and has a small, barely noticeable mole on her right cheek. The camera angle is a close-up, focused on the woman with brown hair's face. The lighting is warm and natural, likely from the setting sun, casting a soft glow on the scene. The scene appears to be real-life footage"
+negative_prompt = "worst quality, inconsistent motion, blurry, jittery, distorted"
+
+video = pipe(
+    prompt=prompt,
+    negative_prompt=negative_prompt,
+    width=768,
+    height=512,
+    num_frames=161,
+    decode_timestep=0.03,
+    decode_noise_scale=0.025,
+    num_inference_steps=50,
+).frames[0]
+export_to_video(video, "output.mp4", fps=24)
+```
+
+Refer to [this section](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cogvideox#memory-optimization) to learn more about optimizing memory consumption.
+
+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`LTXPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, LTXVideoTransformer3DModel, LTXPipeline
+from diffusers.utils import export_to_video
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "Lightricks/LTX-Video",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = LTXVideoTransformer3DModel.from_pretrained(
+    "Lightricks/LTX-Video",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = LTXPipeline.from_pretrained(
+    "Lightricks/LTX-Video",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "A detailed wooden toy ship with intricately carved masts and sails is seen gliding smoothly over a plush, blue carpet that mimics the waves of the sea. The ship's hull is painted a rich brown, with tiny windows. The carpet, soft and textured, provides a perfect backdrop, resembling an oceanic expanse. Surrounding the ship are various other toys and children's items, hinting at a playful environment. The scene captures the innocence and imagination of childhood, with the toy ship's journey symbolizing endless adventures in a whimsical, indoor setting."
+video = pipeline(prompt=prompt, num_frames=161, num_inference_steps=50).frames[0]
+export_to_video(video, "ship.mp4", fps=24)
+```
+
+## LTXPipeline
+
+[[autodoc]] LTXPipeline
+  - all
+  - __call__
+
+## LTXImageToVideoPipeline
+
+[[autodoc]] LTXImageToVideoPipeline
+  - all
+  - __call__
+
+## LTXPipelineOutput
+
+[[autodoc]] pipelines.ltx.pipeline_output.LTXPipelineOutput
@@ -47,7 +47,7 @@ This pipeline was contributed by [PommesPeter](https://github.com/PommesPeter).

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

@@ -82,6 +82,46 @@ pipeline.vae.decode = torch.compile(pipeline.vae.decode, mode="max-autotune", fu
 image = pipeline(prompt="Upper body of a young woman in a Victorian-era outfit with brass goggles and leather straps. Background shows an industrial revolution cityscape with smoky skies and tall, metal structures").images[0]
 ```

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`LuminaText2ImgPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, Transformer2DModel, LuminaText2ImgPipeline
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "Alpha-VLLM/Lumina-Next-SFT-diffusers",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = Transformer2DModel.from_pretrained(
+    "Alpha-VLLM/Lumina-Next-SFT-diffusers",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = LuminaText2ImgPipeline.from_pretrained(
+    "Alpha-VLLM/Lumina-Next-SFT-diffusers",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "a tiny astronaut hatching from an egg on the moon"
+image = pipeline(prompt).images[0]
+image.save("lumina.png")
+```
+
 ## LuminaText2ImgPipeline

 [[autodoc]] LuminaText2ImgPipeline
@@ -13,18 +13,257 @@
 # limitations under the License.
 -->

-# Mochi
+# Mochi 1 Preview

-[Mochi 1 Preview](https://huggingface.co/genmo/mochi-1-preview) from Genmo.
+> [!TIP]
+> Only a research preview of the model weights is available at the moment.
+
+[Mochi 1](https://huggingface.co/genmo/mochi-1-preview) is a video generation model by Genmo with a strong focus on prompt adherence and motion quality. The model features a 10B parameter Asmmetric Diffusion Transformer (AsymmDiT) architecture, and uses non-square QKV and output projection layers to reduce inference memory requirements. A single T5-XXL model is used to encode prompts.

 *Mochi 1 preview is an open state-of-the-art video generation model with high-fidelity motion and strong prompt adherence in preliminary evaluation. This model dramatically closes the gap between closed and open video generation systems. The model is released under a permissive Apache 2.0 license.*

+> [!TIP]
+> Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.
+
+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`MochiPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, MochiTransformer3DModel, MochiPipeline
+from diffusers.utils import export_to_video
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "genmo/mochi-1-preview",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = MochiTransformer3DModel.from_pretrained(
+    "genmo/mochi-1-preview",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = MochiPipeline.from_pretrained(
+    "genmo/mochi-1-preview",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+video = pipeline(
+  "Close-up of a cats eye, with the galaxy reflected in the cats eye. Ultra high resolution 4k.",
+  num_inference_steps=28,
+  guidance_scale=3.5
+).frames[0]
+export_to_video(video, "cat.mp4")
+```
+
+## Generating videos with Mochi-1 Preview
+
+The following example will download the full precision `mochi-1-preview` weights and produce the highest quality results but will require at least 42GB VRAM to run.
+
+```python
+import torch
+from diffusers import MochiPipeline
+from diffusers.utils import export_to_video
+
+pipe = MochiPipeline.from_pretrained("genmo/mochi-1-preview")
+
+# Enable memory savings
+pipe.enable_model_cpu_offload()
+pipe.enable_vae_tiling()
+
+prompt = "Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k."
+
+with torch.autocast("cuda", torch.bfloat16, cache_enabled=False):
+      frames = pipe(prompt, num_frames=85).frames[0]
+
+export_to_video(frames, "mochi.mp4", fps=30)
+```
+
+## Using a lower precision variant to save memory
+
+The following example will use the `bfloat16` variant of the model and requires 22GB VRAM to run. There is a slight drop in the quality of the generated video as a result.
+
+```python
+import torch
+from diffusers import MochiPipeline
+from diffusers.utils import export_to_video
+
+pipe = MochiPipeline.from_pretrained("genmo/mochi-1-preview", variant="bf16", torch_dtype=torch.bfloat16)
+
+# Enable memory savings
+pipe.enable_model_cpu_offload()
+pipe.enable_vae_tiling()
+
+prompt = "Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k."
+frames = pipe(prompt, num_frames=85).frames[0]
+
+export_to_video(frames, "mochi.mp4", fps=30)
+```
+
+## Reproducing the results from the Genmo Mochi repo
+
+The [Genmo Mochi implementation](https://github.com/genmoai/mochi/tree/main) uses different precision values for each stage in the inference process. The text encoder and VAE use `torch.float32`, while the DiT uses `torch.bfloat16` with the [attention kernel](https://pytorch.org/docs/stable/generated/torch.nn.attention.sdpa_kernel.html#torch.nn.attention.sdpa_kernel) set to `EFFICIENT_ATTENTION`. Diffusers pipelines currently do not support setting different `dtypes` for different stages of the pipeline. In order to run inference in the same way as the the original implementation, please refer to the following example.
+
 <Tip>
+The original Mochi implementation zeros out empty prompts. However, enabling this option and placing the entire pipeline under autocast can lead to numerical overflows with the T5 text encoder.

-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.
-
+When enabling `force_zeros_for_empty_prompt`, it is recommended to run the text encoding step outside the autocast context in full precision.
 </Tip>

+<Tip>
+Decoding the latents in full precision is very memory intensive. You will need at least 70GB VRAM to generate the 163 frames in this example. To reduce memory, either reduce the number of frames or run the decoding step in `torch.bfloat16`.
+</Tip>
+
+```python
+import torch
+from torch.nn.attention import SDPBackend, sdpa_kernel
+
+from diffusers import MochiPipeline
+from diffusers.utils import export_to_video
+from diffusers.video_processor import VideoProcessor
+
+pipe = MochiPipeline.from_pretrained("genmo/mochi-1-preview", force_zeros_for_empty_prompt=True)
+pipe.enable_vae_tiling()
+pipe.enable_model_cpu_offload()
+
+prompt =  "An aerial shot of a parade of elephants walking across the African savannah. The camera showcases the herd and the surrounding landscape."
+
+with torch.no_grad():
+    prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask = (
+        pipe.encode_prompt(prompt=prompt)
+    )
+
+with torch.autocast("cuda", torch.bfloat16):
+    with sdpa_kernel(SDPBackend.EFFICIENT_ATTENTION):
+        frames = pipe(
+            prompt_embeds=prompt_embeds,
+            prompt_attention_mask=prompt_attention_mask,
+            negative_prompt_embeds=negative_prompt_embeds,
+            negative_prompt_attention_mask=negative_prompt_attention_mask,
+            guidance_scale=4.5,
+            num_inference_steps=64,
+            height=480,
+            width=848,
+            num_frames=163,
+            generator=torch.Generator("cuda").manual_seed(0),
+            output_type="latent",
+            return_dict=False,
+        )[0]
+
+video_processor = VideoProcessor(vae_scale_factor=8)
+has_latents_mean = hasattr(pipe.vae.config, "latents_mean") and pipe.vae.config.latents_mean is not None
+has_latents_std = hasattr(pipe.vae.config, "latents_std") and pipe.vae.config.latents_std is not None
+if has_latents_mean and has_latents_std:
+    latents_mean = (
+        torch.tensor(pipe.vae.config.latents_mean).view(1, 12, 1, 1, 1).to(frames.device, frames.dtype)
+    )
+    latents_std = (
+        torch.tensor(pipe.vae.config.latents_std).view(1, 12, 1, 1, 1).to(frames.device, frames.dtype)
+    )
+    frames = frames * latents_std / pipe.vae.config.scaling_factor + latents_mean
+else:
+    frames = frames / pipe.vae.config.scaling_factor
+
+with torch.no_grad():
+    video = pipe.vae.decode(frames.to(pipe.vae.dtype), return_dict=False)[0]
+
+video = video_processor.postprocess_video(video)[0]
+export_to_video(video, "mochi.mp4", fps=30)
+```
+
+## Running inference with multiple GPUs
+
+It is possible to split the large Mochi transformer across multiple GPUs using the `device_map` and `max_memory` options in `from_pretrained`. In the following example we split the model across two GPUs, each with 24GB of VRAM.
+
+```python
+import torch
+from diffusers import MochiPipeline, MochiTransformer3DModel
+from diffusers.utils import export_to_video
+
+model_id = "genmo/mochi-1-preview"
+transformer = MochiTransformer3DModel.from_pretrained(
+    model_id,
+    subfolder="transformer",
+    device_map="auto",
+    max_memory={0: "24GB", 1: "24GB"}
+)
+
+pipe = MochiPipeline.from_pretrained(model_id,  transformer=transformer)
+pipe.enable_model_cpu_offload()
+pipe.enable_vae_tiling()
+
+with torch.autocast(device_type="cuda", dtype=torch.bfloat16, cache_enabled=False):
+    frames = pipe(
+        prompt="Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k.",
+        negative_prompt="",
+        height=480,
+        width=848,
+        num_frames=85,
+        num_inference_steps=50,
+        guidance_scale=4.5,
+        num_videos_per_prompt=1,
+        generator=torch.Generator(device="cuda").manual_seed(0),
+        max_sequence_length=256,
+        output_type="pil",
+    ).frames[0]
+
+export_to_video(frames, "output.mp4", fps=30)
+```
+
+## Using single file loading with the Mochi Transformer
+
+You can use `from_single_file` to load the Mochi transformer in its original format.
+
+<Tip>
+Diffusers currently doesn't support using the FP8 scaled versions of the Mochi single file checkpoints.
+</Tip>
+
+```python
+import torch
+from diffusers import MochiPipeline, MochiTransformer3DModel
+from diffusers.utils import export_to_video
+
+model_id = "genmo/mochi-1-preview"
+
+ckpt_path = "https://huggingface.co/Comfy-Org/mochi_preview_repackaged/blob/main/split_files/diffusion_models/mochi_preview_bf16.safetensors"
+
+transformer = MochiTransformer3DModel.from_pretrained(ckpt_path, torch_dtype=torch.bfloat16)
+
+pipe = MochiPipeline.from_pretrained(model_id,  transformer=transformer)
+pipe.enable_model_cpu_offload()
+pipe.enable_vae_tiling()
+
+with torch.autocast(device_type="cuda", dtype=torch.bfloat16, cache_enabled=False):
+    frames = pipe(
+        prompt="Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k.",
+        negative_prompt="",
+        height=480,
+        width=848,
+        num_frames=85,
+        num_inference_steps=50,
+        guidance_scale=4.5,
+        num_videos_per_prompt=1,
+        generator=torch.Generator(device="cuda").manual_seed(0),
+        max_sequence_length=256,
+        output_type="pil",
+    ).frames[0]
+
+export_to_video(frames, "output.mp4", fps=30)
+```
+
 ## MochiPipeline

 [[autodoc]] MochiPipeline
@@ -31,7 +31,7 @@ Some notes about this pipeline:

 <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.
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.

 </Tip>

@@ -0,0 +1,107 @@
+<!-- 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. -->
+
+# SanaPipeline
+
+[SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers](https://huggingface.co/papers/2410.10629) from NVIDIA and MIT HAN Lab, by Enze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, Song Han.
+
+The abstract from the paper is:
+
+*We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096×4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8×, we trained an AE that can compress images 32×, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024×1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released.*
+
+<Tip>
+
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) 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.
+
+</Tip>
+
+This pipeline was contributed by [lawrence-cj](https://github.com/lawrence-cj) and [chenjy2003](https://github.com/chenjy2003). The original codebase can be found [here](https://github.com/NVlabs/Sana). The original weights can be found under [hf.co/Efficient-Large-Model](https://huggingface.co/Efficient-Large-Model).
+
+Available models:
+
+| Model | Recommended dtype |
+|:-----:|:-----------------:|
+| [`Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers) | `torch.bfloat16` |
+| [`Efficient-Large-Model/Sana_1600M_1024px_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_1600M_1024px_diffusers) | `torch.float16` |
+| [`Efficient-Large-Model/Sana_1600M_1024px_MultiLing_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_1600M_1024px_MultiLing_diffusers) | `torch.float16` |
+| [`Efficient-Large-Model/Sana_1600M_512px_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_1600M_512px_diffusers) | `torch.float16` |
+| [`Efficient-Large-Model/Sana_1600M_512px_MultiLing_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_1600M_512px_MultiLing_diffusers) | `torch.float16` |
+| [`Efficient-Large-Model/Sana_600M_1024px_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_600M_1024px_diffusers) | `torch.float16` |
+| [`Efficient-Large-Model/Sana_600M_512px_diffusers`](https://huggingface.co/Efficient-Large-Model/Sana_600M_512px_diffusers) | `torch.float16` |
+
+Refer to [this](https://huggingface.co/collections/Efficient-Large-Model/sana-673efba2a57ed99843f11f9e) collection for more information.
+
+Note: The recommended dtype mentioned is for the transformer weights. The text encoder and VAE weights must stay in `torch.bfloat16` or `torch.float32` for the model to work correctly. Please refer to the inference example below to see how to load the model with the recommended dtype. 
+
+<Tip>
+
+Make sure to pass the `variant` argument for downloaded checkpoints to use lower disk space. Set it to `"fp16"` for models with recommended dtype as `torch.float16`, and `"bf16"` for models with recommended dtype as `torch.bfloat16`. By default, `torch.float32` weights are downloaded, which use twice the amount of disk storage. Additionally, `torch.float32` weights can be downcasted on-the-fly by specifying the `torch_dtype` argument. Read about it in the [docs](https://huggingface.co/docs/diffusers/v0.31.0/en/api/pipelines/overview#diffusers.DiffusionPipeline.from_pretrained).
+
+</Tip>
+
+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`SanaPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, SanaTransformer2DModel, SanaPipeline
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, AutoModelForCausalLM
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = AutoModelForCausalLM.from_pretrained(
+    "Efficient-Large-Model/Sana_1600M_1024px_diffusers",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = SanaTransformer2DModel.from_pretrained(
+    "Efficient-Large-Model/Sana_1600M_1024px_diffusers",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = SanaPipeline.from_pretrained(
+    "Efficient-Large-Model/Sana_1600M_1024px_diffusers",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "a tiny astronaut hatching from an egg on the moon"
+image = pipeline(prompt).images[0]
+image.save("sana.png")
+```
+
+## SanaPipeline
+
+[[autodoc]] SanaPipeline
+  - all
+  - __call__
+
+## SanaPAGPipeline
+
+[[autodoc]] SanaPAGPipeline
+  - all
+  - __call__
+
+## SanaPipelineOutput
+
+[[autodoc]] pipelines.sana.pipeline_output.SanaPipelineOutput
@@ -35,6 +35,57 @@ During inference:
 * The _quality_ of the generated audio sample can be controlled by the `num_inference_steps` argument; higher steps give higher quality audio at the expense of slower inference.
 * Multiple waveforms can be generated in one go: set `num_waveforms_per_prompt` to a value greater than 1 to enable. Automatic scoring will be performed between the generated waveforms and prompt text, and the audios ranked from best to worst accordingly.

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`StableAudioPipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, StableAudioDiTModel, StableAudioPipeline
+from diffusers.utils import export_to_video
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "stabilityai/stable-audio-open-1.0",
+    subfolder="text_encoder",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = StableAudioDiTModel.from_pretrained(
+    "stabilityai/stable-audio-open-1.0",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = StableAudioPipeline.from_pretrained(
+    "stabilityai/stable-audio-open-1.0",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "The sound of a hammer hitting a wooden surface."
+negative_prompt = "Low quality."
+audio = pipeline(
+    prompt,
+    negative_prompt=negative_prompt,
+    num_inference_steps=200,
+    audio_end_in_s=10.0,
+    num_waveforms_per_prompt=3,
+    generator=generator,
+).audios
+
+output = audio[0].T.float().cpu().numpy()
+sf.write("hammer.wav", output, pipeline.vae.sampling_rate)
+```
+

 ## StableAudioPipeline
 [[autodoc]] StableAudioPipeline
@@ -59,9 +59,76 @@ image.save("sd3_hello_world.png")
 - [`stabilityai/stable-diffusion-3.5-large`](https://huggingface.co/stabilityai/stable-diffusion-3-5-large)
 - [`stabilityai/stable-diffusion-3.5-large-turbo`](https://huggingface.co/stabilityai/stable-diffusion-3-5-large-turbo)

+## Image Prompting with IP-Adapters
+
+An IP-Adapter lets you prompt SD3 with images, in addition to the text prompt. This is especially useful when describing complex concepts that are difficult to articulate through text alone and you have reference images. To load and use an IP-Adapter, you need:
+
+- `image_encoder`: Pre-trained vision model used to obtain image features, usually a CLIP image encoder.
+- `feature_extractor`: Image processor that prepares the input image for the chosen `image_encoder`.
+- `ip_adapter_id`: Checkpoint containing parameters of image cross attention layers and image projection. 
+
+IP-Adapters are trained for a specific model architecture, so they also work in finetuned variations of the base model. You can use the [`~SD3IPAdapterMixin.set_ip_adapter_scale`] function to adjust how strongly the output aligns with the image prompt. The higher the value, the more closely the model follows the image prompt. A default value of 0.5 is typically a good balance, ensuring the model considers both the text and image prompts equally.
+
+```python
+import torch
+from PIL import Image
+
+from diffusers import StableDiffusion3Pipeline
+from transformers import SiglipVisionModel, SiglipImageProcessor
+
+image_encoder_id = "google/siglip-so400m-patch14-384"
+ip_adapter_id = "InstantX/SD3.5-Large-IP-Adapter"
+
+feature_extractor = SiglipImageProcessor.from_pretrained(
+    image_encoder_id,
+    torch_dtype=torch.float16
+)
+image_encoder = SiglipVisionModel.from_pretrained(
+    image_encoder_id,
+    torch_dtype=torch.float16
+).to( "cuda")
+
+pipe = StableDiffusion3Pipeline.from_pretrained(
+    "stabilityai/stable-diffusion-3.5-large",
+    torch_dtype=torch.float16,
+    feature_extractor=feature_extractor,
+    image_encoder=image_encoder,
+).to("cuda")
+
+pipe.load_ip_adapter(ip_adapter_id)
+pipe.set_ip_adapter_scale(0.6)
+
+ref_img = Image.open("image.jpg").convert('RGB')
+
+image = pipe(
+    width=1024,
+    height=1024,
+    prompt="a cat",
+    negative_prompt="lowres, low quality, worst quality",
+    num_inference_steps=24,
+    guidance_scale=5.0,
+    ip_adapter_image=ref_img
+).images[0]
+
+image.save("result.jpg")
+```
+
+<div class="justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sd3_ip_adapter_example.png"/>
+    <figcaption class="mt-2 text-sm text-center text-gray-500">IP-Adapter examples with prompt "a cat"</figcaption>
+</div>
+
+
+<Tip>
+
+Check out [IP-Adapter](../../../using-diffusers/ip_adapter) to learn more about how IP-Adapters work.
+
+</Tip>
+
+
 ## Memory Optimisations for SD3

-SD3 uses three text encoders, one if which is the very large T5-XXL model. This makes it challenging to run the model on GPUs with less than 24GB of VRAM, even when using `fp16` precision. The following section outlines a few memory optimizations in Diffusers that make it easier to run SD3 on low resource hardware.
+SD3 uses three text encoders, one of which is the very large T5-XXL model. This makes it challenging to run the model on GPUs with less than 24GB of VRAM, even when using `fp16` precision. The following section outlines a few memory optimizations in Diffusers that make it easier to run SD3 on low resource hardware.

 ### Running Inference with Model Offloading

@@ -201,6 +268,46 @@ image.save("sd3_hello_world.png")

 Check out the full script [here](https://gist.github.com/sayakpaul/508d89d7aad4f454900813da5d42ca97).

+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [`StableDiffusion3Pipeline`] for inference with bitsandbytes.
+
+```py
+import torch
+from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, SD3Transformer2DModel, StableDiffusion3Pipeline
+from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel
+
+quant_config = BitsAndBytesConfig(load_in_8bit=True)
+text_encoder_8bit = T5EncoderModel.from_pretrained(
+    "stabilityai/stable-diffusion-3.5-large",
+    subfolder="text_encoder_3",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
+transformer_8bit = SD3Transformer2DModel.from_pretrained(
+    "stabilityai/stable-diffusion-3.5-large",
+    subfolder="transformer",
+    quantization_config=quant_config,
+    torch_dtype=torch.float16,
+)
+
+pipeline = StableDiffusion3Pipeline.from_pretrained(
+    "stabilityai/stable-diffusion-3.5-large",
+    text_encoder=text_encoder_8bit,
+    transformer=transformer_8bit,
+    torch_dtype=torch.float16,
+    device_map="balanced",
+)
+
+prompt = "a tiny astronaut hatching from an egg on the moon"
+image = pipeline(prompt, num_inference_steps=28, guidance_scale=7.0).images[0]
+image.save("sd3.png")
+```
+
 ## Using Long Prompts with the T5 Text Encoder

 By default, the T5 Text Encoder prompt uses a maximum sequence length of `256`. This can be adjusted by setting the `max_sequence_length` to accept fewer or more tokens. Keep in mind that longer sequences require additional resources and result in longer generation times, such as during batch inference.
@@ -28,6 +28,13 @@ Learn how to quantize models in the [Quantization](../quantization/overview) gui

 [[autodoc]] BitsAndBytesConfig

+## GGUFQuantizationConfig
+
+[[autodoc]] GGUFQuantizationConfig
+## TorchAoConfig
+
+[[autodoc]] TorchAoConfig
+
 ## DiffusersQuantizer

 [[autodoc]] quantizers.base.DiffusersQuantizer
@@ -79,4 +79,8 @@ Happy exploring, and thank you for being part of the Diffusers community!
    <td><a href="https://github.com/Netwrck/stable-diffusion-server"> Stable Diffusion Server </a></td>
    <td>A server configured for Inpainting/Generation/img2img with one stable diffusion model</td>
  </tr>
+  <tr style="border-top: 2px solid black">
+    <td><a href="https://github.com/suzukimain/auto_diffusers"> Model Search </a></td>
+    <td>Search models on Civitai and Hugging Face</td>
+  </tr>
 </table>
@@ -0,0 +1,69 @@
+<!--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.
+
+-->
+
+# GGUF
+
+The GGUF file format is typically used to store models for inference with [GGML](https://github.com/ggerganov/ggml) and supports a variety of block wise quantization options. Diffusers supports loading checkpoints prequantized and saved in the GGUF format via `from_single_file` loading with Model classes. Loading GGUF checkpoints via Pipelines is currently not supported.
+
+The following example will load the [FLUX.1 DEV](https://huggingface.co/black-forest-labs/FLUX.1-dev) transformer model using the GGUF Q2_K quantization variant.
+
+Before starting please install gguf in your environment
+
+```shell
+pip install -U gguf
+```
+
+Since GGUF is a single file format, use [`~FromSingleFileMixin.from_single_file`] to load the model and pass in the [`GGUFQuantizationConfig`].
+
+When using GGUF checkpoints, the quantized weights remain in a low memory `dtype`(typically `torch.uint8`) and are dynamically dequantized and cast to the configured `compute_dtype` during each module's forward pass through the model. The `GGUFQuantizationConfig` allows you to set the `compute_dtype`.
+
+The functions used for dynamic dequantizatation are based on the great work done by [city96](https://github.com/city96/ComfyUI-GGUF), who created the Pytorch ports of the original [`numpy`](https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/gguf/quants.py) implementation by [compilade](https://github.com/compilade).
+
+```python
+import torch
+
+from diffusers import FluxPipeline, FluxTransformer2DModel, GGUFQuantizationConfig
+
+ckpt_path = (
+    "https://huggingface.co/city96/FLUX.1-dev-gguf/blob/main/flux1-dev-Q2_K.gguf"
+)
+transformer = FluxTransformer2DModel.from_single_file(
+    ckpt_path,
+    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
+    torch_dtype=torch.bfloat16,
+)
+pipe = FluxPipeline.from_pretrained(
+    "black-forest-labs/FLUX.1-dev",
+    transformer=transformer,
+    torch_dtype=torch.bfloat16,
+)
+pipe.enable_model_cpu_offload()
+prompt = "A cat holding a sign that says hello world"
+image = pipe(prompt, generator=torch.manual_seed(0)).images[0]
+image.save("flux-gguf.png")
+```
+
+## Supported Quantization Types
+
+- BF16
+- Q4_0
+- Q4_1
+- Q5_0
+- Q5_1
+- Q8_0
+- Q2_K
+- Q3_K
+- Q4_K
+- Q5_K
+- Q6_K
+
@@ -17,7 +17,7 @@ Quantization techniques focus on representing data with less information while a

 <Tip>

-Interested in adding a new quantization method to Transformers? Refer to the [Contribute new quantization method guide](https://huggingface.co/docs/transformers/main/en/quantization/contribute) to learn more about adding a new quantization method.
+Interested in adding a new quantization method to Diffusers? Refer to the [Contribute new quantization method guide](https://huggingface.co/docs/transformers/main/en/quantization/contribute) to learn more about adding a new quantization method.

 </Tip>

@@ -32,4 +32,9 @@ If you are new to the quantization field, we recommend you to check out these be

 ## When to use what?

-This section will be expanded once Diffusers has multiple quantization backends. Currently, we only support `bitsandbytes`. [This resource](https://huggingface.co/docs/transformers/main/en/quantization/overview#when-to-use-what) provides a good overview of the pros and cons of different quantization techniques. 
+Diffusers currently supports the following quantization methods.
+- [BitsandBytes](./bitsandbytes)
+- [TorchAO](./torchao)
+- [GGUF](./gguf)
+
+[This resource](https://huggingface.co/docs/transformers/main/en/quantization/overview#when-to-use-what) provides a good overview of the pros and cons of different quantization techniques.
@@ -0,0 +1,156 @@
+<!-- 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. -->
+
+# torchao
+
+[TorchAO](https://github.com/pytorch/ao) is an architecture optimization library for PyTorch. It provides high-performance dtypes, optimization techniques, and kernels for inference and training, featuring composability with native PyTorch features like [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html), FullyShardedDataParallel (FSDP), and more.
+
+Before you begin, make sure you have Pytorch 2.5+ and TorchAO installed.
+
+```bash
+pip install -U torch torchao
+```
+
+
+Quantize a model by passing [`TorchAoConfig`] to [`~ModelMixin.from_pretrained`] (you can also load pre-quantized models). This works for any model in any modality, as long as it supports loading with [Accelerate](https://hf.co/docs/accelerate/index) and contains `torch.nn.Linear` layers.
+
+The example below only quantizes the weights to int8.
+
+```python
+import torch
+from diffusers import FluxPipeline, FluxTransformer2DModel, TorchAoConfig
+
+model_id = "black-forest-labs/FLUX.1-dev"
+dtype = torch.bfloat16
+
+quantization_config = TorchAoConfig("int8wo")
+transformer = FluxTransformer2DModel.from_pretrained(
+    model_id,
+    subfolder="transformer",
+    quantization_config=quantization_config,
+    torch_dtype=dtype,
+)
+pipe = FluxPipeline.from_pretrained(
+    model_id,
+    transformer=transformer,
+    torch_dtype=dtype,
+)
+pipe.to("cuda")
+
+# Without quantization: ~31.447 GB
+# With quantization: ~20.40 GB
+print(f"Pipeline memory usage: {torch.cuda.max_memory_reserved() / 1024**3:.3f} GB")
+
+prompt = "A cat holding a sign that says hello world"
+image = pipe(
+    prompt, num_inference_steps=50, guidance_scale=4.5, max_sequence_length=512
+).images[0]
+image.save("output.png")
+```
+
+TorchAO is fully compatible with [torch.compile](./optimization/torch2.0#torchcompile), setting it apart from other quantization methods. This makes it easy to speed up inference with just one line of code.
+
+```python
+# In the above code, add the following after initializing the transformer
+transformer = torch.compile(transformer, mode="max-autotune", fullgraph=True)
+```
+
+For speed and memory benchmarks on Flux and CogVideoX, please refer to the table [here](https://github.com/huggingface/diffusers/pull/10009#issue-2688781450). You can also find some torchao [benchmarks](https://github.com/pytorch/ao/tree/main/torchao/quantization#benchmarks) numbers for various hardware.
+
+torchao also supports an automatic quantization API through [autoquant](https://github.com/pytorch/ao/blob/main/torchao/quantization/README.md#autoquantization). Autoquantization determines the best quantization strategy applicable to a model by comparing the performance of each technique on chosen input types and shapes. Currently, this can be used directly on the underlying modeling components. Diffusers will also expose an autoquant configuration option in the future.
+
+The `TorchAoConfig` class accepts three parameters:
+- `quant_type`: A string value mentioning one of the quantization types below.
+- `modules_to_not_convert`: A list of module full/partial module names for which quantization should not be performed. For example, to not perform any quantization of the [`FluxTransformer2DModel`]'s first block, one would specify: `modules_to_not_convert=["single_transformer_blocks.0"]`.
+- `kwargs`: A dict of keyword arguments to pass to the underlying quantization method which will be invoked based on `quant_type`.
+
+## Supported quantization types
+
+torchao supports weight-only quantization and weight and dynamic-activation quantization for int8, float3-float8, and uint1-uint7.
+
+Weight-only quantization stores the model weights in a specific low-bit data type but performs computation with a higher-precision data type, like `bfloat16`. This lowers the memory requirements from model weights but retains the memory peaks for activation computation.
+
+Dynamic activation quantization stores the model weights in a low-bit dtype, while also quantizing the activations on-the-fly to save additional memory. This lowers the memory requirements from model weights, while also lowering the memory overhead from activation computations. However, this may come at a quality tradeoff at times, so it is recommended to test different models thoroughly.
+
+The quantization methods supported are as follows:
+
+| **Category** | **Full Function Names** | **Shorthands** |
+|--------------|-------------------------|----------------|
+| **Integer quantization** | `int4_weight_only`, `int8_dynamic_activation_int4_weight`, `int8_weight_only`, `int8_dynamic_activation_int8_weight` | `int4wo`, `int4dq`, `int8wo`, `int8dq` |
+| **Floating point 8-bit quantization** | `float8_weight_only`, `float8_dynamic_activation_float8_weight`, `float8_static_activation_float8_weight` | `float8wo`, `float8wo_e5m2`, `float8wo_e4m3`, `float8dq`, `float8dq_e4m3`, `float8_e4m3_tensor`, `float8_e4m3_row` |
+| **Floating point X-bit quantization** | `fpx_weight_only` | `fpX_eAwB` where `X` is the number of bits (1-7), `A` is exponent bits, and `B` is mantissa bits. Constraint: `X == A + B + 1` |
+| **Unsigned Integer quantization** | `uintx_weight_only` | `uint1wo`, `uint2wo`, `uint3wo`, `uint4wo`, `uint5wo`, `uint6wo`, `uint7wo` |
+
+Some quantization methods are aliases (for example, `int8wo` is the commonly used shorthand for `int8_weight_only`). This allows using the quantization methods described in the torchao docs as-is, while also making it convenient to remember their shorthand notations.
+
+Refer to the official torchao documentation for a better understanding of the available quantization methods and the exhaustive list of configuration options available.
+
+## Serializing and Deserializing quantized models
+
+To serialize a quantized model in a given dtype, first load the model with the desired quantization dtype and then save it using the [`~ModelMixin.save_pretrained`] method.
+
+```python
+import torch
+from diffusers import FluxTransformer2DModel, TorchAoConfig
+
+quantization_config = TorchAoConfig("int8wo")
+transformer = FluxTransformer2DModel.from_pretrained(
+    "black-forest-labs/Flux.1-Dev",
+    subfolder="transformer",
+    quantization_config=quantization_config,
+    torch_dtype=torch.bfloat16,
+)
+transformer.save_pretrained("/path/to/flux_int8wo", safe_serialization=False)
+```
+
+To load a serialized quantized model, use the [`~ModelMixin.from_pretrained`] method.
+
+```python
+import torch
+from diffusers import FluxPipeline, FluxTransformer2DModel
+
+transformer = FluxTransformer2DModel.from_pretrained("/path/to/flux_int8wo", torch_dtype=torch.bfloat16, use_safetensors=False)
+pipe = FluxPipeline.from_pretrained("black-forest-labs/Flux.1-Dev", transformer=transformer, torch_dtype=torch.bfloat16)
+pipe.to("cuda")
+
+prompt = "A cat holding a sign that says hello world"
+image = pipe(prompt, num_inference_steps=30, guidance_scale=7.0).images[0]
+image.save("output.png")
+```
+
+Some quantization methods, such as `uint4wo`, cannot be loaded directly and may result in an `UnpicklingError` when trying to load the models, but work as expected when saving them. In order to work around this, one can load the state dict manually into the model. Note, however, that this requires using `weights_only=False` in `torch.load`, so it should be run only if the weights were obtained from a trustable source.
+
+```python
+import torch
+from accelerate import init_empty_weights
+from diffusers import FluxPipeline, FluxTransformer2DModel, TorchAoConfig
+
+# Serialize the model
+transformer = FluxTransformer2DModel.from_pretrained(
+    "black-forest-labs/Flux.1-Dev",
+    subfolder="transformer",
+    quantization_config=TorchAoConfig("uint4wo"),
+    torch_dtype=torch.bfloat16,
+)
+transformer.save_pretrained("/path/to/flux_uint4wo", safe_serialization=False, max_shard_size="50GB")
+# ...
+
+# Load the model
+state_dict = torch.load("/path/to/flux_uint4wo/diffusion_pytorch_model.bin", weights_only=False, map_location="cpu")
+with init_empty_weights():
+    transformer = FluxTransformer2DModel.from_config("/path/to/flux_uint4wo/config.json")
+transformer.load_state_dict(state_dict, strict=True, assign=True)
+```
+
+## Resources
+
+- [TorchAO Quantization API](https://github.com/pytorch/ao/blob/main/torchao/quantization/README.md)
+- [Diffusers-TorchAO examples](https://github.com/sayakpaul/diffusers-torchao)
@@ -56,7 +56,7 @@ image

 With the `adapter_name` parameter, it is really easy to use another adapter for inference! Load the [nerijs/pixel-art-xl](https://huggingface.co/nerijs/pixel-art-xl) adapter that has been fine-tuned to generate pixel art images and call it `"pixel"`.

-The pipeline automatically sets the first loaded adapter (`"toy"`) as the active adapter, but you can activate the `"pixel"` adapter with the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method:
+The pipeline automatically sets the first loaded adapter (`"toy"`) as the active adapter, but you can activate the `"pixel"` adapter with the [`~PeftAdapterMixin.set_adapters`] method:

 ```python
 pipe.load_lora_weights("nerijs/pixel-art-xl", weight_name="pixel-art-xl.safetensors", adapter_name="pixel")
@@ -85,7 +85,7 @@ By default, if the most up-to-date versions of PEFT and Transformers are detecte

 You can also merge different adapter checkpoints for inference to blend their styles together.

-Once again, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate the `pixel` and `toy` adapters and specify the weights for how they should be merged.
+Once again, use the [`~PeftAdapterMixin.set_adapters`] method to activate the `pixel` and `toy` adapters and specify the weights for how they should be merged.

 ```python
 pipe.set_adapters(["pixel", "toy"], adapter_weights=[0.5, 1.0])
@@ -114,7 +114,7 @@ Impressive! As you can see, the model generated an image that mixed the characte
 > [!TIP]
 > Through its PEFT integration, Diffusers also offers more efficient merging methods which you can learn about in the [Merge LoRAs](../using-diffusers/merge_loras) guide!

-To return to only using one adapter, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`] method to activate the `"toy"` adapter:
+To return to only using one adapter, use the [`~PeftAdapterMixin.set_adapters`] method to activate the `"toy"` adapter:

 ```python
 pipe.set_adapters("toy")
@@ -127,7 +127,7 @@ image = pipe(
 image
 ```

-Or to disable all adapters entirely, use the [`~diffusers.loaders.UNet2DConditionLoadersMixin.disable_lora`] method to return the base model.
+Or to disable all adapters entirely, use the [`~PeftAdapterMixin.disable_lora`] method to return the base model.

 ```python
 pipe.disable_lora()
@@ -140,7 +140,8 @@ image
 ![no-lora](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_20_1.png)

 ### Customize adapters strength
-For even more customization, you can control how strongly the adapter affects each part of the pipeline. For this, pass a dictionary with the control strengths (called "scales") to [`~diffusers.loaders.UNet2DConditionLoadersMixin.set_adapters`].
+
+For even more customization, you can control how strongly the adapter affects each part of the pipeline. For this, pass a dictionary with the control strengths (called "scales") to [`~PeftAdapterMixin.set_adapters`].

 For example, here's how you can turn on the adapter for the `down` parts, but turn it off for the `mid` and `up` parts:
 ```python
@@ -195,7 +196,7 @@ image

 ![block-lora-mixed](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/peft_integration/diffusers_peft_lora_inference_block_mixed.png)

-## Manage active adapters
+## Manage adapters

 You have attached multiple adapters in this tutorial, and if you're feeling a bit lost on what adapters have been attached to the pipeline's components, use the [`~diffusers.loaders.StableDiffusionLoraLoaderMixin.get_active_adapters`] method to check the list of active adapters:

@@ -212,3 +213,11 @@ list_adapters_component_wise = pipe.get_list_adapters()
 list_adapters_component_wise
 {"text_encoder": ["toy", "pixel"], "unet": ["toy", "pixel"], "text_encoder_2": ["toy", "pixel"]}
 ```
+
+The [`~PeftAdapterMixin.delete_adapters`] function completely removes an adapter and their LoRA layers from a model.
+
+```py
+pipe.delete_adapters("toy")
+pipe.get_active_adapters()
+["pixel"]
+```
@@ -1,4 +1,4 @@
-<!--Copyright 2024 The HuggingFace Team. All rights reserved.
+ <!--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
@@ -10,31 +10,20 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->

-# Text or image-to-video
+# Video generation

-Driven by the success of text-to-image diffusion models, generative video models are able to generate short clips of video from a text prompt or an initial image. These models extend a pretrained diffusion model to generate videos by adding some type of temporal and/or spatial convolution layer to the architecture. A mixed dataset of images and videos are used to train the model which learns to output a series of video frames based on the text or image conditioning.
+Video generation models include a temporal dimension to bring images, or frames, together to create a video. These models are trained on large-scale datasets of high-quality text-video pairs to learn how to combine the modalities to ensure the generated video is coherent and realistic.

-This guide will show you how to generate videos, how to configure video model parameters, and how to control video generation.
+[Explore](https://huggingface.co/models?other=video-generation) some of the more popular open-source video generation models available from Diffusers below.

-## Popular models
+<hfoptions id="popular-models">
+<hfoption id="CogVideoX">

-> [!TIP]
-> Discover other cool and trending video generation models on the Hub [here](https://huggingface.co/models?pipeline_tag=text-to-video&sort=trending)!
+[CogVideoX](https://huggingface.co/collections/THUDM/cogvideo-66c08e62f1685a3ade464cce) uses a 3D causal Variational Autoencoder (VAE) to compress videos along the spatial and temporal dimensions, and it includes a stack of expert transformer blocks with a 3D full attention mechanism to better capture visual, semantic, and motion information in the data.

-[Stable Video Diffusions (SVD)](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid), [I2VGen-XL](https://huggingface.co/ali-vilab/i2vgen-xl/), [AnimateDiff](https://huggingface.co/guoyww/animatediff), and [ModelScopeT2V](https://huggingface.co/ali-vilab/text-to-video-ms-1.7b) are popular models used for video diffusion. Each model is distinct. For example, AnimateDiff inserts a motion modeling module into a frozen text-to-image model to generate personalized animated images, whereas SVD is entirely pretrained from scratch with a three-stage training process to generate short high-quality videos.
+The CogVideoX family also includes models capable of generating videos from images and videos in addition to text. The image-to-video models are indicated by **I2V** in the checkpoint name, and they should be used with the [`CogVideoXImageToVideoPipeline`]. The regular checkpoints support video-to-video through the [`CogVideoXVideoToVideoPipeline`].

-[CogVideoX](https://huggingface.co/collections/THUDM/cogvideo-66c08e62f1685a3ade464cce) is another popular video generation model. The model is a multidimensional transformer that integrates text, time, and space. It employs full attention in the attention module and includes an expert block at the layer level to spatially align text and video.
-
-### CogVideoX
- 
-[CogVideoX](../api/pipelines/cogvideox) uses a 3D Variational Autoencoder (VAE) to compress videos along the spatial and temporal dimensions.
-
-Begin by loading the [`CogVideoXPipeline`] and passing an initial text or image to generate a video.
-<Tip>
-
-CogVideoX is available for image-to-video and text-to-video. [THUDM/CogVideoX-5b-I2V](https://huggingface.co/THUDM/CogVideoX-5b-I2V) uses the [`CogVideoXImageToVideoPipeline`] for image-to-video. [THUDM/CogVideoX-5b](https://huggingface.co/THUDM/CogVideoX-5b) and [THUDM/CogVideoX-2b](https://huggingface.co/THUDM/CogVideoX-2b) are available for text-to-video with the [`CogVideoXPipeline`].
- 
-</Tip>
+The example below demonstrates how to generate a video from an image and text prompt with [THUDM/CogVideoX-5b-I2V](https://huggingface.co/THUDM/CogVideoX-5b-I2V).

 ```py
 import torch
@@ -42,12 +31,13 @@ from diffusers import CogVideoXImageToVideoPipeline
 from diffusers.utils import export_to_video, load_image

 prompt = "A vast, shimmering ocean flows gracefully under a twilight sky, its waves undulating in a mesmerizing dance of blues and greens. The surface glints with the last rays of the setting sun, casting golden highlights that ripple across the water. Seagulls soar above, their cries blending with the gentle roar of the waves. The horizon stretches infinitely, where the ocean meets the sky in a seamless blend of hues. Close-ups reveal the intricate patterns of the waves, capturing the fluidity and dynamic beauty of the sea in motion."
-image = load_image(image="cogvideox_rocket.png")
+image = load_image(image="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cogvideox/cogvideox_rocket.png")
 pipe = CogVideoXImageToVideoPipeline.from_pretrained(
    "THUDM/CogVideoX-5b-I2V",
    torch_dtype=torch.bfloat16
 )
- 
+
+# reduce memory requirements 
 pipe.vae.enable_tiling()
 pipe.vae.enable_slicing()

@@ -60,7 +50,6 @@ video = pipe(
    guidance_scale=6,
    generator=torch.Generator(device="cuda").manual_seed(42),
 ).frames[0]
-
 export_to_video(video, "output.mp4", fps=8)
 ```

@@ -75,12 +64,103 @@ export_to_video(video, "output.mp4", fps=8)
  </div>
 </div>

- 
-### Stable Video Diffusion
+</hfoption>
+<hfoption id="HunyuanVideo">

-[SVD](../api/pipelines/svd) is based on the Stable Diffusion 2.1 model and it is trained on images, then low-resolution videos, and finally a smaller dataset of high-resolution videos. This model generates a short 2-4 second video from an initial image. You can learn more details about model, like micro-conditioning, in the [Stable Video Diffusion](../using-diffusers/svd) guide.
+> [!TIP]
+> HunyuanVideo is a 13B parameter model and requires a lot of memory. Refer to the HunyuanVideo [Quantization](../api/pipelines/hunyuan_video#quantization) guide to learn how to quantize the model. CogVideoX and LTX-Video are more lightweight options that can still generate high-quality videos.

-Begin by loading the [`StableVideoDiffusionPipeline`] and passing an initial image to generate a video from.
+[HunyuanVideo](https://huggingface.co/tencent/HunyuanVideo) features a dual-stream to single-stream diffusion transformer (DiT) for learning video and text tokens separately, and then subsequently concatenating the video and text tokens to combine their information. A single multimodal large language model (MLLM) serves as the text encoder, and videos are also spatio-temporally compressed with a 3D causal VAE.
+
+```py
+import torch
+from diffusers import HunyuanVideoPipeline, HunyuanVideoTransformer3DModel
+from diffusers.utils import export_to_video
+
+transformer = HunyuanVideoTransformer3DModel.from_pretrained(
+    "tencent/HunyuanVideo", subfolder="transformer", torch_dtype=torch.bfloat16
+)
+pipe = HunyuanVideoPipeline.from_pretrained(
+  "tencent/HunyuanVideo", transformer=transformer, torch_dtype=torch.float16
+)
+
+# reduce memory requirements
+pipe.vae.enable_tiling()
+pipe.to("cuda")
+
+video = pipe(
+    prompt="A cat walks on the grass, realistic",
+    height=320,
+    width=512,
+    num_frames=61,
+    num_inference_steps=30,
+).frames[0]
+export_to_video(video, "output.mp4", fps=15)
+```
+
+<div class="flex justify-center">
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/hunyuan-video-output.gif"/>
+</div>
+
+</hfoption>
+<hfoption id="LTX-Video">
+
+[LTX-Video (LTXV)](https://huggingface.co/Lightricks/LTX-Video) is a diffusion transformer (DiT) with a focus on speed. It generates 768x512 resolution videos at 24 frames per second (fps), enabling near real-time generation of high-quality videos. LTXV is relatively lightweight compared to other modern video generation models, making it possible to run on consumer GPUs.
+
+```py
+import torch
+from diffusers import LTXPipeline
+from diffusers.utils import export_to_video
+
+pipe = LTXPipeline.from_pretrained("Lightricks/LTX-Video", torch_dtype=torch.bfloat16).to("cuda")
+
+prompt = "A man walks towards a window, looks out, and then turns around. He has short, dark hair, dark skin, and is wearing a brown coat over a red and gray scarf. He walks from left to right towards a window, his gaze fixed on something outside. The camera follows him from behind at a medium distance. The room is brightly lit, with white walls and a large window covered by a white curtain. As he approaches the window, he turns his head slightly to the left, then back to the right. He then turns his entire body to the right, facing the window. The camera remains stationary as he stands in front of the window. The scene is captured in real-life footage."
+video = pipe(
+    prompt=prompt,
+    width=704,
+    height=480,
+    num_frames=161,
+    num_inference_steps=50,
+).frames[0]
+export_to_video(video, "output.mp4", fps=24)
+```
+
+<div class="flex justify-center">
+  <img src="https://huggingface.co/Lightricks/LTX-Video/resolve/main/media/ltx-video_example_00014.gif"/>
+</div>
+
+</hfoption>
+<hfoption id="Mochi-1">
+
+> [!TIP]
+> Mochi-1 is a 10B parameter model and requires a lot of memory. Refer to the Mochi [Quantization](../api/pipelines/mochi#quantization) guide to learn how to quantize the model. CogVideoX and LTX-Video are more lightweight options that can still generate high-quality videos.
+
+[Mochi-1](https://huggingface.co/genmo/mochi-1-preview) introduces the Asymmetric Diffusion Transformer (AsymmDiT) and Asymmetric Variational Autoencoder (AsymmVAE) to reduces memory requirements. AsymmVAE causally compresses videos 128x to improve memory efficiency, and AsymmDiT jointly attends to the compressed video tokens and user text tokens. This model is noted for generating videos with high-quality motion dynamics and strong prompt adherence.
+
+```py
+import torch
+from diffusers import MochiPipeline
+from diffusers.utils import export_to_video
+
+pipe = MochiPipeline.from_pretrained("genmo/mochi-1-preview", variant="bf16", torch_dtype=torch.bfloat16)
+
+# reduce memory requirements
+pipe.enable_model_cpu_offload()
+pipe.enable_vae_tiling()
+
+prompt = "Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k."
+video = pipe(prompt, num_frames=84).frames[0]
+export_to_video(video, "output.mp4", fps=30)
+```
+
+<div class="flex justify-center">
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/mochi-video-output.gif"/>
+</div>
+
+</hfoption>
+<hfoption id="StableVideoDiffusion">
+
+[StableVideoDiffusion (SVD)](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt) is based on the Stable Diffusion 2.1 model and it is trained on images, then low-resolution videos, and finally a smaller dataset of high-resolution videos. This model generates a short 2-4 second video from an initial image.

 ```py
 import torch
@@ -90,6 +170,8 @@ from diffusers.utils import load_image, export_to_video
 pipeline = StableVideoDiffusionPipeline.from_pretrained(
    "stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16"
 )
+
+# reduce memory requirements
 pipeline.enable_model_cpu_offload()

 image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png")
@@ -111,54 +193,12 @@ export_to_video(frames, "generated.mp4", fps=7)
  </div>
 </div>

-### I2VGen-XL
+</hfoption>
+<hfoption id="AnimateDiff">

-[I2VGen-XL](../api/pipelines/i2vgenxl) is a diffusion model that can generate higher resolution videos than SVD and it is also capable of accepting text prompts in addition to images. The model is trained with two hierarchical encoders (detail and global encoder) to better capture low and high-level details in images. These learned details are used to train a video diffusion model which refines the video resolution and details in the generated video.
+[AnimateDiff](https://huggingface.co/guoyww/animatediff) is an adapter model that inserts a motion module into a pretrained diffusion model to animate an image. The adapter is trained on video clips to learn motion which is used to condition the generation process to create a video. It is faster and easier to only train the adapter and it can be loaded into most diffusion models, effectively turning them into “video models”.

-You can use I2VGen-XL by loading the [`I2VGenXLPipeline`], and passing a text and image prompt to generate a video.
-
-```py
-import torch
-from diffusers import I2VGenXLPipeline
-from diffusers.utils import export_to_gif, load_image
-
-pipeline = I2VGenXLPipeline.from_pretrained("ali-vilab/i2vgen-xl", torch_dtype=torch.float16, variant="fp16")
-pipeline.enable_model_cpu_offload()
-
-image_url = "https://huggingface.co/datasets/diffusers/docs-images/resolve/main/i2vgen_xl_images/img_0009.png"
-image = load_image(image_url).convert("RGB")
-
-prompt = "Papers were floating in the air on a table in the library"
-negative_prompt = "Distorted, discontinuous, Ugly, blurry, low resolution, motionless, static, disfigured, disconnected limbs, Ugly faces, incomplete arms"
-generator = torch.manual_seed(8888)
-
-frames = pipeline(
-    prompt=prompt,
-    image=image,
-    num_inference_steps=50,
-    negative_prompt=negative_prompt,
-    guidance_scale=9.0,
-    generator=generator
-).frames[0]
-export_to_gif(frames, "i2v.gif")
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/i2vgen_xl_images/img_0009.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/i2vgen-xl-example.gif"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated video</figcaption>
-  </div>
-</div>
-
-### AnimateDiff
-
-[AnimateDiff](../api/pipelines/animatediff) is an adapter model that inserts a motion module into a pretrained diffusion model to animate an image. The adapter is trained on video clips to learn motion which is used to condition the generation process to create a video. It is faster and easier to only train the adapter and it can be loaded into most diffusion models, effectively turning them into "video models".
-
-Start by loading a [`MotionAdapter`].
+Load a `MotionAdapter` and pass it to the [`AnimateDiffPipeline`].

 ```py
 import torch
@@ -166,11 +206,6 @@ from diffusers import AnimateDiffPipeline, DDIMScheduler, MotionAdapter
 from diffusers.utils import export_to_gif

 adapter = MotionAdapter.from_pretrained("guoyww/animatediff-motion-adapter-v1-5-2", torch_dtype=torch.float16)
-```
-
-Then load a finetuned Stable Diffusion model with the [`AnimateDiffPipeline`].
-
-```py
 pipeline = AnimateDiffPipeline.from_pretrained("emilianJR/epiCRealism", motion_adapter=adapter, torch_dtype=torch.float16)
 scheduler = DDIMScheduler.from_pretrained(
    "emilianJR/epiCRealism",
@@ -181,13 +216,11 @@ scheduler = DDIMScheduler.from_pretrained(
    steps_offset=1,
 )
 pipeline.scheduler = scheduler
+
+# reduce memory requirements
 pipeline.enable_vae_slicing()
 pipeline.enable_model_cpu_offload()
-```

-Create a prompt and generate the video.
-
-```py
 output = pipeline(
    prompt="A space rocket with trails of smoke behind it launching into space from the desert, 4k, high resolution",
    negative_prompt="bad quality, worse quality, low resolution",
@@ -201,38 +234,11 @@ export_to_gif(frames, "animation.gif")
 ```

 <div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff.gif"/>
+  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/animatediff.gif"/>
 </div>

-### ModelscopeT2V
-
-[ModelscopeT2V](../api/pipelines/text_to_video) adds spatial and temporal convolutions and attention to a UNet, and it is trained on image-text and video-text datasets to enhance what it learns during training. The model takes a prompt, encodes it and creates text embeddings which are denoised by the UNet, and then decoded by a VQGAN into a video.
-
-<Tip>
-
-ModelScopeT2V generates watermarked videos due to the datasets it was trained on. To use a watermark-free model, try the [cerspense/zeroscope_v2_76w](https://huggingface.co/cerspense/zeroscope_v2_576w) model with the [`TextToVideoSDPipeline`] first, and then upscale it's output with the [cerspense/zeroscope_v2_XL](https://huggingface.co/cerspense/zeroscope_v2_XL) checkpoint using the [`VideoToVideoSDPipeline`].
-
-</Tip>
-
-Load a ModelScopeT2V checkpoint into the [`DiffusionPipeline`] along with a prompt to generate a video.
-
-```py
-import torch
-from diffusers import DiffusionPipeline
-from diffusers.utils import export_to_video
-
-pipeline = DiffusionPipeline.from_pretrained("damo-vilab/text-to-video-ms-1.7b", torch_dtype=torch.float16, variant="fp16")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_vae_slicing()
-
-prompt = "Confident teddy bear surfer rides the wave in the tropics"
-video_frames = pipeline(prompt).frames[0]
-export_to_video(video_frames, "modelscopet2v.mp4", fps=10)
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/modelscopet2v.gif" />
-</div>
+</hfoption>
+</hfoptions>

 ## Configure model parameters

@@ -548,3 +554,9 @@ If memory is not an issue and you want to optimize for speed, try wrapping the U
 + pipeline.to("cuda")
 + pipeline.unet = torch.compile(pipeline.unet, mode="reduce-overhead", fullgraph=True)
 ```
+
+## Quantization
+
+Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.
+
+Refer to the [Quantization](../../quantization/overview) to learn more about supported quantization backends (bitsandbytes, torchao, gguf) and selecting a quantization backend that supports your use case.
@@ -74,7 +74,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -73,7 +73,7 @@ from diffusers.utils.import_utils import is_xformers_available


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -79,7 +79,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -61,7 +61,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -52,7 +52,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -241,27 +241,15 @@ from diffusers import StableDiffusionPipeline
 from diffusers.callbacks import PipelineCallback, MultiPipelineCallbacks
 from diffusers.configuration_utils import register_to_config
 import torch
-from typing import Any, Dict, Optional
+from typing import Any, Dict, Tuple, Union


-pipeline: StableDiffusionPipeline = StableDiffusionPipeline.from_pretrained(
-    "stable-diffusion-v1-5/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    variant="fp16",
-    use_safetensors=True,
-).to("cuda")
-pipeline.safety_checker = None
-pipeline.requires_safety_checker = False
-
-
-class SDPromptScheduleCallback(PipelineCallback):
+class SDPromptSchedulingCallback(PipelineCallback):
    @register_to_config
    def __init__(
        self,
-        prompt: str,
-        negative_prompt: Optional[str] = None,
-        num_images_per_prompt: int = 1,
-        cutoff_step_ratio=1.0,
+        encoded_prompt: Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]],
+        cutoff_step_ratio=None,
        cutoff_step_index=None,
    ):
        super().__init__(
@@ -275,6 +263,10 @@ class SDPromptScheduleCallback(PipelineCallback):
    ) -> Dict[str, Any]:
        cutoff_step_ratio = self.config.cutoff_step_ratio
        cutoff_step_index = self.config.cutoff_step_index
+        if isinstance(self.config.encoded_prompt, tuple):
+            prompt_embeds, negative_prompt_embeds = self.config.encoded_prompt
+        else:
+            prompt_embeds = self.config.encoded_prompt

        # Use cutoff_step_index if it's not None, otherwise use cutoff_step_ratio
        cutoff_step = (
@@ -284,34 +276,164 @@ class SDPromptScheduleCallback(PipelineCallback):
        )

        if step_index == cutoff_step:
-            prompt_embeds, negative_prompt_embeds = pipeline.encode_prompt(
-                prompt=self.config.prompt,
-                negative_prompt=self.config.negative_prompt,
-                device=pipeline._execution_device,
-                num_images_per_prompt=self.config.num_images_per_prompt,
-                do_classifier_free_guidance=pipeline.do_classifier_free_guidance,
-            )
            if pipeline.do_classifier_free_guidance:
                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
            callback_kwargs[self.tensor_inputs[0]] = prompt_embeds
        return callback_kwargs

+
+pipeline: StableDiffusionPipeline = StableDiffusionPipeline.from_pretrained(
+    "stable-diffusion-v1-5/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    variant="fp16",
+    use_safetensors=True,
+).to("cuda")
+pipeline.safety_checker = None
+pipeline.requires_safety_checker = False
+
 callback = MultiPipelineCallbacks(
    [
-        SDPromptScheduleCallback(
-            prompt="Official portrait of a smiling world war ii general, female, cheerful, happy, detailed face, 20th century, highly detailed, cinematic lighting, digital art painting by Greg Rutkowski",
-            negative_prompt="Deformed, ugly, bad anatomy",
-            cutoff_step_ratio=0.25,
-        )
+        SDPromptSchedulingCallback(
+            encoded_prompt=pipeline.encode_prompt(
+                prompt=f"prompt {index}",
+                negative_prompt=f"negative prompt {index}",
+                device=pipeline._execution_device,
+                num_images_per_prompt=1,
+                # pipeline.do_classifier_free_guidance can't be accessed until after pipeline is ran
+                do_classifier_free_guidance=True,
+            ),
+            cutoff_step_index=index,
+        ) for index in range(1, 20)
    ]
 )

 image = pipeline(
-    prompt="Official portrait of a smiling world war ii general, male, cheerful, happy, detailed face, 20th century, highly detailed, cinematic lighting, digital art painting by Greg Rutkowski",
-    negative_prompt="Deformed, ugly, bad anatomy",
+    prompt="prompt"
+    negative_prompt="negative prompt",
    callback_on_step_end=callback,
    callback_on_step_end_tensor_inputs=["prompt_embeds"],
 ).images[0]
 torch.cuda.empty_cache()
 image.save('image.png')
 ```
+
+```python
+from diffusers import StableDiffusionXLPipeline
+from diffusers.callbacks import PipelineCallback, MultiPipelineCallbacks
+from diffusers.configuration_utils import register_to_config
+import torch
+from typing import Any, Dict, Tuple, Union
+
+
+class SDXLPromptSchedulingCallback(PipelineCallback):
+    @register_to_config
+    def __init__(
+        self,
+        encoded_prompt: Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]],
+        add_text_embeds: Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]],
+        add_time_ids: Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]],
+        cutoff_step_ratio=None,
+        cutoff_step_index=None,
+    ):
+        super().__init__(
+            cutoff_step_ratio=cutoff_step_ratio, cutoff_step_index=cutoff_step_index
+        )
+
+    tensor_inputs = ["prompt_embeds", "add_text_embeds", "add_time_ids"]
+
+    def callback_fn(
+        self, pipeline, step_index, timestep, callback_kwargs
+    ) -> Dict[str, Any]:
+        cutoff_step_ratio = self.config.cutoff_step_ratio
+        cutoff_step_index = self.config.cutoff_step_index
+        if isinstance(self.config.encoded_prompt, tuple):
+            prompt_embeds, negative_prompt_embeds = self.config.encoded_prompt
+        else:
+            prompt_embeds = self.config.encoded_prompt
+        if isinstance(self.config.add_text_embeds, tuple):
+            add_text_embeds, negative_add_text_embeds = self.config.add_text_embeds
+        else:
+            add_text_embeds = self.config.add_text_embeds
+        if isinstance(self.config.add_time_ids, tuple):
+            add_time_ids, negative_add_time_ids = self.config.add_time_ids
+        else:
+            add_time_ids = self.config.add_time_ids
+
+        # Use cutoff_step_index if it's not None, otherwise use cutoff_step_ratio
+        cutoff_step = (
+            cutoff_step_index
+            if cutoff_step_index is not None
+            else int(pipeline.num_timesteps * cutoff_step_ratio)
+        )
+
+        if step_index == cutoff_step:
+            if pipeline.do_classifier_free_guidance:
+                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+                add_text_embeds = torch.cat([negative_add_text_embeds, add_text_embeds])
+                add_time_ids = torch.cat([negative_add_time_ids, add_time_ids])
+            callback_kwargs[self.tensor_inputs[0]] = prompt_embeds
+            callback_kwargs[self.tensor_inputs[1]] = add_text_embeds
+            callback_kwargs[self.tensor_inputs[2]] = add_time_ids
+        return callback_kwargs
+
+
+pipeline: StableDiffusionXLPipeline = StableDiffusionXLPipeline.from_pretrained(
+    "stabilityai/stable-diffusion-xl-base-1.0",
+    torch_dtype=torch.float16,
+    variant="fp16",
+    use_safetensors=True,
+).to("cuda")
+
+callbacks = []
+for index in range(1, 20):
+    (
+        prompt_embeds,
+        negative_prompt_embeds,
+        pooled_prompt_embeds,
+        negative_pooled_prompt_embeds,
+    ) = pipeline.encode_prompt(
+        prompt=f"prompt {index}",
+        negative_prompt=f"prompt {index}",
+        device=pipeline._execution_device,
+        num_images_per_prompt=1,
+        # pipeline.do_classifier_free_guidance can't be accessed until after pipeline is ran
+        do_classifier_free_guidance=True,
+    )
+    text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+    add_time_ids = pipeline._get_add_time_ids(
+        (1024, 1024),
+        (0, 0),
+        (1024, 1024),
+        dtype=prompt_embeds.dtype,
+        text_encoder_projection_dim=text_encoder_projection_dim,
+    )
+    negative_add_time_ids = pipeline._get_add_time_ids(
+        (1024, 1024),
+        (0, 0),
+        (1024, 1024),
+        dtype=prompt_embeds.dtype,
+        text_encoder_projection_dim=text_encoder_projection_dim,
+    )
+    callbacks.append(
+        SDXLPromptSchedulingCallback(
+            encoded_prompt=(prompt_embeds, negative_prompt_embeds),
+            add_text_embeds=(pooled_prompt_embeds, negative_pooled_prompt_embeds),
+            add_time_ids=(add_time_ids, negative_add_time_ids),
+            cutoff_step_index=index,
+        )
+    )
+
+
+callback = MultiPipelineCallbacks(callbacks)
+
+image = pipeline(
+    prompt="prompt",
+    negative_prompt="negative prompt",
+    callback_on_step_end=callback,
+    callback_on_step_end_tensor_inputs=[
+        "prompt_embeds",
+        "add_text_embeds",
+        "add_time_ids",
+    ],
+).images[0]
+```
@@ -43,7 +43,7 @@ from diffusers.utils import BaseOutput, check_min_version


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")


 class MarigoldDepthOutput(BaseOutput):
@@ -648,6 +648,8 @@ class RFInversionFluxPipeline(
        height: Optional[int] = None,
        width: Optional[int] = None,
        eta: float = 1.0,
+        decay_eta: Optional[bool] = False,
+        eta_decay_power: Optional[float] = 1.0,
        strength: float = 1.0,
        start_timestep: float = 0,
        stop_timestep: float = 0.25,
@@ -880,12 +882,9 @@ class RFInversionFluxPipeline(
                    v_t = -noise_pred
                    v_t_cond = (y_0 - latents) / (1 - t_i)
                    eta_t = eta if start_timestep <= i < stop_timestep else 0.0
-                    if start_timestep <= i < stop_timestep:
-                        # controlled vector field
-                        v_hat_t = v_t + eta * (v_t_cond - v_t)
-
-                    else:
-                        v_hat_t = v_t
+                    if decay_eta:
+                        eta_t = eta_t * (1 - i / num_inference_steps) ** eta_decay_power  # Decay eta over the loop
+                    v_hat_t = v_t + eta_t * (v_t_cond - v_t)

                    # SDE Eq: 17 from https://arxiv.org/pdf/2410.10792
                    latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
@@ -1008,6 +1008,8 @@ class HunyuanDiTDifferentialImg2ImgPipeline(DiffusionPipeline):
            self.transformer.inner_dim // self.transformer.num_heads,
            grid_crops_coords,
            (grid_height, grid_width),
+            device=device,
+            output_type="pt",
        )

        style = torch.tensor([0], device=device)
@@ -129,7 +129,7 @@ class RegionalPromptingStableDiffusionPipeline(StableDiffusionPipeline):
        self.power = int(rp_args["power"]) if "power" in rp_args else 1

        prompts = prompt if isinstance(prompt, list) else [prompt]
-        n_prompts = negative_prompt if isinstance(prompt, list) else [negative_prompt]
+        n_prompts = negative_prompt if isinstance(negative_prompt, list) else [negative_prompt]
        self.batch = batch = num_images_per_prompt * len(prompts)

        if use_base:
@@ -73,7 +73,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -66,7 +66,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -79,7 +79,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -72,7 +72,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -78,7 +78,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -60,7 +60,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -60,7 +60,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = logging.getLogger(__name__)

@@ -65,7 +65,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)
 if is_torch_npu_available():
@@ -59,7 +59,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -61,7 +61,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)
 if is_torch_npu_available():
@@ -63,7 +63,7 @@ from diffusers.utils.import_utils import is_xformers_available


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -0,0 +1,127 @@
+# DreamBooth training example for SANA
+
+[DreamBooth](https://arxiv.org/abs/2208.12242) is a method to personalize text2image models like stable diffusion given just a few (3~5) images of a subject.
+
+The `train_dreambooth_lora_sana.py` script shows how to implement the training procedure with [LoRA](https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora) and adapt it for [SANA](https://arxiv.org/abs/2410.10629). 
+
+
+This will also allow us to push the trained model parameters to the Hugging Face Hub platform.
+
+## Running locally with PyTorch
+
+### Installing the dependencies
+
+Before running the scripts, make sure to install the library's training dependencies:
+
+**Important**
+
+To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
+
+```bash
+git clone https://github.com/huggingface/diffusers
+cd diffusers
+pip install -e .
+```
+
+Then cd in the `examples/dreambooth` folder and run
+```bash
+pip install -r requirements_sana.txt
+```
+
+And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
+
+```bash
+accelerate config
+```
+
+Or for a default accelerate configuration without answering questions about your environment
+
+```bash
+accelerate config default
+```
+
+Or if your environment doesn't support an interactive shell (e.g., a notebook)
+
+```python
+from accelerate.utils import write_basic_config
+write_basic_config()
+```
+
+When running `accelerate config`, if we specify torch compile mode to True there can be dramatic speedups.
+Note also that we use PEFT library as backend for LoRA training, make sure to have `peft>=0.14.0` installed in your environment.
+
+
+### Dog toy example
+
+Now let's get our dataset. For this example we will use some dog images: https://huggingface.co/datasets/diffusers/dog-example.
+
+Let's first download it locally:
+
+```python
+from huggingface_hub import snapshot_download
+
+local_dir = "./dog"
+snapshot_download(
+    "diffusers/dog-example",
+    local_dir=local_dir, repo_type="dataset",
+    ignore_patterns=".gitattributes",
+)
+```
+
+This will also allow us to push the trained LoRA parameters to the Hugging Face Hub platform.
+
+Now, we can launch training using:
+
+```bash
+export MODEL_NAME="Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers"
+export INSTANCE_DIR="dog"
+export OUTPUT_DIR="trained-sana-lora"
+
+accelerate launch train_dreambooth_lora_sana.py \
+  --pretrained_model_name_or_path=$MODEL_NAME  \
+  --instance_data_dir=$INSTANCE_DIR \
+  --output_dir=$OUTPUT_DIR \
+  --mixed_precision="bf16" \
+  --instance_prompt="a photo of sks dog" \
+  --resolution=1024 \
+  --train_batch_size=1 \
+  --gradient_accumulation_steps=4 \
+  --use_8bit_adam \
+  --learning_rate=1e-4 \
+  --report_to="wandb" \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --max_train_steps=500 \
+  --validation_prompt="A photo of sks dog in a bucket" \
+  --validation_epochs=25 \
+  --seed="0" \
+  --push_to_hub
+```
+
+For using `push_to_hub`, make you're logged into your Hugging Face account:
+
+```bash
+huggingface-cli login
+```
+
+To better track our training experiments, we're using the following flags in the command above:
+
+* `report_to="wandb` will ensure the training runs are tracked on [Weights and Biases](https://wandb.ai/site). To use it, be sure to install `wandb` with `pip install wandb`. Don't forget to call `wandb login <your_api_key>` before training if you haven't done it before.
+* `validation_prompt` and `validation_epochs` to allow the script to do a few validation inference runs. This allows us to qualitatively check if the training is progressing as expected.
+
+## Notes
+
+Additionally, we welcome you to explore the following CLI arguments:
+
+* `--lora_layers`: The transformer modules to apply LoRA training on. Please specify the layers in a comma seperated. E.g. - "to_k,to_q,to_v" will result in lora training of attention layers only.
+* `--complex_human_instruction`: Instructions for complex human attention as shown in [here](https://github.com/NVlabs/Sana/blob/main/configs/sana_app_config/Sana_1600M_app.yaml#L55).
+* `--max_sequence_length`: Maximum sequence length to use for text embeddings.
+
+
+We provide several options for optimizing memory optimization:
+
+* `--offload`: When enabled, we will offload the text encoder and VAE to CPU, when they are not used.
+* `cache_latents`: When enabled, we will pre-compute the latents from the input images with the VAE and remove the VAE from memory once done.
+* `--use_8bit_adam`: When enabled, we will use the 8bit version of AdamW provided by the `bitsandbytes` library.
+
+Refer to the [official documentation](https://huggingface.co/docs/diffusers/main/en/api/pipelines/sana) of the `SanaPipeline` to know more about the models available under the SANA family and their preferred dtypes during inference.
@@ -0,0 +1,8 @@
+accelerate>=1.0.0
+torchvision
+transformers>=4.47.0
+ftfy
+tensorboard
+Jinja2
+peft>=0.14.0
+sentencepiece
@@ -0,0 +1,206 @@
+# coding=utf-8
+# Copyright 2024 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 logging
+import os
+import sys
+import tempfile
+
+import safetensors
+
+
+sys.path.append("..")
+from test_examples_utils import ExamplesTestsAccelerate, run_command  # noqa: E402
+
+
+logging.basicConfig(level=logging.DEBUG)
+
+logger = logging.getLogger()
+stream_handler = logging.StreamHandler(sys.stdout)
+logger.addHandler(stream_handler)
+
+
+class DreamBoothLoRASANA(ExamplesTestsAccelerate):
+    instance_data_dir = "docs/source/en/imgs"
+    pretrained_model_name_or_path = "hf-internal-testing/tiny-sana-pipe"
+    script_path = "examples/dreambooth/train_dreambooth_lora_sana.py"
+    transformer_layer_type = "transformer_blocks.0.attn1.to_k"
+
+    def test_dreambooth_lora_sana(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                {self.script_path}
+                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
+                --instance_data_dir {self.instance_data_dir}
+                --resolution 32
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --max_sequence_length 16
+                """.split()
+
+            test_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"transformer"` in their names.
+            starts_with_transformer = all(key.startswith("transformer") for key in lora_state_dict.keys())
+            self.assertTrue(starts_with_transformer)
+
+    def test_dreambooth_lora_latent_caching(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                {self.script_path}
+                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
+                --instance_data_dir {self.instance_data_dir}
+                --resolution 32
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --cache_latents
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --max_sequence_length 16
+                """.split()
+
+            test_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"transformer"` in their names.
+            starts_with_transformer = all(key.startswith("transformer") for key in lora_state_dict.keys())
+            self.assertTrue(starts_with_transformer)
+
+    def test_dreambooth_lora_layers(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                {self.script_path}
+                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
+                --instance_data_dir {self.instance_data_dir}
+                --resolution 32
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --cache_latents
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lora_layers {self.transformer_layer_type}
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                --max_sequence_length 16
+                """.split()
+
+            test_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"transformer"` in their names. In this test, we only params of
+            # `self.transformer_layer_type` should be in the state dict.
+            starts_with_transformer = all(self.transformer_layer_type in key for key in lora_state_dict)
+            self.assertTrue(starts_with_transformer)
+
+    def test_dreambooth_lora_sana_checkpointing_checkpoints_total_limit(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            {self.script_path}
+            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
+            --instance_data_dir={self.instance_data_dir}
+            --output_dir={tmpdir}
+            --resolution=32
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --max_train_steps=6
+            --checkpoints_total_limit=2
+            --checkpointing_steps=2
+            --max_sequence_length 16
+            """.split()
+
+            test_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + test_args)
+
+            self.assertEqual(
+                {x for x in os.listdir(tmpdir) if "checkpoint" in x},
+                {"checkpoint-4", "checkpoint-6"},
+            )
+
+    def test_dreambooth_lora_sana_checkpointing_checkpoints_total_limit_removes_multiple_checkpoints(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+            {self.script_path}
+            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
+            --instance_data_dir={self.instance_data_dir}
+            --output_dir={tmpdir}
+            --resolution=32
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --max_train_steps=4
+            --checkpointing_steps=2
+            --max_sequence_length 166
+            """.split()
+
+            test_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + test_args)
+
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-2", "checkpoint-4"})
+
+            resume_run_args = f"""
+            {self.script_path}
+            --pretrained_model_name_or_path={self.pretrained_model_name_or_path}
+            --instance_data_dir={self.instance_data_dir}
+            --output_dir={tmpdir}
+            --resolution=32
+            --train_batch_size=1
+            --gradient_accumulation_steps=1
+            --max_train_steps=8
+            --checkpointing_steps=2
+            --resume_from_checkpoint=checkpoint-4
+            --checkpoints_total_limit=2
+            --max_sequence_length 16
+            """.split()
+
+            resume_run_args.extend(["--instance_prompt", ""])
+            run_command(self._launch_args + resume_run_args)
+
+            self.assertEqual({x for x in os.listdir(tmpdir) if "checkpoint" in x}, {"checkpoint-6", "checkpoint-8"})
@@ -63,7 +63,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -1300,16 +1300,17 @@ def main(args):
                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
                    # This is discussed in Section 4.2 of the same paper.
                    snr = compute_snr(noise_scheduler, timesteps)
-                    base_weight = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )

                    if noise_scheduler.config.prediction_type == "v_prediction":
                        # Velocity objective needs to be floored to an SNR weight of one.
-                        mse_loss_weights = base_weight + 1
+                        divisor = snr + 1
                    else:
-                        # Epsilon and sample both use the same loss weights.
-                        mse_loss_weights = base_weight
+                        divisor = snr
+
+                    mse_loss_weights = (
+                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / divisor
+                    )
+
                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
                    loss = loss.mean()
@@ -35,7 +35,7 @@ from diffusers.utils import check_min_version


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 # Cache compiled models across invocations of this script.
 cc.initialize_cache(os.path.expanduser("~/.cache/jax/compilation_cache"))
@@ -65,7 +65,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -70,7 +70,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -72,7 +72,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -29,7 +29,7 @@ import numpy as np
 import torch
 import torch.utils.checkpoint
 import transformers
-from accelerate import Accelerator
+from accelerate import Accelerator, DistributedType
 from accelerate.logging import get_logger
 from accelerate.utils import DistributedDataParallelKwargs, ProjectConfiguration, set_seed
 from huggingface_hub import create_repo, upload_folder
@@ -72,7 +72,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -1292,11 +1292,17 @@ def main(args):
            text_encoder_two_lora_layers_to_save = None

            for model in models:
-                if isinstance(model, type(unwrap_model(transformer))):
+                if isinstance(unwrap_model(model), type(unwrap_model(transformer))):
+                    model = unwrap_model(model)
+                    if args.upcast_before_saving:
+                        model = model.to(torch.float32)
                    transformer_lora_layers_to_save = get_peft_model_state_dict(model)
-                elif isinstance(model, type(unwrap_model(text_encoder_one))):  # or text_encoder_two
+                elif args.train_text_encoder and isinstance(
+                    unwrap_model(model), type(unwrap_model(text_encoder_one))
+                ):  # or text_encoder_two
                    # both text encoders are of the same class, so we check hidden size to distinguish between the two
-                    hidden_size = unwrap_model(model).config.hidden_size
+                    model = unwrap_model(model)
+                    hidden_size = model.config.hidden_size
                    if hidden_size == 768:
                        text_encoder_one_lora_layers_to_save = get_peft_model_state_dict(model)
                    elif hidden_size == 1280:
@@ -1305,7 +1311,8 @@ def main(args):
                    raise ValueError(f"unexpected save model: {model.__class__}")

                # make sure to pop weight so that corresponding model is not saved again
-                weights.pop()
+                if weights:
+                    weights.pop()

            StableDiffusion3Pipeline.save_lora_weights(
                output_dir,
@@ -1319,17 +1326,31 @@ def main(args):
        text_encoder_one_ = None
        text_encoder_two_ = None

-        while len(models) > 0:
-            model = models.pop()
+        if not accelerator.distributed_type == DistributedType.DEEPSPEED:
+            while len(models) > 0:
+                model = models.pop()

-            if isinstance(model, type(unwrap_model(transformer))):
-                transformer_ = model
-            elif isinstance(model, type(unwrap_model(text_encoder_one))):
-                text_encoder_one_ = model
-            elif isinstance(model, type(unwrap_model(text_encoder_two))):
-                text_encoder_two_ = model
-            else:
-                raise ValueError(f"unexpected save model: {model.__class__}")
+                if isinstance(unwrap_model(model), type(unwrap_model(transformer))):
+                    transformer_ = unwrap_model(model)
+                elif isinstance(unwrap_model(model), type(unwrap_model(text_encoder_one))):
+                    text_encoder_one_ = unwrap_model(model)
+                elif isinstance(unwrap_model(model), type(unwrap_model(text_encoder_two))):
+                    text_encoder_two_ = unwrap_model(model)
+                else:
+                    raise ValueError(f"unexpected save model: {model.__class__}")
+
+        else:
+            transformer_ = SD3Transformer2DModel.from_pretrained(
+                args.pretrained_model_name_or_path, subfolder="transformer"
+            )
+            transformer_.add_adapter(transformer_lora_config)
+            if args.train_text_encoder:
+                text_encoder_one_ = text_encoder_cls_one.from_pretrained(
+                    args.pretrained_model_name_or_path, subfolder="text_encoder"
+                )
+                text_encoder_two_ = text_encoder_cls_two.from_pretrained(
+                    args.pretrained_model_name_or_path, subfolder="text_encoder_2"
+                )

        lora_state_dict = StableDiffusion3Pipeline.lora_state_dict(input_dir)

@@ -1829,7 +1850,7 @@ def main(args):
                progress_bar.update(1)
                global_step += 1

-                if accelerator.is_main_process:
+                if accelerator.is_main_process or accelerator.distributed_type == DistributedType.DEEPSPEED:
                    if global_step % args.checkpointing_steps == 0:
                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
                        if args.checkpoints_total_limit is not None:
@@ -79,7 +79,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -63,7 +63,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -0,0 +1,204 @@
+# Training Flux Control
+
+This (experimental) example shows how to train Control LoRAs with [Flux](https://huggingface.co/black-forest-labs/FLUX.1-dev) by conditioning it with additional structural controls (like depth maps, poses, etc.). We provide a script for full fine-tuning, too, refer to [this section](#full-fine-tuning). To know more about Flux Control family, refer to the following resources:
+
+* [Docs](https://github.com/black-forest-labs/flux/blob/main/docs/structural-conditioning.md) by Black Forest Labs
+* Diffusers docs ([1](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux#canny-control), [2](https://huggingface.co/docs/diffusers/main/en/api/pipelines/flux#depth-control))
+
+To incorporate additional condition latents, we expand the input features of Flux.1-Dev from 64 to 128. The first 64 channels correspond to the original input latents to be denoised, while the latter 64 channels correspond to control latents. This expansion happens on the `x_embedder` layer, where the combined latents are projected to the expected feature dimension of rest of the network. Inference is performed using the `FluxControlPipeline`.
+
+> [!NOTE]
+> **Gated model**
+>
+> As the model is gated, before using it with diffusers you first need to go to the [FLUX.1 [dev] Hugging Face page](https://huggingface.co/black-forest-labs/FLUX.1-dev), fill in the form and accept the gate. Once you are in, you need to log in so that your system knows you’ve accepted the gate. Use the command below to log in:
+
+```bash
+huggingface-cli login
+```
+
+The example command below shows how to launch fine-tuning for pose conditions. The dataset ([`raulc0399/open_pose_controlnet`](https://huggingface.co/datasets/raulc0399/open_pose_controlnet)) being used here already has the pose conditions of the original images, so we don't have to compute them.
+
+```bash
+accelerate launch train_control_lora_flux.py \
+  --pretrained_model_name_or_path="black-forest-labs/FLUX.1-dev" \
+  --dataset_name="raulc0399/open_pose_controlnet" \
+  --output_dir="pose-control-lora" \
+  --mixed_precision="bf16" \
+  --train_batch_size=1 \
+  --rank=64 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --use_8bit_adam \
+  --learning_rate=1e-4 \
+  --report_to="wandb" \
+  --lr_scheduler="constant" \
+  --lr_warmup_steps=0 \
+  --max_train_steps=5000 \
+  --validation_image="openpose.png" \
+  --validation_prompt="A couple, 4k photo, highly detailed" \
+  --offload \
+  --seed="0" \
+  --push_to_hub
+```
+
+`openpose.png` comes from [here](https://huggingface.co/Adapter/t2iadapter/resolve/main/openpose.png).
+
+You need to install `diffusers` from the branch of [this PR](https://github.com/huggingface/diffusers/pull/9999). When it's merged, you should install `diffusers` from the `main`.
+
+The training script exposes additional CLI args that might be useful to experiment with:
+
+* `use_lora_bias`: When set, additionally trains the biases of the `lora_B` layer. 
+* `train_norm_layers`: When set, additionally trains the normalization scales. Takes care of saving and loading.
+* `lora_layers`: Specify the layers you want to apply LoRA to. If you specify "all-linear", all the linear layers will be LoRA-attached.
+
+### Training with DeepSpeed
+
+It's possible to train with [DeepSpeed](https://github.com/microsoft/DeepSpeed), specifically leveraging the Zero2 system optimization. To use it, save the following config to an YAML file (feel free to modify as needed):
+
+```yaml
+compute_environment: LOCAL_MACHINE
+debug: false
+deepspeed_config:
+  gradient_accumulation_steps: 1
+  gradient_clipping: 1.0
+  offload_optimizer_device: cpu
+  offload_param_device: cpu
+  zero3_init_flag: false
+  zero_stage: 2
+distributed_type: DEEPSPEED
+downcast_bf16: 'no'
+enable_cpu_affinity: false
+machine_rank: 0
+main_training_function: main
+mixed_precision: bf16
+num_machines: 1
+num_processes: 1
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false
+```
+
+And then while launching training, pass the config file:
+
+```bash
+accelerate launch --config_file=CONFIG_FILE.yaml ...
+```
+
+### Inference
+
+The pose images in our dataset were computed using the [`controlnet_aux`](https://github.com/huggingface/controlnet_aux) library. Let's install it first:
+
+```bash
+pip install controlnet_aux
+```
+
+And then we are ready:
+
+```py
+from controlnet_aux import OpenposeDetector
+from diffusers import FluxControlPipeline
+from diffusers.utils import load_image
+from PIL import Image
+import numpy as np
+import torch 
+
+pipe = FluxControlPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16).to("cuda")
+pipe.load_lora_weights("...") # change this.
+
+open_pose = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
+
+# prepare pose condition.
+url = "https://huggingface.co/Adapter/t2iadapter/resolve/main/people.jpg"
+image = load_image(url)
+image = open_pose(image, detect_resolution=512, image_resolution=1024)
+image = np.array(image)[:, :, ::-1]           
+image = Image.fromarray(np.uint8(image))
+
+prompt = "A couple, 4k photo, highly detailed"
+
+gen_images = pipe(
+  prompt=prompt,
+  condition_image=image,
+  num_inference_steps=50,
+  joint_attention_kwargs={"scale": 0.9},
+  guidance_scale=25., 
+).images[0]
+gen_images.save("output.png")
+```
+
+## Full fine-tuning
+
+We provide a non-LoRA version of the training script `train_control_flux.py`. Here is an example command:
+
+```bash
+accelerate launch --config_file=accelerate_ds2.yaml train_control_flux.py \
+  --pretrained_model_name_or_path="black-forest-labs/FLUX.1-dev" \
+  --dataset_name="raulc0399/open_pose_controlnet" \
+  --output_dir="pose-control" \
+  --mixed_precision="bf16" \
+  --train_batch_size=2 \
+  --dataloader_num_workers=4 \
+  --gradient_accumulation_steps=4 \
+  --gradient_checkpointing \
+  --use_8bit_adam \
+  --proportion_empty_prompts=0.2 \
+  --learning_rate=5e-5 \
+  --adam_weight_decay=1e-4 \
+  --report_to="wandb" \
+  --lr_scheduler="cosine" \
+  --lr_warmup_steps=1000 \
+  --checkpointing_steps=1000 \
+  --max_train_steps=10000 \
+  --validation_steps=200 \
+  --validation_image "2_pose_1024.jpg" "3_pose_1024.jpg" \
+  --validation_prompt "two friends sitting by each other enjoying a day at the park, full hd, cinematic" "person enjoying a day at the park, full hd, cinematic" \
+  --offload \
+  --seed="0" \
+  --push_to_hub
+```
+
+Change the `validation_image` and `validation_prompt` as needed.
+
+For inference, this time, we will run:
+
+```py
+from controlnet_aux import OpenposeDetector
+from diffusers import FluxControlPipeline, FluxTransformer2DModel
+from diffusers.utils import load_image
+from PIL import Image
+import numpy as np
+import torch 
+
+transformer = FluxTransformer2DModel.from_pretrained("...") # change this.
+pipe = FluxControlPipeline.from_pretrained(
+  "black-forest-labs/FLUX.1-dev",  transformer=transformer, torch_dtype=torch.bfloat16
+).to("cuda")
+
+open_pose = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
+
+# prepare pose condition.
+url = "https://huggingface.co/Adapter/t2iadapter/resolve/main/people.jpg"
+image = load_image(url)
+image = open_pose(image, detect_resolution=512, image_resolution=1024)
+image = np.array(image)[:, :, ::-1]           
+image = Image.fromarray(np.uint8(image))
+
+prompt = "A couple, 4k photo, highly detailed"
+
+gen_images = pipe(
+  prompt=prompt,
+  condition_image=image,
+  num_inference_steps=50,
+  guidance_scale=25., 
+).images[0]
+gen_images.save("output.png")
+```
+
+## Things to note
+
+* The scripts provided in this directory are experimental and educational. This means we may have to tweak things around to get good results on a given condition. We believe this is best done with the community 🤗
+* The scripts are not memory-optimized but we offload the VAE and the text encoders to CPU when they are not used. 
+* We can extract LoRAs from the fully fine-tuned model. While we currently don't provide any utilities for that, users are welcome to refer to [this script](https://github.com/Stability-AI/stability-ComfyUI-nodes/blob/master/control_lora_create.py) that provides a similar functionality. 
@@ -0,0 +1,6 @@
+transformers==4.47.0
+wandb
+torch
+torchvision
+accelerate==1.2.0
+peft>=0.14.0
@@ -57,7 +57,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -60,7 +60,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -52,7 +52,7 @@ if is_wandb_available():


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -46,7 +46,7 @@ from diffusers.utils import check_min_version, is_wandb_available


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -46,7 +46,7 @@ from diffusers.utils import check_min_version, is_wandb_available


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -51,7 +51,7 @@ if is_wandb_available():


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -60,7 +60,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -57,7 +57,7 @@ if is_wandb_available():


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -49,7 +49,7 @@ from diffusers.utils import check_min_version


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = logging.getLogger(__name__)

@@ -56,7 +56,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -68,7 +68,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)
 if is_torch_npu_available():
@@ -55,7 +55,7 @@ from diffusers.utils.torch_utils import is_compiled_module


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)
 if is_torch_npu_available():
@@ -81,7 +81,7 @@ else:


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -56,7 +56,7 @@ else:
 # ------------------------------------------------------------------------------

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = logging.getLogger(__name__)

@@ -76,7 +76,7 @@ else:


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__)

@@ -29,7 +29,7 @@ from diffusers.utils.import_utils import is_xformers_available


 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -50,7 +50,7 @@ if is_wandb_available():
    import wandb

 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.32.0.dev0")
+check_min_version("0.33.0.dev0")

 logger = get_logger(__name__, log_level="INFO")

@@ -0,0 +1,97 @@
+import argparse
+from contextlib import nullcontext
+
+import safetensors.torch
+from accelerate import init_empty_weights
+from huggingface_hub import hf_hub_download
+
+from diffusers.utils.import_utils import is_accelerate_available, is_transformers_available
+
+
+if is_transformers_available():
+    from transformers import CLIPVisionModelWithProjection
+
+    vision = True
+else:
+    vision = False
+
+"""
+python scripts/convert_flux_xlabs_ipadapter_to_diffusers.py  \
+--original_state_dict_repo_id "XLabs-AI/flux-ip-adapter" \
+--filename "flux-ip-adapter.safetensors"
+--output_path "flux-ip-adapter-hf/"
+"""
+
+
+CTX = init_empty_weights if is_accelerate_available else nullcontext
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--original_state_dict_repo_id", default=None, type=str)
+parser.add_argument("--filename", default="flux.safetensors", type=str)
+parser.add_argument("--checkpoint_path", default=None, type=str)
+parser.add_argument("--output_path", type=str)
+parser.add_argument("--vision_pretrained_or_path", default="openai/clip-vit-large-patch14", type=str)
+
+args = parser.parse_args()
+
+
+def load_original_checkpoint(args):
+    if args.original_state_dict_repo_id is not None:
+        ckpt_path = hf_hub_download(repo_id=args.original_state_dict_repo_id, filename=args.filename)
+    elif args.checkpoint_path is not None:
+        ckpt_path = args.checkpoint_path
+    else:
+        raise ValueError(" please provide either `original_state_dict_repo_id` or a local `checkpoint_path`")
+
+    original_state_dict = safetensors.torch.load_file(ckpt_path)
+    return original_state_dict
+
+
+def convert_flux_ipadapter_checkpoint_to_diffusers(original_state_dict, num_layers):
+    converted_state_dict = {}
+
+    # image_proj
+    ## norm
+    converted_state_dict["image_proj.norm.weight"] = original_state_dict.pop("ip_adapter_proj_model.norm.weight")
+    converted_state_dict["image_proj.norm.bias"] = original_state_dict.pop("ip_adapter_proj_model.norm.bias")
+    ## proj
+    converted_state_dict["image_proj.proj.weight"] = original_state_dict.pop("ip_adapter_proj_model.norm.weight")
+    converted_state_dict["image_proj.proj.bias"] = original_state_dict.pop("ip_adapter_proj_model.norm.bias")
+
+    # double transformer blocks
+    for i in range(num_layers):
+        block_prefix = f"ip_adapter.{i}."
+        # to_k_ip
+        converted_state_dict[f"{block_prefix}to_k_ip.bias"] = original_state_dict.pop(
+            f"double_blocks.{i}.processor.ip_adapter_double_stream_k_proj.bias"
+        )
+        converted_state_dict[f"{block_prefix}to_k_ip.weight"] = original_state_dict.pop(
+            f"double_blocks.{i}.processor.ip_adapter_double_stream_k_proj.weight"
+        )
+        # to_v_ip
+        converted_state_dict[f"{block_prefix}to_v_ip.bias"] = original_state_dict.pop(
+            f"double_blocks.{i}.processor.ip_adapter_double_stream_v_proj.bias"
+        )
+        converted_state_dict[f"{block_prefix}to_k_ip.weight"] = original_state_dict.pop(
+            f"double_blocks.{i}.processor.ip_adapter_double_stream_v_proj.weight"
+        )
+
+    return converted_state_dict
+
+
+def main(args):
+    original_ckpt = load_original_checkpoint(args)
+
+    num_layers = 19
+    converted_ip_adapter_state_dict = convert_flux_ipadapter_checkpoint_to_diffusers(original_ckpt, num_layers)
+
+    print("Saving Flux IP-Adapter in Diffusers format.")
+    safetensors.torch.save_file(converted_ip_adapter_state_dict, f"{args.output_path}/model.safetensors")
+
+    if vision:
+        model = CLIPVisionModelWithProjection.from_pretrained(args.vision_pretrained_or_path)
+        model.save_pretrained(f"{args.output_path}/image_encoder")
+
+
+if __name__ == "__main__":
+    main(args)
@@ -0,0 +1,257 @@
+import argparse
+from typing import Any, Dict
+
+import torch
+from accelerate import init_empty_weights
+from transformers import AutoModel, AutoTokenizer, CLIPTextModel, CLIPTokenizer
+
+from diffusers import (
+    AutoencoderKLHunyuanVideo,
+    FlowMatchEulerDiscreteScheduler,
+    HunyuanVideoPipeline,
+    HunyuanVideoTransformer3DModel,
+)
+
+
+def remap_norm_scale_shift_(key, state_dict):
+    weight = state_dict.pop(key)
+    shift, scale = weight.chunk(2, dim=0)
+    new_weight = torch.cat([scale, shift], dim=0)
+    state_dict[key.replace("final_layer.adaLN_modulation.1", "norm_out.linear")] = new_weight
+
+
+def remap_txt_in_(key, state_dict):
+    def rename_key(key):
+        new_key = key.replace("individual_token_refiner.blocks", "token_refiner.refiner_blocks")
+        new_key = new_key.replace("adaLN_modulation.1", "norm_out.linear")
+        new_key = new_key.replace("txt_in", "context_embedder")
+        new_key = new_key.replace("t_embedder.mlp.0", "time_text_embed.timestep_embedder.linear_1")
+        new_key = new_key.replace("t_embedder.mlp.2", "time_text_embed.timestep_embedder.linear_2")
+        new_key = new_key.replace("c_embedder", "time_text_embed.text_embedder")
+        new_key = new_key.replace("mlp", "ff")
+        return new_key
+
+    if "self_attn_qkv" in key:
+        weight = state_dict.pop(key)
+        to_q, to_k, to_v = weight.chunk(3, dim=0)
+        state_dict[rename_key(key.replace("self_attn_qkv", "attn.to_q"))] = to_q
+        state_dict[rename_key(key.replace("self_attn_qkv", "attn.to_k"))] = to_k
+        state_dict[rename_key(key.replace("self_attn_qkv", "attn.to_v"))] = to_v
+    else:
+        state_dict[rename_key(key)] = state_dict.pop(key)
+
+
+def remap_img_attn_qkv_(key, state_dict):
+    weight = state_dict.pop(key)
+    to_q, to_k, to_v = weight.chunk(3, dim=0)
+    state_dict[key.replace("img_attn_qkv", "attn.to_q")] = to_q
+    state_dict[key.replace("img_attn_qkv", "attn.to_k")] = to_k
+    state_dict[key.replace("img_attn_qkv", "attn.to_v")] = to_v
+
+
+def remap_txt_attn_qkv_(key, state_dict):
+    weight = state_dict.pop(key)
+    to_q, to_k, to_v = weight.chunk(3, dim=0)
+    state_dict[key.replace("txt_attn_qkv", "attn.add_q_proj")] = to_q
+    state_dict[key.replace("txt_attn_qkv", "attn.add_k_proj")] = to_k
+    state_dict[key.replace("txt_attn_qkv", "attn.add_v_proj")] = to_v
+
+
+def remap_single_transformer_blocks_(key, state_dict):
+    hidden_size = 3072
+
+    if "linear1.weight" in key:
+        linear1_weight = state_dict.pop(key)
+        split_size = (hidden_size, hidden_size, hidden_size, linear1_weight.size(0) - 3 * hidden_size)
+        q, k, v, mlp = torch.split(linear1_weight, split_size, dim=0)
+        new_key = key.replace("single_blocks", "single_transformer_blocks").removesuffix(".linear1.weight")
+        state_dict[f"{new_key}.attn.to_q.weight"] = q
+        state_dict[f"{new_key}.attn.to_k.weight"] = k
+        state_dict[f"{new_key}.attn.to_v.weight"] = v
+        state_dict[f"{new_key}.proj_mlp.weight"] = mlp
+
+    elif "linear1.bias" in key:
+        linear1_bias = state_dict.pop(key)
+        split_size = (hidden_size, hidden_size, hidden_size, linear1_bias.size(0) - 3 * hidden_size)
+        q_bias, k_bias, v_bias, mlp_bias = torch.split(linear1_bias, split_size, dim=0)
+        new_key = key.replace("single_blocks", "single_transformer_blocks").removesuffix(".linear1.bias")
+        state_dict[f"{new_key}.attn.to_q.bias"] = q_bias
+        state_dict[f"{new_key}.attn.to_k.bias"] = k_bias
+        state_dict[f"{new_key}.attn.to_v.bias"] = v_bias
+        state_dict[f"{new_key}.proj_mlp.bias"] = mlp_bias
+
+    else:
+        new_key = key.replace("single_blocks", "single_transformer_blocks")
+        new_key = new_key.replace("linear2", "proj_out")
+        new_key = new_key.replace("q_norm", "attn.norm_q")
+        new_key = new_key.replace("k_norm", "attn.norm_k")
+        state_dict[new_key] = state_dict.pop(key)
+
+
+TRANSFORMER_KEYS_RENAME_DICT = {
+    "img_in": "x_embedder",
+    "time_in.mlp.0": "time_text_embed.timestep_embedder.linear_1",
+    "time_in.mlp.2": "time_text_embed.timestep_embedder.linear_2",
+    "guidance_in.mlp.0": "time_text_embed.guidance_embedder.linear_1",
+    "guidance_in.mlp.2": "time_text_embed.guidance_embedder.linear_2",
+    "vector_in.in_layer": "time_text_embed.text_embedder.linear_1",
+    "vector_in.out_layer": "time_text_embed.text_embedder.linear_2",
+    "double_blocks": "transformer_blocks",
+    "img_attn_q_norm": "attn.norm_q",
+    "img_attn_k_norm": "attn.norm_k",
+    "img_attn_proj": "attn.to_out.0",
+    "txt_attn_q_norm": "attn.norm_added_q",
+    "txt_attn_k_norm": "attn.norm_added_k",
+    "txt_attn_proj": "attn.to_add_out",
+    "img_mod.linear": "norm1.linear",
+    "img_norm1": "norm1.norm",
+    "img_norm2": "norm2",
+    "img_mlp": "ff",
+    "txt_mod.linear": "norm1_context.linear",
+    "txt_norm1": "norm1.norm",
+    "txt_norm2": "norm2_context",
+    "txt_mlp": "ff_context",
+    "self_attn_proj": "attn.to_out.0",
+    "modulation.linear": "norm.linear",
+    "pre_norm": "norm.norm",
+    "final_layer.norm_final": "norm_out.norm",
+    "final_layer.linear": "proj_out",
+    "fc1": "net.0.proj",
+    "fc2": "net.2",
+    "input_embedder": "proj_in",
+}
+
+TRANSFORMER_SPECIAL_KEYS_REMAP = {
+    "txt_in": remap_txt_in_,
+    "img_attn_qkv": remap_img_attn_qkv_,
+    "txt_attn_qkv": remap_txt_attn_qkv_,
+    "single_blocks": remap_single_transformer_blocks_,
+    "final_layer.adaLN_modulation.1": remap_norm_scale_shift_,
+}
+
+VAE_KEYS_RENAME_DICT = {}
+
+VAE_SPECIAL_KEYS_REMAP = {}
+
+
+def update_state_dict_(state_dict: Dict[str, Any], old_key: str, new_key: str) -> Dict[str, Any]:
+    state_dict[new_key] = state_dict.pop(old_key)
+
+
+def get_state_dict(saved_dict: Dict[str, Any]) -> Dict[str, Any]:
+    state_dict = saved_dict
+    if "model" in saved_dict.keys():
+        state_dict = state_dict["model"]
+    if "module" in saved_dict.keys():
+        state_dict = state_dict["module"]
+    if "state_dict" in saved_dict.keys():
+        state_dict = state_dict["state_dict"]
+    return state_dict
+
+
+def convert_transformer(ckpt_path: str):
+    original_state_dict = get_state_dict(torch.load(ckpt_path, map_location="cpu", weights_only=True))
+
+    with init_empty_weights():
+        transformer = HunyuanVideoTransformer3DModel()
+
+    for key in list(original_state_dict.keys()):
+        new_key = key[:]
+        for replace_key, rename_key in TRANSFORMER_KEYS_RENAME_DICT.items():
+            new_key = new_key.replace(replace_key, rename_key)
+        update_state_dict_(original_state_dict, key, new_key)
+
+    for key in list(original_state_dict.keys()):
+        for special_key, handler_fn_inplace in TRANSFORMER_SPECIAL_KEYS_REMAP.items():
+            if special_key not in key:
+                continue
+            handler_fn_inplace(key, original_state_dict)
+
+    transformer.load_state_dict(original_state_dict, strict=True, assign=True)
+    return transformer
+
+
+def convert_vae(ckpt_path: str):
+    original_state_dict = get_state_dict(torch.load(ckpt_path, map_location="cpu", weights_only=True))
+
+    with init_empty_weights():
+        vae = AutoencoderKLHunyuanVideo()
+
+    for key in list(original_state_dict.keys()):
+        new_key = key[:]
+        for replace_key, rename_key in VAE_KEYS_RENAME_DICT.items():
+            new_key = new_key.replace(replace_key, rename_key)
+        update_state_dict_(original_state_dict, key, new_key)
+
+    for key in list(original_state_dict.keys()):
+        for special_key, handler_fn_inplace in VAE_SPECIAL_KEYS_REMAP.items():
+            if special_key not in key:
+                continue
+            handler_fn_inplace(key, original_state_dict)
+
+    vae.load_state_dict(original_state_dict, strict=True, assign=True)
+    return vae
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--transformer_ckpt_path", type=str, default=None, help="Path to original transformer checkpoint"
+    )
+    parser.add_argument("--vae_ckpt_path", type=str, default=None, help="Path to original VAE checkpoint")
+    parser.add_argument("--text_encoder_path", type=str, default=None, help="Path to original llama checkpoint")
+    parser.add_argument("--tokenizer_path", type=str, default=None, help="Path to original llama tokenizer")
+    parser.add_argument("--text_encoder_2_path", type=str, default=None, help="Path to original clip checkpoint")
+    parser.add_argument("--save_pipeline", action="store_true")
+    parser.add_argument("--output_path", type=str, required=True, help="Path where converted model should be saved")
+    parser.add_argument("--dtype", default="bf16", help="Torch dtype to save the transformer in.")
+    return parser.parse_args()
+
+
+DTYPE_MAPPING = {
+    "fp32": torch.float32,
+    "fp16": torch.float16,
+    "bf16": torch.bfloat16,
+}
+
+
+if __name__ == "__main__":
+    args = get_args()
+
+    transformer = None
+    dtype = DTYPE_MAPPING[args.dtype]
+
+    if args.save_pipeline:
+        assert args.transformer_ckpt_path is not None and args.vae_ckpt_path is not None
+        assert args.text_encoder_path is not None
+        assert args.tokenizer_path is not None
+        assert args.text_encoder_2_path is not None
+
+    if args.transformer_ckpt_path is not None:
+        transformer = convert_transformer(args.transformer_ckpt_path)
+        transformer = transformer.to(dtype=dtype)
+        if not args.save_pipeline:
+            transformer.save_pretrained(args.output_path, safe_serialization=True, max_shard_size="5GB")
+
+    if args.vae_ckpt_path is not None:
+        vae = convert_vae(args.vae_ckpt_path)
+        if not args.save_pipeline:
+            vae.save_pretrained(args.output_path, safe_serialization=True, max_shard_size="5GB")
+
+    if args.save_pipeline:
+        text_encoder = AutoModel.from_pretrained(args.text_encoder_path, torch_dtype=torch.float16)
+        tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_path, padding_side="right")
+        text_encoder_2 = CLIPTextModel.from_pretrained(args.text_encoder_2_path, torch_dtype=torch.float16)
+        tokenizer_2 = CLIPTokenizer.from_pretrained(args.text_encoder_2_path)
+        scheduler = FlowMatchEulerDiscreteScheduler(shift=7.0)
+
+        pipe = HunyuanVideoPipeline(
+            transformer=transformer,
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            text_encoder_2=text_encoder_2,
+            tokenizer_2=tokenizer_2,
+            scheduler=scheduler,
+        )
+        pipe.save_pretrained(args.output_path, safe_serialization=True, max_shard_size="5GB")
@@ -0,0 +1,297 @@
+import argparse
+from pathlib import Path
+from typing import Any, Dict
+
+import torch
+from accelerate import init_empty_weights
+from safetensors.torch import load_file
+from transformers import T5EncoderModel, T5Tokenizer
+
+from diffusers import AutoencoderKLLTXVideo, FlowMatchEulerDiscreteScheduler, LTXPipeline, LTXVideoTransformer3DModel
+
+
+def remove_keys_(key: str, state_dict: Dict[str, Any]):
+    state_dict.pop(key)
+
+
+TOKENIZER_MAX_LENGTH = 128
+
+TRANSFORMER_KEYS_RENAME_DICT = {
+    "patchify_proj": "proj_in",
+    "adaln_single": "time_embed",
+    "q_norm": "norm_q",
+    "k_norm": "norm_k",
+}
+
+TRANSFORMER_SPECIAL_KEYS_REMAP = {
+    "vae": remove_keys_,
+}
+
+VAE_KEYS_RENAME_DICT = {
+    # decoder
+    "up_blocks.0": "mid_block",
+    "up_blocks.1": "up_blocks.0",
+    "up_blocks.2": "up_blocks.1.upsamplers.0",
+    "up_blocks.3": "up_blocks.1",
+    "up_blocks.4": "up_blocks.2.conv_in",
+    "up_blocks.5": "up_blocks.2.upsamplers.0",
+    "up_blocks.6": "up_blocks.2",
+    "up_blocks.7": "up_blocks.3.conv_in",
+    "up_blocks.8": "up_blocks.3.upsamplers.0",
+    "up_blocks.9": "up_blocks.3",
+    # encoder
+    "down_blocks.0": "down_blocks.0",
+    "down_blocks.1": "down_blocks.0.downsamplers.0",
+    "down_blocks.2": "down_blocks.0.conv_out",
+    "down_blocks.3": "down_blocks.1",
+    "down_blocks.4": "down_blocks.1.downsamplers.0",
+    "down_blocks.5": "down_blocks.1.conv_out",
+    "down_blocks.6": "down_blocks.2",
+    "down_blocks.7": "down_blocks.2.downsamplers.0",
+    "down_blocks.8": "down_blocks.3",
+    "down_blocks.9": "mid_block",
+    # common
+    "conv_shortcut": "conv_shortcut.conv",
+    "res_blocks": "resnets",
+    "norm3.norm": "norm3",
+    "per_channel_statistics.mean-of-means": "latents_mean",
+    "per_channel_statistics.std-of-means": "latents_std",
+}
+
+VAE_091_RENAME_DICT = {
+    # decoder
+    "up_blocks.0": "mid_block",
+    "up_blocks.1": "up_blocks.0.upsamplers.0",
+    "up_blocks.2": "up_blocks.0",
+    "up_blocks.3": "up_blocks.1.upsamplers.0",
+    "up_blocks.4": "up_blocks.1",
+    "up_blocks.5": "up_blocks.2.upsamplers.0",
+    "up_blocks.6": "up_blocks.2",
+    "up_blocks.7": "up_blocks.3.upsamplers.0",
+    "up_blocks.8": "up_blocks.3",
+    # common
+    "last_time_embedder": "time_embedder",
+    "last_scale_shift_table": "scale_shift_table",
+}
+
+VAE_SPECIAL_KEYS_REMAP = {
+    "per_channel_statistics.channel": remove_keys_,
+    "per_channel_statistics.mean-of-means": remove_keys_,
+    "per_channel_statistics.mean-of-stds": remove_keys_,
+    "model.diffusion_model": remove_keys_,
+}
+
+VAE_091_SPECIAL_KEYS_REMAP = {
+    "timestep_scale_multiplier": remove_keys_,
+}
+
+
+def get_state_dict(saved_dict: Dict[str, Any]) -> Dict[str, Any]:
+    state_dict = saved_dict
+    if "model" in saved_dict.keys():
+        state_dict = state_dict["model"]
+    if "module" in saved_dict.keys():
+        state_dict = state_dict["module"]
+    if "state_dict" in saved_dict.keys():
+        state_dict = state_dict["state_dict"]
+    return state_dict
+
+
+def update_state_dict_inplace(state_dict: Dict[str, Any], old_key: str, new_key: str) -> Dict[str, Any]:
+    state_dict[new_key] = state_dict.pop(old_key)
+
+
+def convert_transformer(
+    ckpt_path: str,
+    dtype: torch.dtype,
+):
+    PREFIX_KEY = "model.diffusion_model."
+
+    original_state_dict = get_state_dict(load_file(ckpt_path))
+    with init_empty_weights():
+        transformer = LTXVideoTransformer3DModel()
+
+    for key in list(original_state_dict.keys()):
+        new_key = key[:]
+        if new_key.startswith(PREFIX_KEY):
+            new_key = key[len(PREFIX_KEY) :]
+        for replace_key, rename_key in TRANSFORMER_KEYS_RENAME_DICT.items():
+            new_key = new_key.replace(replace_key, rename_key)
+        update_state_dict_inplace(original_state_dict, key, new_key)
+
+    for key in list(original_state_dict.keys()):
+        for special_key, handler_fn_inplace in TRANSFORMER_SPECIAL_KEYS_REMAP.items():
+            if special_key not in key:
+                continue
+            handler_fn_inplace(key, original_state_dict)
+
+    transformer.load_state_dict(original_state_dict, strict=True, assign=True)
+    return transformer
+
+
+def convert_vae(ckpt_path: str, config, dtype: torch.dtype):
+    PREFIX_KEY = "vae."
+
+    original_state_dict = get_state_dict(load_file(ckpt_path))
+    with init_empty_weights():
+        vae = AutoencoderKLLTXVideo(**config)
+
+    for key in list(original_state_dict.keys()):
+        new_key = key[:]
+        if new_key.startswith(PREFIX_KEY):
+            new_key = key[len(PREFIX_KEY) :]
+        for replace_key, rename_key in VAE_KEYS_RENAME_DICT.items():
+            new_key = new_key.replace(replace_key, rename_key)
+        update_state_dict_inplace(original_state_dict, key, new_key)
+
+    for key in list(original_state_dict.keys()):
+        for special_key, handler_fn_inplace in VAE_SPECIAL_KEYS_REMAP.items():
+            if special_key not in key:
+                continue
+            handler_fn_inplace(key, original_state_dict)
+
+    vae.load_state_dict(original_state_dict, strict=True, assign=True)
+    return vae
+
+
+def get_vae_config(version: str) -> Dict[str, Any]:
+    if version == "0.9.0":
+        config = {
+            "in_channels": 3,
+            "out_channels": 3,
+            "latent_channels": 128,
+            "block_out_channels": (128, 256, 512, 512),
+            "decoder_block_out_channels": (128, 256, 512, 512),
+            "layers_per_block": (4, 3, 3, 3, 4),
+            "decoder_layers_per_block": (4, 3, 3, 3, 4),
+            "spatio_temporal_scaling": (True, True, True, False),
+            "decoder_spatio_temporal_scaling": (True, True, True, False),
+            "decoder_inject_noise": (False, False, False, False, False),
+            "upsample_residual": (False, False, False, False),
+            "upsample_factor": (1, 1, 1, 1),
+            "patch_size": 4,
+            "patch_size_t": 1,
+            "resnet_norm_eps": 1e-6,
+            "scaling_factor": 1.0,
+            "encoder_causal": True,
+            "decoder_causal": False,
+            "timestep_conditioning": False,
+        }
+    elif version == "0.9.1":
+        config = {
+            "in_channels": 3,
+            "out_channels": 3,
+            "latent_channels": 128,
+            "block_out_channels": (128, 256, 512, 512),
+            "decoder_block_out_channels": (256, 512, 1024),
+            "layers_per_block": (4, 3, 3, 3, 4),
+            "decoder_layers_per_block": (5, 6, 7, 8),
+            "spatio_temporal_scaling": (True, True, True, False),
+            "decoder_spatio_temporal_scaling": (True, True, True),
+            "decoder_inject_noise": (True, True, True, False),
+            "upsample_residual": (True, True, True),
+            "upsample_factor": (2, 2, 2),
+            "timestep_conditioning": True,
+            "patch_size": 4,
+            "patch_size_t": 1,
+            "resnet_norm_eps": 1e-6,
+            "scaling_factor": 1.0,
+            "encoder_causal": True,
+            "decoder_causal": False,
+        }
+        VAE_KEYS_RENAME_DICT.update(VAE_091_RENAME_DICT)
+        VAE_SPECIAL_KEYS_REMAP.update(VAE_091_SPECIAL_KEYS_REMAP)
+    return config
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--transformer_ckpt_path", type=str, default=None, help="Path to original transformer checkpoint"
+    )
+    parser.add_argument("--vae_ckpt_path", type=str, default=None, help="Path to original vae checkpoint")
+    parser.add_argument(
+        "--text_encoder_cache_dir", type=str, default=None, help="Path to text encoder cache directory"
+    )
+    parser.add_argument(
+        "--typecast_text_encoder",
+        action="store_true",
+        default=False,
+        help="Whether or not to apply fp16/bf16 precision to text_encoder",
+    )
+    parser.add_argument("--save_pipeline", action="store_true")
+    parser.add_argument("--output_path", type=str, required=True, help="Path where converted model should be saved")
+    parser.add_argument("--dtype", default="fp32", help="Torch dtype to save the model in.")
+    parser.add_argument(
+        "--version", type=str, default="0.9.0", choices=["0.9.0", "0.9.1"], help="Version of the LTX model"
+    )
+    return parser.parse_args()
+
+
+DTYPE_MAPPING = {
+    "fp32": torch.float32,
+    "fp16": torch.float16,
+    "bf16": torch.bfloat16,
+}
+
+VARIANT_MAPPING = {
+    "fp32": None,
+    "fp16": "fp16",
+    "bf16": "bf16",
+}
+
+
+if __name__ == "__main__":
+    args = get_args()
+
+    transformer = None
+    dtype = DTYPE_MAPPING[args.dtype]
+    variant = VARIANT_MAPPING[args.dtype]
+    output_path = Path(args.output_path)
+
+    if args.save_pipeline:
+        assert args.transformer_ckpt_path is not None and args.vae_ckpt_path is not None
+
+    if args.transformer_ckpt_path is not None:
+        transformer: LTXVideoTransformer3DModel = convert_transformer(args.transformer_ckpt_path, dtype)
+        if not args.save_pipeline:
+            transformer.save_pretrained(
+                output_path / "transformer", safe_serialization=True, max_shard_size="5GB", variant=variant
+            )
+
+    if args.vae_ckpt_path is not None:
+        config = get_vae_config(args.version)
+        vae: AutoencoderKLLTXVideo = convert_vae(args.vae_ckpt_path, config, dtype)
+        if not args.save_pipeline:
+            vae.save_pretrained(output_path / "vae", safe_serialization=True, max_shard_size="5GB", variant=variant)
+
+    if args.save_pipeline:
+        text_encoder_id = "google/t5-v1_1-xxl"
+        tokenizer = T5Tokenizer.from_pretrained(text_encoder_id, model_max_length=TOKENIZER_MAX_LENGTH)
+        text_encoder = T5EncoderModel.from_pretrained(text_encoder_id, cache_dir=args.text_encoder_cache_dir)
+
+        if args.typecast_text_encoder:
+            text_encoder = text_encoder.to(dtype=dtype)
+
+        # Apparently, the conversion does not work anymore without this :shrug:
+        for param in text_encoder.parameters():
+            param.data = param.data.contiguous()
+
+        scheduler = FlowMatchEulerDiscreteScheduler(
+            use_dynamic_shifting=True,
+            base_shift=0.95,
+            max_shift=2.05,
+            base_image_seq_len=1024,
+            max_image_seq_len=4096,
+            shift_terminal=0.1,
+        )
+
+        pipe = LTXPipeline(
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+        )
+
+        pipe.save_pretrained(args.output_path, safe_serialization=True, variant=variant, max_shard_size="5GB")
@@ -0,0 +1,314 @@
+#!/usr/bin/env python
+from __future__ import annotations
+
+import argparse
+import os
+from contextlib import nullcontext
+
+import torch
+from accelerate import init_empty_weights
+from huggingface_hub import hf_hub_download, snapshot_download
+from termcolor import colored
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+from diffusers import (
+    AutoencoderDC,
+    DPMSolverMultistepScheduler,
+    FlowMatchEulerDiscreteScheduler,
+    SanaPipeline,
+    SanaTransformer2DModel,
+)
+from diffusers.models.modeling_utils import load_model_dict_into_meta
+from diffusers.utils.import_utils import is_accelerate_available
+
+
+CTX = init_empty_weights if is_accelerate_available else nullcontext
+
+ckpt_ids = [
+    "Efficient-Large-Model/Sana_1600M_2Kpx_BF16/checkpoints/Sana_1600M_2Kpx_BF16.pth",
+    "Efficient-Large-Model/Sana_1600M_1024px_MultiLing/checkpoints/Sana_1600M_1024px_MultiLing.pth",
+    "Efficient-Large-Model/Sana_1600M_1024px_BF16/checkpoints/Sana_1600M_1024px_BF16.pth",
+    "Efficient-Large-Model/Sana_1600M_512px_MultiLing/checkpoints/Sana_1600M_512px_MultiLing.pth",
+    "Efficient-Large-Model/Sana_1600M_1024px/checkpoints/Sana_1600M_1024px.pth",
+    "Efficient-Large-Model/Sana_1600M_512px/checkpoints/Sana_1600M_512px.pth",
+    "Efficient-Large-Model/Sana_600M_1024px/checkpoints/Sana_600M_1024px_MultiLing.pth",
+    "Efficient-Large-Model/Sana_600M_512px/checkpoints/Sana_600M_512px_MultiLing.pth",
+]
+# https://github.com/NVlabs/Sana/blob/main/scripts/inference.py
+
+
+def main(args):
+    cache_dir_path = os.path.expanduser("~/.cache/huggingface/hub")
+
+    if args.orig_ckpt_path is None or args.orig_ckpt_path in ckpt_ids:
+        ckpt_id = args.orig_ckpt_path or ckpt_ids[0]
+        snapshot_download(
+            repo_id=f"{'/'.join(ckpt_id.split('/')[:2])}",
+            cache_dir=cache_dir_path,
+            repo_type="model",
+        )
+        file_path = hf_hub_download(
+            repo_id=f"{'/'.join(ckpt_id.split('/')[:2])}",
+            filename=f"{'/'.join(ckpt_id.split('/')[2:])}",
+            cache_dir=cache_dir_path,
+            repo_type="model",
+        )
+    else:
+        file_path = args.orig_ckpt_path
+
+    print(colored(f"Loading checkpoint from {file_path}", "green", attrs=["bold"]))
+    all_state_dict = torch.load(file_path, weights_only=True)
+    state_dict = all_state_dict.pop("state_dict")
+    converted_state_dict = {}
+
+    # Patch embeddings.
+    converted_state_dict["patch_embed.proj.weight"] = state_dict.pop("x_embedder.proj.weight")
+    converted_state_dict["patch_embed.proj.bias"] = state_dict.pop("x_embedder.proj.bias")
+
+    # Caption projection.
+    converted_state_dict["caption_projection.linear_1.weight"] = state_dict.pop("y_embedder.y_proj.fc1.weight")
+    converted_state_dict["caption_projection.linear_1.bias"] = state_dict.pop("y_embedder.y_proj.fc1.bias")
+    converted_state_dict["caption_projection.linear_2.weight"] = state_dict.pop("y_embedder.y_proj.fc2.weight")
+    converted_state_dict["caption_projection.linear_2.bias"] = state_dict.pop("y_embedder.y_proj.fc2.bias")
+
+    # AdaLN-single LN
+    converted_state_dict["time_embed.emb.timestep_embedder.linear_1.weight"] = state_dict.pop(
+        "t_embedder.mlp.0.weight"
+    )
+    converted_state_dict["time_embed.emb.timestep_embedder.linear_1.bias"] = state_dict.pop("t_embedder.mlp.0.bias")
+    converted_state_dict["time_embed.emb.timestep_embedder.linear_2.weight"] = state_dict.pop(
+        "t_embedder.mlp.2.weight"
+    )
+    converted_state_dict["time_embed.emb.timestep_embedder.linear_2.bias"] = state_dict.pop("t_embedder.mlp.2.bias")
+
+    # Shared norm.
+    converted_state_dict["time_embed.linear.weight"] = state_dict.pop("t_block.1.weight")
+    converted_state_dict["time_embed.linear.bias"] = state_dict.pop("t_block.1.bias")
+
+    # y norm
+    converted_state_dict["caption_norm.weight"] = state_dict.pop("attention_y_norm.weight")
+
+    # scheduler
+    flow_shift = 3.0
+
+    # model config
+    if args.model_type == "SanaMS_1600M_P1_D20":
+        layer_num = 20
+    elif args.model_type == "SanaMS_600M_P1_D28":
+        layer_num = 28
+    else:
+        raise ValueError(f"{args.model_type} is not supported.")
+    # Positional embedding interpolation scale.
+    interpolation_scale = {512: None, 1024: None, 2048: 1.0}
+
+    for depth in range(layer_num):
+        # Transformer blocks.
+        converted_state_dict[f"transformer_blocks.{depth}.scale_shift_table"] = state_dict.pop(
+            f"blocks.{depth}.scale_shift_table"
+        )
+
+        # Linear Attention is all you need 🤘
+        # Self attention.
+        q, k, v = torch.chunk(state_dict.pop(f"blocks.{depth}.attn.qkv.weight"), 3, dim=0)
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_q.weight"] = q
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_k.weight"] = k
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_v.weight"] = v
+        # Projection.
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_out.0.weight"] = state_dict.pop(
+            f"blocks.{depth}.attn.proj.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.attn1.to_out.0.bias"] = state_dict.pop(
+            f"blocks.{depth}.attn.proj.bias"
+        )
+
+        # Feed-forward.
+        converted_state_dict[f"transformer_blocks.{depth}.ff.conv_inverted.weight"] = state_dict.pop(
+            f"blocks.{depth}.mlp.inverted_conv.conv.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.conv_inverted.bias"] = state_dict.pop(
+            f"blocks.{depth}.mlp.inverted_conv.conv.bias"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.conv_depth.weight"] = state_dict.pop(
+            f"blocks.{depth}.mlp.depth_conv.conv.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.conv_depth.bias"] = state_dict.pop(
+            f"blocks.{depth}.mlp.depth_conv.conv.bias"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.ff.conv_point.weight"] = state_dict.pop(
+            f"blocks.{depth}.mlp.point_conv.conv.weight"
+        )
+
+        # Cross-attention.
+        q = state_dict.pop(f"blocks.{depth}.cross_attn.q_linear.weight")
+        q_bias = state_dict.pop(f"blocks.{depth}.cross_attn.q_linear.bias")
+        k, v = torch.chunk(state_dict.pop(f"blocks.{depth}.cross_attn.kv_linear.weight"), 2, dim=0)
+        k_bias, v_bias = torch.chunk(state_dict.pop(f"blocks.{depth}.cross_attn.kv_linear.bias"), 2, dim=0)
+
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_q.weight"] = q
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_q.bias"] = q_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_k.weight"] = k
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_k.bias"] = k_bias
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_v.weight"] = v
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_v.bias"] = v_bias
+
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_out.0.weight"] = state_dict.pop(
+            f"blocks.{depth}.cross_attn.proj.weight"
+        )
+        converted_state_dict[f"transformer_blocks.{depth}.attn2.to_out.0.bias"] = state_dict.pop(
+            f"blocks.{depth}.cross_attn.proj.bias"
+        )
+
+    # Final block.
+    converted_state_dict["proj_out.weight"] = state_dict.pop("final_layer.linear.weight")
+    converted_state_dict["proj_out.bias"] = state_dict.pop("final_layer.linear.bias")
+    converted_state_dict["scale_shift_table"] = state_dict.pop("final_layer.scale_shift_table")
+
+    # Transformer
+    with CTX():
+        transformer = SanaTransformer2DModel(
+            in_channels=32,
+            out_channels=32,
+            num_attention_heads=model_kwargs[args.model_type]["num_attention_heads"],
+            attention_head_dim=model_kwargs[args.model_type]["attention_head_dim"],
+            num_layers=model_kwargs[args.model_type]["num_layers"],
+            num_cross_attention_heads=model_kwargs[args.model_type]["num_cross_attention_heads"],
+            cross_attention_head_dim=model_kwargs[args.model_type]["cross_attention_head_dim"],
+            cross_attention_dim=model_kwargs[args.model_type]["cross_attention_dim"],
+            caption_channels=2304,
+            mlp_ratio=2.5,
+            attention_bias=False,
+            sample_size=args.image_size // 32,
+            patch_size=1,
+            norm_elementwise_affine=False,
+            norm_eps=1e-6,
+            interpolation_scale=interpolation_scale[args.image_size],
+        )
+
+    if is_accelerate_available():
+        load_model_dict_into_meta(transformer, converted_state_dict)
+    else:
+        transformer.load_state_dict(converted_state_dict, strict=True, assign=True)
+
+    try:
+        state_dict.pop("y_embedder.y_embedding")
+        state_dict.pop("pos_embed")
+    except KeyError:
+        print("y_embedder.y_embedding or pos_embed not found in the state_dict")
+
+    assert len(state_dict) == 0, f"State dict is not empty, {state_dict.keys()}"
+
+    num_model_params = sum(p.numel() for p in transformer.parameters())
+    print(f"Total number of transformer parameters: {num_model_params}")
+
+    transformer = transformer.to(weight_dtype)
+
+    if not args.save_full_pipeline:
+        print(
+            colored(
+                f"Only saving transformer model of {args.model_type}. "
+                f"Set --save_full_pipeline to save the whole SanaPipeline",
+                "green",
+                attrs=["bold"],
+            )
+        )
+        transformer.save_pretrained(
+            os.path.join(args.dump_path, "transformer"), safe_serialization=True, max_shard_size="5GB", variant=variant
+        )
+    else:
+        print(colored(f"Saving the whole SanaPipeline containing {args.model_type}", "green", attrs=["bold"]))
+        # VAE
+        ae = AutoencoderDC.from_pretrained("mit-han-lab/dc-ae-f32c32-sana-1.0-diffusers", torch_dtype=torch.float32)
+
+        # Text Encoder
+        text_encoder_model_path = "google/gemma-2-2b-it"
+        tokenizer = AutoTokenizer.from_pretrained(text_encoder_model_path)
+        tokenizer.padding_side = "right"
+        text_encoder = AutoModelForCausalLM.from_pretrained(
+            text_encoder_model_path, torch_dtype=torch.bfloat16
+        ).get_decoder()
+
+        # Scheduler
+        if args.scheduler_type == "flow-dpm_solver":
+            scheduler = DPMSolverMultistepScheduler(
+                flow_shift=flow_shift,
+                use_flow_sigmas=True,
+                prediction_type="flow_prediction",
+            )
+        elif args.scheduler_type == "flow-euler":
+            scheduler = FlowMatchEulerDiscreteScheduler(shift=flow_shift)
+        else:
+            raise ValueError(f"Scheduler type {args.scheduler_type} is not supported")
+
+        pipe = SanaPipeline(
+            tokenizer=tokenizer,
+            text_encoder=text_encoder,
+            transformer=transformer,
+            vae=ae,
+            scheduler=scheduler,
+        )
+        pipe.save_pretrained(args.dump_path, safe_serialization=True, max_shard_size="5GB", variant=variant)
+
+
+DTYPE_MAPPING = {
+    "fp32": torch.float32,
+    "fp16": torch.float16,
+    "bf16": torch.bfloat16,
+}
+
+VARIANT_MAPPING = {
+    "fp32": None,
+    "fp16": "fp16",
+    "bf16": "bf16",
+}
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--orig_ckpt_path", default=None, type=str, required=False, help="Path to the checkpoint to convert."
+    )
+    parser.add_argument(
+        "--image_size",
+        default=1024,
+        type=int,
+        choices=[512, 1024, 2048],
+        required=False,
+        help="Image size of pretrained model, 512, 1024 or 2048.",
+    )
+    parser.add_argument(
+        "--model_type", default="SanaMS_1600M_P1_D20", type=str, choices=["SanaMS_1600M_P1_D20", "SanaMS_600M_P1_D28"]
+    )
+    parser.add_argument(
+        "--scheduler_type", default="flow-dpm_solver", type=str, choices=["flow-dpm_solver", "flow-euler"]
+    )
+    parser.add_argument("--dump_path", default=None, type=str, required=True, help="Path to the output pipeline.")
+    parser.add_argument("--save_full_pipeline", action="store_true", help="save all the pipelien elemets in one.")
+    parser.add_argument("--dtype", default="fp32", type=str, choices=["fp32", "fp16", "bf16"], help="Weight dtype.")
+
+    args = parser.parse_args()
+
+    model_kwargs = {
+        "SanaMS_1600M_P1_D20": {
+            "num_attention_heads": 70,
+            "attention_head_dim": 32,
+            "num_cross_attention_heads": 20,
+            "cross_attention_head_dim": 112,
+            "cross_attention_dim": 2240,
+            "num_layers": 20,
+        },
+        "SanaMS_600M_P1_D28": {
+            "num_attention_heads": 36,
+            "attention_head_dim": 32,
+            "num_cross_attention_heads": 16,
+            "cross_attention_head_dim": 72,
+            "cross_attention_dim": 1152,
+            "num_layers": 28,
+        },
+    }
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    weight_dtype = DTYPE_MAPPING[args.dtype]
+    variant = VARIANT_MAPPING[args.dtype]
+
+    main(args)
--- a/Show More
+++ b/Show More