Merge branch 'main' into enable-all-gpus

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

.github/workflows/benchmark.yml

@@ -25,7 +25,7 @@ jobs:
       group: aws-g6e-4xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3

.github/workflows/nightly_tests.yml

@@ -61,7 +61,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3
@@ -107,7 +107,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash
@@ -222,7 +222,7 @@ jobs:
       group: aws-g6e-xlarge-plus
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3
@@ -270,7 +270,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-minimum-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash

.github/workflows/pr_tests_gpu.yml

@@ -118,7 +118,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3
@@ -183,7 +183,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash

.github/workflows/push_tests.yml

@@ -64,7 +64,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3
@@ -109,7 +109,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash

.github/workflows/release_tests_fast.yml

@@ -62,7 +62,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
         uses: actions/checkout@v3
@@ -107,7 +107,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash
@@ -163,7 +163,7 @@ jobs:
       group: aws-g4dn-2xlarge
     container:
       image: diffusers/diffusers-pytorch-minimum-cuda
-      options: --shm-size "16gb" --ipc host --gpus 0
+      options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
         shell: bash

.github/workflows/ssh-runner.yml

@@ -31,7 +31,7 @@ jobs:
       group: "${{ github.event.inputs.runner_type }}"
     container:
       image: ${{ github.event.inputs.docker_image }}
-      options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface/diffusers:/mnt/cache/ --gpus 0 --privileged
+      options: --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface/diffusers:/mnt/cache/ --gpus all --privileged
 
     steps:
       - name: Checkout diffusers

docs/source/en/_toctree.yml

@@ -366,6 +366,8 @@
         title: PixArtTransformer2DModel
       - local: api/models/prior_transformer
         title: PriorTransformer
+      - local: api/models/qwenimage_transformer2d
+        title: QwenImageTransformer2DModel
       - local: api/models/sana_transformer2d
         title: SanaTransformer2DModel
       - local: api/models/sd3_transformer2d
@@ -418,6 +420,8 @@
         title: AutoencoderKLMagvit
       - local: api/models/autoencoderkl_mochi
         title: AutoencoderKLMochi
+      - local: api/models/autoencoderkl_qwenimage
+        title: AutoencoderKLQwenImage
       - local: api/models/autoencoder_kl_wan
         title: AutoencoderKLWan
       - local: api/models/consistency_decoder_vae
@@ -554,6 +558,8 @@
       title: PixArt-α
     - local: api/pipelines/pixart_sigma
       title: PixArt-Σ
+    - local: api/pipelines/qwenimage
+      title: QwenImage
     - local: api/pipelines/sana
       title: Sana
     - local: api/pipelines/sana_sprint

docs/source/en/api/models/autoencoderkl_qwenimage.md

@@ -0,0 +1,35 @@
+<!-- Copyright 2025 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. -->
+
+# AutoencoderKLQwenImage
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import AutoencoderKLQwenImage
+
+vae = AutoencoderKLQwenImage.from_pretrained("Qwen/QwenImage-20B", subfolder="vae")
+```
+
+## AutoencoderKLQwenImage
+
+[[autodoc]] AutoencoderKLQwenImage
+    - decode
+    - encode
+    - all
+
+## AutoencoderKLOutput
+
+[[autodoc]] models.autoencoders.autoencoder_kl.AutoencoderKLOutput
+
+## DecoderOutput
+
+[[autodoc]] models.autoencoders.vae.DecoderOutput

docs/source/en/api/models/qwenimage_transformer2d.md

@@ -0,0 +1,28 @@
+<!-- Copyright 2025 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. -->
+
+# QwenImageTransformer2DModel
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import QwenImageTransformer2DModel
+
+transformer = QwenImageTransformer2DModel.from_pretrained("Qwen/QwenImage-20B", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
+## QwenImageTransformer2DModel
+
+[[autodoc]] QwenImageTransformer2DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput

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

@@ -0,0 +1,33 @@
+<!-- Copyright 2025 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. -->
+
+# QwenImage
+
+<!-- TODO: update this section when model is out -->
+
+<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.
+
+</Tip>
+
+## QwenImagePipeline
+
+[[autodoc]] QwenImagePipeline
+  - all
+  - __call__
+
+## QwenImagePipeline
+
+[[autodoc]] pipelines.qwenimage.pipeline_output.QwenImagePipelineOutput

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

@@ -29,6 +29,7 @@
 You can find all the original Wan2.1 checkpoints under the [Wan-AI](https://huggingface.co/Wan-AI) organization.
 
 The following Wan models are supported in Diffusers:
+
 - [Wan 2.1 T2V 1.3B](https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B-Diffusers)
 - [Wan 2.1 T2V 14B](https://huggingface.co/Wan-AI/Wan2.1-T2V-14B-Diffusers)
 - [Wan 2.1 I2V 14B - 480P](https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P-Diffusers)
@@ -36,6 +37,9 @@ The following Wan models are supported in Diffusers:
 - [Wan 2.1 FLF2V 14B - 720P](https://huggingface.co/Wan-AI/Wan2.1-FLF2V-14B-720P-diffusers)
 - [Wan 2.1 VACE 1.3B](https://huggingface.co/Wan-AI/Wan2.1-VACE-1.3B-diffusers)
 - [Wan 2.1 VACE 14B](https://huggingface.co/Wan-AI/Wan2.1-VACE-14B-diffusers)
+- [Wan 2.2 T2V 14B](https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B-Diffusers)
+- [Wan 2.2 I2V 14B](https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B-Diffusers)
+- [Wan 2.2 TI2V 5B](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B-Diffusers)
 
 > [!TIP]
 > Click on the Wan2.1 models in the right sidebar for more examples of video generation.
@@ -327,6 +331,8 @@ The general rule of thumb to keep in mind when preparing inputs for the VACE pip
 
 - Try lower `shift` values (`2.0` to `5.0`) for lower resolution videos and higher `shift` values (`7.0` to `12.0`) for higher resolution images.
 
+- Wan 2.1 and 2.2 support using [LightX2V LoRAs](https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v) to speed up inference. Using them on Wan 2.2 is slightly more involed. Refer to [this code snippet](https://github.com/huggingface/diffusers/pull/12040#issuecomment-3144185272) to learn more.
+
 ## WanPipeline
 
 [[autodoc]] WanPipeline

src/diffusers/__init__.py

@@ -174,6 +174,7 @@ else:
             "AutoencoderKLLTXVideo",
             "AutoencoderKLMagvit",
             "AutoencoderKLMochi",
+            "AutoencoderKLQwenImage",
             "AutoencoderKLTemporalDecoder",
             "AutoencoderKLWan",
             "AutoencoderOobleck",
@@ -215,6 +216,7 @@ else:
             "OmniGenTransformer2DModel",
             "PixArtTransformer2DModel",
             "PriorTransformer",
+            "QwenImageTransformer2DModel",
             "SanaControlNetModel",
             "SanaTransformer2DModel",
             "SD3ControlNetModel",
@@ -486,6 +488,7 @@ else:
             "PixArtAlphaPipeline",
             "PixArtSigmaPAGPipeline",
             "PixArtSigmaPipeline",
+            "QwenImagePipeline",
             "ReduxImageEncoder",
             "SanaControlNetPipeline",
             "SanaPAGPipeline",
@@ -832,6 +835,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             AutoencoderKLLTXVideo,
             AutoencoderKLMagvit,
             AutoencoderKLMochi,
+            AutoencoderKLQwenImage,
             AutoencoderKLTemporalDecoder,
             AutoencoderKLWan,
             AutoencoderOobleck,
@@ -873,6 +877,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             OmniGenTransformer2DModel,
             PixArtTransformer2DModel,
             PriorTransformer,
+            QwenImageTransformer2DModel,
             SanaControlNetModel,
             SanaTransformer2DModel,
             SD3ControlNetModel,
@@ -1119,6 +1124,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             PixArtAlphaPipeline,
             PixArtSigmaPAGPipeline,
             PixArtSigmaPipeline,
+            QwenImagePipeline,
             ReduxImageEncoder,
             SanaControlNetPipeline,
             SanaPAGPipeline,

src/diffusers/hooks/_helpers.py

@@ -153,6 +153,7 @@ def _register_transformer_blocks_metadata():
     )
     from ..models.transformers.transformer_ltx import LTXVideoTransformerBlock
     from ..models.transformers.transformer_mochi import MochiTransformerBlock
+    from ..models.transformers.transformer_qwenimage import QwenImageTransformerBlock
     from ..models.transformers.transformer_wan import WanTransformerBlock
 
     # BasicTransformerBlock
@@ -255,6 +256,15 @@ def _register_transformer_blocks_metadata():
         ),
     )
 
+    # QwenImage
+    TransformerBlockRegistry.register(
+        model_class=QwenImageTransformerBlock,
+        metadata=TransformerBlockMetadata(
+            return_hidden_states_index=1,
+            return_encoder_hidden_states_index=0,
+        ),
+    )
+
 
 # fmt: off
 def _skip_attention___ret___hidden_states(self, *args, **kwargs):

src/diffusers/loaders/lora_conversion_utils.py

@@ -1974,6 +1974,10 @@ def _convert_non_diffusers_wan_lora_to_diffusers(state_dict):
             converted_key = f"condition_embedder.image_embedder.{img_ours}.lora_B.weight"
             if original_key in original_state_dict:
                 converted_state_dict[converted_key] = original_state_dict.pop(original_key)
+                bias_key_theirs = original_key.removesuffix(f".{lora_up_key}.weight") + ".diff_b"
+                if bias_key_theirs in original_state_dict:
+                    bias_key = converted_key.removesuffix(".weight") + ".bias"
+                    converted_state_dict[bias_key] = original_state_dict.pop(bias_key_theirs)
 
     if len(original_state_dict) > 0:
         diff = all(".diff" in k for k in original_state_dict)

src/diffusers/models/__init__.py

@@ -38,6 +38,7 @@ if is_torch_available():
     _import_structure["autoencoders.autoencoder_kl_ltx"] = ["AutoencoderKLLTXVideo"]
     _import_structure["autoencoders.autoencoder_kl_magvit"] = ["AutoencoderKLMagvit"]
     _import_structure["autoencoders.autoencoder_kl_mochi"] = ["AutoencoderKLMochi"]
+    _import_structure["autoencoders.autoencoder_kl_qwenimage"] = ["AutoencoderKLQwenImage"]
     _import_structure["autoencoders.autoencoder_kl_temporal_decoder"] = ["AutoencoderKLTemporalDecoder"]
     _import_structure["autoencoders.autoencoder_kl_wan"] = ["AutoencoderKLWan"]
     _import_structure["autoencoders.autoencoder_oobleck"] = ["AutoencoderOobleck"]
@@ -88,6 +89,7 @@ if is_torch_available():
     _import_structure["transformers.transformer_lumina2"] = ["Lumina2Transformer2DModel"]
     _import_structure["transformers.transformer_mochi"] = ["MochiTransformer3DModel"]
     _import_structure["transformers.transformer_omnigen"] = ["OmniGenTransformer2DModel"]
+    _import_structure["transformers.transformer_qwenimage"] = ["QwenImageTransformer2DModel"]
     _import_structure["transformers.transformer_sd3"] = ["SD3Transformer2DModel"]
     _import_structure["transformers.transformer_skyreels_v2"] = ["SkyReelsV2Transformer3DModel"]
     _import_structure["transformers.transformer_temporal"] = ["TransformerTemporalModel"]
@@ -126,6 +128,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             AutoencoderKLLTXVideo,
             AutoencoderKLMagvit,
             AutoencoderKLMochi,
+            AutoencoderKLQwenImage,
             AutoencoderKLTemporalDecoder,
             AutoencoderKLWan,
             AutoencoderOobleck,
@@ -177,6 +180,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             OmniGenTransformer2DModel,
             PixArtTransformer2DModel,
             PriorTransformer,
+            QwenImageTransformer2DModel,
             SanaTransformer2DModel,
             SD3Transformer2DModel,
             SkyReelsV2Transformer3DModel,

src/diffusers/models/autoencoders/__init__.py

@@ -8,6 +8,7 @@ from .autoencoder_kl_hunyuan_video import AutoencoderKLHunyuanVideo
 from .autoencoder_kl_ltx import AutoencoderKLLTXVideo
 from .autoencoder_kl_magvit import AutoencoderKLMagvit
 from .autoencoder_kl_mochi import AutoencoderKLMochi
+from .autoencoder_kl_qwenimage import AutoencoderKLQwenImage
 from .autoencoder_kl_temporal_decoder import AutoencoderKLTemporalDecoder
 from .autoencoder_kl_wan import AutoencoderKLWan
 from .autoencoder_oobleck import AutoencoderOobleck

src/diffusers/models/autoencoders/autoencoder_kl.py

@@ -90,7 +90,7 @@ class AutoencoderKL(ModelMixin, ConfigMixin, FromOriginalModelMixin, PeftAdapter
         shift_factor: Optional[float] = None,
         latents_mean: Optional[Tuple[float]] = None,
         latents_std: Optional[Tuple[float]] = None,
-        force_upcast: float = True,
+        force_upcast: bool = True,
         use_quant_conv: bool = True,
         use_post_quant_conv: bool = True,
         mid_block_add_attention: bool = True,

src/diffusers/models/autoencoders/autoencoder_kl_cosmos.py

@@ -168,7 +168,9 @@ class CosmosPatchEmbed3d(nn.Module):
         batch_size, num_channels, num_frames, height, width = hidden_states.shape
         p = self.patch_size
 
-        hidden_states = torch.reshape(batch_size, num_channels, num_frames // p, p, height // p, p, width // p, p)
+        hidden_states = hidden_states.reshape(
+            batch_size, num_channels, num_frames // p, p, height // p, p, width // p, p
+        )
         hidden_states = hidden_states.permute(0, 1, 3, 5, 7, 2, 4, 6).flatten(1, 4).contiguous()
         return hidden_states
 

src/diffusers/models/autoencoders/autoencoder_kl_wan.py

@@ -913,38 +913,21 @@ def patchify(x, patch_size):
     if patch_size == 1:
         return x
 
-    if x.dim() == 4:
-        # x shape: [batch_size, channels, height, width]
-        batch_size, channels, height, width = x.shape
-
-        # Ensure height and width are divisible by patch_size
-        if height % patch_size != 0 or width % patch_size != 0:
-            raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")
-
-        # Reshape to [batch_size, channels, height//patch_size, patch_size, width//patch_size, patch_size]
-        x = x.view(batch_size, channels, height // patch_size, patch_size, width // patch_size, patch_size)
-
-        # Rearrange to [batch_size, channels * patch_size * patch_size, height//patch_size, width//patch_size]
-        x = x.permute(0, 1, 3, 5, 2, 4).contiguous()
-        x = x.view(batch_size, channels * patch_size * patch_size, height // patch_size, width // patch_size)
-
-    elif x.dim() == 5:
-        # x shape: [batch_size, channels, frames, height, width]
-        batch_size, channels, frames, height, width = x.shape
-
-        # Ensure height and width are divisible by patch_size
-        if height % patch_size != 0 or width % patch_size != 0:
-            raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")
-
-        # Reshape to [batch_size, channels, frames, height//patch_size, patch_size, width//patch_size, patch_size]
-        x = x.view(batch_size, channels, frames, height // patch_size, patch_size, width // patch_size, patch_size)
-
-        # Rearrange to [batch_size, channels * patch_size * patch_size, frames, height//patch_size, width//patch_size]
-        x = x.permute(0, 1, 4, 6, 2, 3, 5).contiguous()
-        x = x.view(batch_size, channels * patch_size * patch_size, frames, height // patch_size, width // patch_size)
-
-    else:
+    if x.dim() != 5:
         raise ValueError(f"Invalid input shape: {x.shape}")
+    # x shape: [batch_size, channels, frames, height, width]
+    batch_size, channels, frames, height, width = x.shape
+
+    # Ensure height and width are divisible by patch_size
+    if height % patch_size != 0 or width % patch_size != 0:
+        raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")
+
+    # Reshape to [batch_size, channels, frames, height//patch_size, patch_size, width//patch_size, patch_size]
+    x = x.view(batch_size, channels, frames, height // patch_size, patch_size, width // patch_size, patch_size)
+
+    # Rearrange to [batch_size, channels * patch_size * patch_size, frames, height//patch_size, width//patch_size]
+    x = x.permute(0, 1, 6, 4, 2, 3, 5).contiguous()
+    x = x.view(batch_size, channels * patch_size * patch_size, frames, height // patch_size, width // patch_size)
 
     return x
 
@@ -953,29 +936,18 @@ def unpatchify(x, patch_size):
     if patch_size == 1:
         return x
 
-    if x.dim() == 4:
-        # x shape: [b, (c * patch_size * patch_size), h, w]
-        batch_size, c_patches, height, width = x.shape
-        channels = c_patches // (patch_size * patch_size)
+    if x.dim() != 5:
+        raise ValueError(f"Invalid input shape: {x.shape}")
+    # x shape: [batch_size, (channels * patch_size * patch_size), frame, height, width]
+    batch_size, c_patches, frames, height, width = x.shape
+    channels = c_patches // (patch_size * patch_size)
 
-        # Reshape to [b, c, patch_size, patch_size, h, w]
-        x = x.view(batch_size, channels, patch_size, patch_size, height, width)
+    # Reshape to [b, c, patch_size, patch_size, f, h, w]
+    x = x.view(batch_size, channels, patch_size, patch_size, frames, height, width)
 
-        # Rearrange to [b, c, h * patch_size, w * patch_size]
-        x = x.permute(0, 1, 4, 2, 5, 3).contiguous()
-        x = x.view(batch_size, channels, height * patch_size, width * patch_size)
-
-    elif x.dim() == 5:
-        # x shape: [batch_size, (channels * patch_size * patch_size), frame, height, width]
-        batch_size, c_patches, frames, height, width = x.shape
-        channels = c_patches // (patch_size * patch_size)
-
-        # Reshape to [b, c, patch_size, patch_size, f, h, w]
-        x = x.view(batch_size, channels, patch_size, patch_size, frames, height, width)
-
-        # Rearrange to [b, c, f, h * patch_size, w * patch_size]
-        x = x.permute(0, 1, 4, 5, 2, 6, 3).contiguous()
-        x = x.view(batch_size, channels, frames, height * patch_size, width * patch_size)
+    # Rearrange to [b, c, f, h * patch_size, w * patch_size]
+    x = x.permute(0, 1, 4, 5, 3, 6, 2).contiguous()
+    x = x.view(batch_size, channels, frames, height * patch_size, width * patch_size)
 
     return x
 
@@ -1044,7 +1016,6 @@ class AutoencoderKLWan(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         patch_size: Optional[int] = None,
         scale_factor_temporal: Optional[int] = 4,
         scale_factor_spatial: Optional[int] = 8,
-        clip_output: bool = True,
     ) -> None:
         super().__init__()
 
@@ -1244,10 +1215,11 @@ class AutoencoderKLWan(ModelMixin, ConfigMixin, FromOriginalModelMixin):
                 out_ = self.decoder(x[:, :, i : i + 1, :, :], feat_cache=self._feat_map, feat_idx=self._conv_idx)
                 out = torch.cat([out, out_], 2)
 
-        if self.config.clip_output:
-            out = torch.clamp(out, min=-1.0, max=1.0)
         if self.config.patch_size is not None:
             out = unpatchify(out, patch_size=self.config.patch_size)
+
+        out = torch.clamp(out, min=-1.0, max=1.0)
+
         self.clear_cache()
         if not return_dict:
             return (out,)

src/diffusers/models/transformers/__init__.py

@@ -30,6 +30,7 @@ if is_torch_available():
     from .transformer_lumina2 import Lumina2Transformer2DModel
     from .transformer_mochi import MochiTransformer3DModel
     from .transformer_omnigen import OmniGenTransformer2DModel
+    from .transformer_qwenimage import QwenImageTransformer2DModel
     from .transformer_sd3 import SD3Transformer2DModel
     from .transformer_skyreels_v2 import SkyReelsV2Transformer3DModel
     from .transformer_temporal import TransformerTemporalModel

src/diffusers/models/transformers/transformer_qwenimage.py

@@ -0,0 +1,628 @@
+# Copyright 2025 Qwen-Image Team, The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import math
+from typing import Any, Dict, List, Optional, Tuple, Union
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from ...configuration_utils import ConfigMixin, register_to_config
+from ...loaders import FromOriginalModelMixin, PeftAdapterMixin
+from ...utils import USE_PEFT_BACKEND, logging, scale_lora_layers, unscale_lora_layers
+from ...utils.torch_utils import maybe_allow_in_graph
+from ..attention import FeedForward
+from ..attention_dispatch import dispatch_attention_fn
+from ..attention_processor import Attention
+from ..cache_utils import CacheMixin
+from ..embeddings import TimestepEmbedding, Timesteps
+from ..modeling_outputs import Transformer2DModelOutput
+from ..modeling_utils import ModelMixin
+from ..normalization import AdaLayerNormContinuous, RMSNorm
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+def get_timestep_embedding(
+    timesteps: torch.Tensor,
+    embedding_dim: int,
+    flip_sin_to_cos: bool = False,
+    downscale_freq_shift: float = 1,
+    scale: float = 1,
+    max_period: int = 10000,
+) -> torch.Tensor:
+    """
+    This matches the implementation in Denoising Diffusion Probabilistic Models: Create sinusoidal timestep embeddings.
+
+    Args
+        timesteps (torch.Tensor):
+            a 1-D Tensor of N indices, one per batch element. These may be fractional.
+        embedding_dim (int):
+            the dimension of the output.
+        flip_sin_to_cos (bool):
+            Whether the embedding order should be `cos, sin` (if True) or `sin, cos` (if False)
+        downscale_freq_shift (float):
+            Controls the delta between frequencies between dimensions
+        scale (float):
+            Scaling factor applied to the embeddings.
+        max_period (int):
+            Controls the maximum frequency of the embeddings
+    Returns
+        torch.Tensor: an [N x dim] Tensor of positional embeddings.
+    """
+    assert len(timesteps.shape) == 1, "Timesteps should be a 1d-array"
+
+    half_dim = embedding_dim // 2
+    exponent = -math.log(max_period) * torch.arange(
+        start=0, end=half_dim, dtype=torch.float32, device=timesteps.device
+    )
+    exponent = exponent / (half_dim - downscale_freq_shift)
+
+    emb = torch.exp(exponent).to(timesteps.dtype)
+    emb = timesteps[:, None].float() * emb[None, :]
+
+    # scale embeddings
+    emb = scale * emb
+
+    # concat sine and cosine embeddings
+    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=-1)
+
+    # flip sine and cosine embeddings
+    if flip_sin_to_cos:
+        emb = torch.cat([emb[:, half_dim:], emb[:, :half_dim]], dim=-1)
+
+    # zero pad
+    if embedding_dim % 2 == 1:
+        emb = torch.nn.functional.pad(emb, (0, 1, 0, 0))
+    return emb
+
+
+def apply_rotary_emb_qwen(
+    x: torch.Tensor,
+    freqs_cis: Union[torch.Tensor, Tuple[torch.Tensor]],
+    use_real: bool = True,
+    use_real_unbind_dim: int = -1,
+) -> Tuple[torch.Tensor, torch.Tensor]:
+    """
+    Apply rotary embeddings to input tensors using the given frequency tensor. This function applies rotary embeddings
+    to the given query or key 'x' tensors using the provided frequency tensor 'freqs_cis'. The input tensors are
+    reshaped as complex numbers, and the frequency tensor is reshaped for broadcasting compatibility. The resulting
+    tensors contain rotary embeddings and are returned as real tensors.
+
+    Args:
+        x (`torch.Tensor`):
+            Query or key tensor to apply rotary embeddings. [B, S, H, D] xk (torch.Tensor): Key tensor to apply
+        freqs_cis (`Tuple[torch.Tensor]`): Precomputed frequency tensor for complex exponentials. ([S, D], [S, D],)
+
+    Returns:
+        Tuple[torch.Tensor, torch.Tensor]: Tuple of modified query tensor and key tensor with rotary embeddings.
+    """
+    if use_real:
+        cos, sin = freqs_cis  # [S, D]
+        cos = cos[None, None]
+        sin = sin[None, None]
+        cos, sin = cos.to(x.device), sin.to(x.device)
+
+        if use_real_unbind_dim == -1:
+            # Used for flux, cogvideox, hunyuan-dit
+            x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1)  # [B, S, H, D//2]
+            x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(3)
+        elif use_real_unbind_dim == -2:
+            # Used for Stable Audio, OmniGen, CogView4 and Cosmos
+            x_real, x_imag = x.reshape(*x.shape[:-1], 2, -1).unbind(-2)  # [B, S, H, D//2]
+            x_rotated = torch.cat([-x_imag, x_real], dim=-1)
+        else:
+            raise ValueError(f"`use_real_unbind_dim={use_real_unbind_dim}` but should be -1 or -2.")
+
+        out = (x.float() * cos + x_rotated.float() * sin).to(x.dtype)
+
+        return out
+    else:
+        x_rotated = torch.view_as_complex(x.float().reshape(*x.shape[:-1], -1, 2))
+        freqs_cis = freqs_cis.unsqueeze(1)
+        x_out = torch.view_as_real(x_rotated * freqs_cis).flatten(3)
+
+        return x_out.type_as(x)
+
+
+class QwenTimestepProjEmbeddings(nn.Module):
+    def __init__(self, embedding_dim):
+        super().__init__()
+
+        self.time_proj = Timesteps(num_channels=256, flip_sin_to_cos=True, downscale_freq_shift=0, scale=1000)
+        self.timestep_embedder = TimestepEmbedding(in_channels=256, time_embed_dim=embedding_dim)
+
+    def forward(self, timestep, hidden_states):
+        timesteps_proj = self.time_proj(timestep)
+        timesteps_emb = self.timestep_embedder(timesteps_proj.to(dtype=hidden_states.dtype))  # (N, D)
+
+        conditioning = timesteps_emb
+
+        return conditioning
+
+
+class QwenEmbedRope(nn.Module):
+    def __init__(self, theta: int, axes_dim: List[int], scale_rope=False):
+        super().__init__()
+        self.theta = theta
+        self.axes_dim = axes_dim
+        pos_index = torch.arange(1024)
+        neg_index = torch.arange(1024).flip(0) * -1 - 1
+        self.pos_freqs = torch.cat(
+            [
+                self.rope_params(pos_index, self.axes_dim[0], self.theta),
+                self.rope_params(pos_index, self.axes_dim[1], self.theta),
+                self.rope_params(pos_index, self.axes_dim[2], self.theta),
+            ],
+            dim=1,
+        )
+        self.neg_freqs = torch.cat(
+            [
+                self.rope_params(neg_index, self.axes_dim[0], self.theta),
+                self.rope_params(neg_index, self.axes_dim[1], self.theta),
+                self.rope_params(neg_index, self.axes_dim[2], self.theta),
+            ],
+            dim=1,
+        )
+        self.rope_cache = {}
+
+        # 是否使用 scale rope
+        self.scale_rope = scale_rope
+
+    def rope_params(self, index, dim, theta=10000):
+        """
+        Args:
+            index: [0, 1, 2, 3] 1D Tensor representing the position index of the token
+        """
+        assert dim % 2 == 0
+        freqs = torch.outer(index, 1.0 / torch.pow(theta, torch.arange(0, dim, 2).to(torch.float32).div(dim)))
+        freqs = torch.polar(torch.ones_like(freqs), freqs)
+        return freqs
+
+    def forward(self, video_fhw, txt_seq_lens, device):
+        """
+        Args: video_fhw: [frame, height, width] a list of 3 integers representing the shape of the video Args:
+        txt_length: [bs] a list of 1 integers representing the length of the text
+        """
+        if self.pos_freqs.device != device:
+            self.pos_freqs = self.pos_freqs.to(device)
+            self.neg_freqs = self.neg_freqs.to(device)
+
+        if isinstance(video_fhw, list):
+            video_fhw = video_fhw[0]
+        frame, height, width = video_fhw
+        rope_key = f"{frame}_{height}_{width}"
+
+        if rope_key not in self.rope_cache:
+            seq_lens = frame * height * width
+            freqs_pos = self.pos_freqs.split([x // 2 for x in self.axes_dim], dim=1)
+            freqs_neg = self.neg_freqs.split([x // 2 for x in self.axes_dim], dim=1)
+            freqs_frame = freqs_pos[0][:frame].view(frame, 1, 1, -1).expand(frame, height, width, -1)
+            if self.scale_rope:
+                freqs_height = torch.cat([freqs_neg[1][-(height - height // 2) :], freqs_pos[1][: height // 2]], dim=0)
+                freqs_height = freqs_height.view(1, height, 1, -1).expand(frame, height, width, -1)
+                freqs_width = torch.cat([freqs_neg[2][-(width - width // 2) :], freqs_pos[2][: width // 2]], dim=0)
+                freqs_width = freqs_width.view(1, 1, width, -1).expand(frame, height, width, -1)
+
+            else:
+                freqs_height = freqs_pos[1][:height].view(1, height, 1, -1).expand(frame, height, width, -1)
+                freqs_width = freqs_pos[2][:width].view(1, 1, width, -1).expand(frame, height, width, -1)
+
+            freqs = torch.cat([freqs_frame, freqs_height, freqs_width], dim=-1).reshape(seq_lens, -1)
+            self.rope_cache[rope_key] = freqs.clone().contiguous()
+        vid_freqs = self.rope_cache[rope_key]
+
+        if self.scale_rope:
+            max_vid_index = max(height // 2, width // 2)
+        else:
+            max_vid_index = max(height, width)
+
+        max_len = max(txt_seq_lens)
+        txt_freqs = self.pos_freqs[max_vid_index : max_vid_index + max_len, ...]
+
+        return vid_freqs, txt_freqs
+
+
+class QwenDoubleStreamAttnProcessor2_0:
+    """
+    Attention processor for Qwen double-stream architecture, matching DoubleStreamLayerMegatron logic. This processor
+    implements joint attention computation where text and image streams are processed together.
+    """
+
+    _attention_backend = None
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError(
+                "QwenDoubleStreamAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
+            )
+
+    def __call__(
+        self,
+        attn: Attention,
+        hidden_states: torch.FloatTensor,  # Image stream
+        encoder_hidden_states: torch.FloatTensor = None,  # Text stream
+        encoder_hidden_states_mask: torch.FloatTensor = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        image_rotary_emb: Optional[torch.Tensor] = None,
+    ) -> torch.FloatTensor:
+        if encoder_hidden_states is None:
+            raise ValueError("QwenDoubleStreamAttnProcessor2_0 requires encoder_hidden_states (text stream)")
+
+        seq_txt = encoder_hidden_states.shape[1]
+
+        # Compute QKV for image stream (sample projections)
+        img_query = attn.to_q(hidden_states)
+        img_key = attn.to_k(hidden_states)
+        img_value = attn.to_v(hidden_states)
+
+        # Compute QKV for text stream (context projections)
+        txt_query = attn.add_q_proj(encoder_hidden_states)
+        txt_key = attn.add_k_proj(encoder_hidden_states)
+        txt_value = attn.add_v_proj(encoder_hidden_states)
+
+        # Reshape for multi-head attention
+        img_query = img_query.unflatten(-1, (attn.heads, -1))
+        img_key = img_key.unflatten(-1, (attn.heads, -1))
+        img_value = img_value.unflatten(-1, (attn.heads, -1))
+
+        txt_query = txt_query.unflatten(-1, (attn.heads, -1))
+        txt_key = txt_key.unflatten(-1, (attn.heads, -1))
+        txt_value = txt_value.unflatten(-1, (attn.heads, -1))
+
+        # Apply QK normalization
+        if attn.norm_q is not None:
+            img_query = attn.norm_q(img_query)
+        if attn.norm_k is not None:
+            img_key = attn.norm_k(img_key)
+        if attn.norm_added_q is not None:
+            txt_query = attn.norm_added_q(txt_query)
+        if attn.norm_added_k is not None:
+            txt_key = attn.norm_added_k(txt_key)
+
+        # Apply RoPE
+        if image_rotary_emb is not None:
+            img_freqs, txt_freqs = image_rotary_emb
+            img_query = apply_rotary_emb_qwen(img_query, img_freqs, use_real=False)
+            img_key = apply_rotary_emb_qwen(img_key, img_freqs, use_real=False)
+            txt_query = apply_rotary_emb_qwen(txt_query, txt_freqs, use_real=False)
+            txt_key = apply_rotary_emb_qwen(txt_key, txt_freqs, use_real=False)
+
+        # Concatenate for joint attention
+        # Order: [text, image]
+        joint_query = torch.cat([txt_query, img_query], dim=1)
+        joint_key = torch.cat([txt_key, img_key], dim=1)
+        joint_value = torch.cat([txt_value, img_value], dim=1)
+
+        # Compute joint attention
+        joint_hidden_states = dispatch_attention_fn(
+            joint_query,
+            joint_key,
+            joint_value,
+            attn_mask=attention_mask,
+            dropout_p=0.0,
+            is_causal=False,
+            backend=self._attention_backend,
+        )
+
+        # Reshape back
+        joint_hidden_states = joint_hidden_states.flatten(2, 3)
+        joint_hidden_states = joint_hidden_states.to(joint_query.dtype)
+
+        # Split attention outputs back
+        txt_attn_output = joint_hidden_states[:, :seq_txt, :]  # Text part
+        img_attn_output = joint_hidden_states[:, seq_txt:, :]  # Image part
+
+        # Apply output projections
+        img_attn_output = attn.to_out[0](img_attn_output)
+        if len(attn.to_out) > 1:
+            img_attn_output = attn.to_out[1](img_attn_output)  # dropout
+
+        txt_attn_output = attn.to_add_out(txt_attn_output)
+
+        return img_attn_output, txt_attn_output
+
+
+@maybe_allow_in_graph
+class QwenImageTransformerBlock(nn.Module):
+    def __init__(
+        self, dim: int, num_attention_heads: int, attention_head_dim: int, qk_norm: str = "rms_norm", eps: float = 1e-6
+    ):
+        super().__init__()
+
+        self.dim = dim
+        self.num_attention_heads = num_attention_heads
+        self.attention_head_dim = attention_head_dim
+
+        # Image processing modules
+        self.img_mod = nn.Sequential(
+            nn.SiLU(),
+            nn.Linear(dim, 6 * dim, bias=True),  # For scale, shift, gate for norm1 and norm2
+        )
+        self.img_norm1 = nn.LayerNorm(dim, elementwise_affine=False, eps=eps)
+        self.attn = Attention(
+            query_dim=dim,
+            cross_attention_dim=None,  # Enable cross attention for joint computation
+            added_kv_proj_dim=dim,  # Enable added KV projections for text stream
+            dim_head=attention_head_dim,
+            heads=num_attention_heads,
+            out_dim=dim,
+            context_pre_only=False,
+            bias=True,
+            processor=QwenDoubleStreamAttnProcessor2_0(),
+            qk_norm=qk_norm,
+            eps=eps,
+        )
+        self.img_norm2 = nn.LayerNorm(dim, elementwise_affine=False, eps=eps)
+        self.img_mlp = FeedForward(dim=dim, dim_out=dim, activation_fn="gelu-approximate")
+
+        # Text processing modules
+        self.txt_mod = nn.Sequential(
+            nn.SiLU(),
+            nn.Linear(dim, 6 * dim, bias=True),  # For scale, shift, gate for norm1 and norm2
+        )
+        self.txt_norm1 = nn.LayerNorm(dim, elementwise_affine=False, eps=eps)
+        # Text doesn't need separate attention - it's handled by img_attn joint computation
+        self.txt_norm2 = nn.LayerNorm(dim, elementwise_affine=False, eps=eps)
+        self.txt_mlp = FeedForward(dim=dim, dim_out=dim, activation_fn="gelu-approximate")
+
+    def _modulate(self, x, mod_params):
+        """Apply modulation to input tensor"""
+        shift, scale, gate = mod_params.chunk(3, dim=-1)
+        return x * (1 + scale.unsqueeze(1)) + shift.unsqueeze(1), gate.unsqueeze(1)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+        encoder_hidden_states_mask: torch.Tensor,
+        temb: torch.Tensor,
+        image_rotary_emb: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        # Get modulation parameters for both streams
+        img_mod_params = self.img_mod(temb)  # [B, 6*dim]
+        txt_mod_params = self.txt_mod(temb)  # [B, 6*dim]
+
+        # Split modulation parameters for norm1 and norm2
+        img_mod1, img_mod2 = img_mod_params.chunk(2, dim=-1)  # Each [B, 3*dim]
+        txt_mod1, txt_mod2 = txt_mod_params.chunk(2, dim=-1)  # Each [B, 3*dim]
+
+        # Process image stream - norm1 + modulation
+        img_normed = self.img_norm1(hidden_states)
+        img_modulated, img_gate1 = self._modulate(img_normed, img_mod1)
+
+        # Process text stream - norm1 + modulation
+        txt_normed = self.txt_norm1(encoder_hidden_states)
+        txt_modulated, txt_gate1 = self._modulate(txt_normed, txt_mod1)
+
+        # Use QwenAttnProcessor2_0 for joint attention computation
+        # This directly implements the DoubleStreamLayerMegatron logic:
+        # 1. Computes QKV for both streams
+        # 2. Applies QK normalization and RoPE
+        # 3. Concatenates and runs joint attention
+        # 4. Splits results back to separate streams
+        joint_attention_kwargs = joint_attention_kwargs or {}
+        attn_output = self.attn(
+            hidden_states=img_modulated,  # Image stream (will be processed as "sample")
+            encoder_hidden_states=txt_modulated,  # Text stream (will be processed as "context")
+            encoder_hidden_states_mask=encoder_hidden_states_mask,
+            image_rotary_emb=image_rotary_emb,
+            **joint_attention_kwargs,
+        )
+
+        # QwenAttnProcessor2_0 returns (img_output, txt_output) when encoder_hidden_states is provided
+        img_attn_output, txt_attn_output = attn_output
+
+        # Apply attention gates and add residual (like in Megatron)
+        hidden_states = hidden_states + img_gate1 * img_attn_output
+        encoder_hidden_states = encoder_hidden_states + txt_gate1 * txt_attn_output
+
+        # Process image stream - norm2 + MLP
+        img_normed2 = self.img_norm2(hidden_states)
+        img_modulated2, img_gate2 = self._modulate(img_normed2, img_mod2)
+        img_mlp_output = self.img_mlp(img_modulated2)
+        hidden_states = hidden_states + img_gate2 * img_mlp_output
+
+        # Process text stream - norm2 + MLP
+        txt_normed2 = self.txt_norm2(encoder_hidden_states)
+        txt_modulated2, txt_gate2 = self._modulate(txt_normed2, txt_mod2)
+        txt_mlp_output = self.txt_mlp(txt_modulated2)
+        encoder_hidden_states = encoder_hidden_states + txt_gate2 * txt_mlp_output
+
+        # Clip to prevent overflow for fp16
+        if encoder_hidden_states.dtype == torch.float16:
+            encoder_hidden_states = encoder_hidden_states.clip(-65504, 65504)
+        if hidden_states.dtype == torch.float16:
+            hidden_states = hidden_states.clip(-65504, 65504)
+
+        return encoder_hidden_states, hidden_states
+
+
+class QwenImageTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOriginalModelMixin, CacheMixin):
+    """
+    The Transformer model introduced in Qwen.
+
+    Args:
+        patch_size (`int`, defaults to `2`):
+            Patch size to turn the input data into small patches.
+        in_channels (`int`, defaults to `64`):
+            The number of channels in the input.
+        out_channels (`int`, *optional*, defaults to `None`):
+            The number of channels in the output. If not specified, it defaults to `in_channels`.
+        num_layers (`int`, defaults to `60`):
+            The number of layers of dual stream DiT blocks to use.
+        attention_head_dim (`int`, defaults to `128`):
+            The number of dimensions to use for each attention head.
+        num_attention_heads (`int`, defaults to `24`):
+            The number of attention heads to use.
+        joint_attention_dim (`int`, defaults to `3584`):
+            The number of dimensions to use for the joint attention (embedding/channel dimension of
+            `encoder_hidden_states`).
+        guidance_embeds (`bool`, defaults to `False`):
+            Whether to use guidance embeddings for guidance-distilled variant of the model.
+        axes_dims_rope (`Tuple[int]`, defaults to `(16, 56, 56)`):
+            The dimensions to use for the rotary positional embeddings.
+    """
+
+    _supports_gradient_checkpointing = True
+    _no_split_modules = ["QwenImageTransformerBlock"]
+    _skip_layerwise_casting_patterns = ["pos_embed", "norm"]
+
+    @register_to_config
+    def __init__(
+        self,
+        patch_size: int = 2,
+        in_channels: int = 64,
+        out_channels: Optional[int] = 16,
+        num_layers: int = 60,
+        attention_head_dim: int = 128,
+        num_attention_heads: int = 24,
+        joint_attention_dim: int = 3584,
+        guidance_embeds: bool = False,  # TODO: this should probably be removed
+        axes_dims_rope: Tuple[int, int, int] = (16, 56, 56),
+    ):
+        super().__init__()
+        self.out_channels = out_channels or in_channels
+        self.inner_dim = num_attention_heads * attention_head_dim
+
+        self.pos_embed = QwenEmbedRope(theta=10000, axes_dim=list(axes_dims_rope), scale_rope=True)
+
+        self.time_text_embed = QwenTimestepProjEmbeddings(embedding_dim=self.inner_dim)
+
+        self.txt_norm = RMSNorm(joint_attention_dim, eps=1e-6)
+
+        self.img_in = nn.Linear(in_channels, self.inner_dim)
+        self.txt_in = nn.Linear(joint_attention_dim, self.inner_dim)
+
+        self.transformer_blocks = nn.ModuleList(
+            [
+                QwenImageTransformerBlock(
+                    dim=self.inner_dim,
+                    num_attention_heads=num_attention_heads,
+                    attention_head_dim=attention_head_dim,
+                )
+                for _ in range(num_layers)
+            ]
+        )
+
+        self.norm_out = AdaLayerNormContinuous(self.inner_dim, self.inner_dim, elementwise_affine=False, eps=1e-6)
+        self.proj_out = nn.Linear(self.inner_dim, patch_size * patch_size * self.out_channels, bias=True)
+
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor = None,
+        encoder_hidden_states_mask: torch.Tensor = None,
+        timestep: torch.LongTensor = None,
+        img_shapes: Optional[List[Tuple[int, int, int]]] = None,
+        txt_seq_lens: Optional[List[int]] = None,
+        guidance: torch.Tensor = None,  # TODO: this should probably be removed
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        return_dict: bool = True,
+    ) -> Union[torch.Tensor, Transformer2DModelOutput]:
+        """
+        The [`QwenTransformer2DModel`] forward method.
+
+        Args:
+            hidden_states (`torch.Tensor` of shape `(batch_size, image_sequence_length, in_channels)`):
+                Input `hidden_states`.
+            encoder_hidden_states (`torch.Tensor` of shape `(batch_size, text_sequence_length, joint_attention_dim)`):
+                Conditional embeddings (embeddings computed from the input conditions such as prompts) to use.
+            encoder_hidden_states_mask (`torch.Tensor` of shape `(batch_size, text_sequence_length)`):
+                Mask of the input conditions.
+            timestep ( `torch.LongTensor`):
+                Used to indicate denoising step.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain
+                tuple.
+
+        Returns:
+            If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
+            `tuple` where the first element is the sample tensor.
+        """
+        if attention_kwargs is not None:
+            attention_kwargs = attention_kwargs.copy()
+            lora_scale = attention_kwargs.pop("scale", 1.0)
+        else:
+            lora_scale = 1.0
+
+        if USE_PEFT_BACKEND:
+            # weight the lora layers by setting `lora_scale` for each PEFT layer
+            scale_lora_layers(self, lora_scale)
+        else:
+            if attention_kwargs is not None and attention_kwargs.get("scale", None) is not None:
+                logger.warning(
+                    "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective."
+                )
+
+        hidden_states = self.img_in(hidden_states)
+
+        timestep = timestep.to(hidden_states.dtype)
+        encoder_hidden_states = self.txt_norm(encoder_hidden_states)
+        encoder_hidden_states = self.txt_in(encoder_hidden_states)
+
+        if guidance is not None:
+            guidance = guidance.to(hidden_states.dtype) * 1000
+
+        temb = (
+            self.time_text_embed(timestep, hidden_states)
+            if guidance is None
+            else self.time_text_embed(timestep, guidance, hidden_states)
+        )
+
+        image_rotary_emb = self.pos_embed(img_shapes, txt_seq_lens, device=hidden_states.device)
+
+        for index_block, block in enumerate(self.transformer_blocks):
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
+                encoder_hidden_states, hidden_states = self._gradient_checkpointing_func(
+                    block,
+                    hidden_states,
+                    encoder_hidden_states,
+                    encoder_hidden_states_mask,
+                    temb,
+                    image_rotary_emb,
+                )
+
+            else:
+                encoder_hidden_states, hidden_states = block(
+                    hidden_states=hidden_states,
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_hidden_states_mask=encoder_hidden_states_mask,
+                    temb=temb,
+                    image_rotary_emb=image_rotary_emb,
+                    joint_attention_kwargs=attention_kwargs,
+                )
+
+        # Use only the image part (hidden_states) from the dual-stream blocks
+        hidden_states = self.norm_out(hidden_states, temb)
+        output = self.proj_out(hidden_states)
+
+        if USE_PEFT_BACKEND:
+            # remove `lora_scale` from each PEFT layer
+            unscale_lora_layers(self, lora_scale)
+
+        if not return_dict:
+            return (output,)
+
+        return Transformer2DModelOutput(sample=output)

src/diffusers/models/transformers/transformer_wan.py

@@ -180,6 +180,7 @@ class WanAttention(torch.nn.Module, AttentionModuleMixin):
         added_kv_proj_dim: Optional[int] = None,
         cross_attention_dim_head: Optional[int] = None,
         processor=None,
+        is_cross_attention=None,
     ):
         super().__init__()
 
@@ -207,6 +208,8 @@ class WanAttention(torch.nn.Module, AttentionModuleMixin):
             self.add_v_proj = torch.nn.Linear(added_kv_proj_dim, self.inner_dim, bias=True)
             self.norm_added_k = torch.nn.RMSNorm(dim_head * heads, eps=eps)
 
+        self.is_cross_attention = cross_attention_dim_head is not None
+
         self.set_processor(processor)
 
     def fuse_projections(self):
@@ -324,7 +327,7 @@ class WanTimeTextImageEmbedding(nn.Module):
     ):
         timestep = self.timesteps_proj(timestep)
         if timestep_seq_len is not None:
-            timestep = timestep.unflatten(0, (1, timestep_seq_len))
+            timestep = timestep.unflatten(0, (-1, timestep_seq_len))
 
         time_embedder_dtype = next(iter(self.time_embedder.parameters())).dtype
         if timestep.dtype != time_embedder_dtype and time_embedder_dtype != torch.int8:

src/diffusers/pipelines/__init__.py

@@ -387,6 +387,7 @@ else:
         "SkyReelsV2ImageToVideoPipeline",
         "SkyReelsV2Pipeline",
     ]
+    _import_structure["qwenimage"] = ["QwenImagePipeline"]
 try:
     if not is_onnx_available():
         raise OptionalDependencyNotAvailable()
@@ -703,6 +704,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .paint_by_example import PaintByExamplePipeline
         from .pia import PIAPipeline
         from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
+        from .qwenimage import QwenImagePipeline
         from .sana import SanaControlNetPipeline, SanaPipeline, SanaSprintImg2ImgPipeline, SanaSprintPipeline
         from .semantic_stable_diffusion import SemanticStableDiffusionPipeline
         from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline

src/diffusers/pipelines/qwenimage/__init__.py

@@ -0,0 +1,49 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_additional_imports = {}
+_import_structure = {"pipeline_output": ["QwenImagePipelineOutput", "QwenImagePriorReduxPipelineOutput"]}
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["modeling_qwenimage"] = ["ReduxImageEncoder"]
+    _import_structure["pipeline_qwenimage"] = ["QwenImagePipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
+    else:
+        from .pipeline_qwenimage import QwenImagePipeline
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)
+    for name, value in _additional_imports.items():
+        setattr(sys.modules[__name__], name, value)

src/diffusers/pipelines/qwenimage/pipeline_output.py

@@ -0,0 +1,21 @@
+from dataclasses import dataclass
+from typing import List, Union
+
+import numpy as np
+import PIL.Image
+
+from ...utils import BaseOutput
+
+
+@dataclass
+class QwenImagePipelineOutput(BaseOutput):
+    """
+    Output class for Stable Diffusion pipelines.
+
+    Args:
+        images (`List[PIL.Image.Image]` or `np.ndarray`)
+            List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
+            num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
+    """
+
+    images: Union[List[PIL.Image.Image], np.ndarray]

src/diffusers/pipelines/qwenimage/pipeline_qwenimage.py

@@ -0,0 +1,729 @@
+# Copyright 2025 Qwen-Image Team and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import numpy as np
+import torch
+from transformers import Qwen2_5_VLForConditionalGeneration, Qwen2Tokenizer
+
+from ...image_processor import VaeImageProcessor
+from ...models import AutoencoderKLQwenImage, QwenImageTransformer2DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import QwenImagePipelineOutput
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import QwenImagePipeline
+
+        >>> pipe = QwenImagePipeline.from_pretrained("Qwen/QwenImage-20B", torch_dtype=torch.bfloat16)
+        >>> pipe.to("cuda")
+        >>> prompt = "A cat holding a sign that says hello world"
+        >>> # Depending on the variant being used, the pipeline call will slightly vary.
+        >>> # Refer to the pipeline documentation for more details.
+        >>> image = pipe(prompt, num_inference_steps=4, guidance_scale=0.0).images[0]
+        >>> image.save("qwenimage.png")
+        ```
+"""
+
+
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.15,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class QwenImagePipeline(DiffusionPipeline):
+    r"""
+    The QwenImage pipeline for text-to-image generation.
+
+    Args:
+        transformer ([`QwenImageTransformer2DModel`]):
+            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`Qwen2.5-VL-7B-Instruct`]):
+            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct), specifically the
+            [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) variant.
+        tokenizer (`QwenTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+    """
+
+    model_cpu_offload_seq = "text_encoder->transformer->vae"
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        vae: AutoencoderKLQwenImage,
+        text_encoder: Qwen2_5_VLForConditionalGeneration,
+        tokenizer: Qwen2Tokenizer,
+        transformer: QwenImageTransformer2DModel,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            transformer=transformer,
+            scheduler=scheduler,
+        )
+        self.vae_scale_factor = 2 ** len(self.vae.temperal_downsample) if getattr(self, "vae", None) else 8
+        # QwenImage latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
+        # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
+        self.tokenizer_max_length = 1024
+        self.prompt_template_encode = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+        self.prompt_template_encode_start_idx = 34
+        self.default_sample_size = 128
+
+    def _extract_masked_hidden(self, hidden_states: torch.Tensor, mask: torch.Tensor):
+        bool_mask = mask.bool()
+        valid_lengths = bool_mask.sum(dim=1)
+        selected = hidden_states[bool_mask]
+        split_result = torch.split(selected, valid_lengths.tolist(), dim=0)
+
+        return split_result
+
+    def _get_qwen_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]] = None,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+
+        template = self.prompt_template_encode
+        drop_idx = self.prompt_template_encode_start_idx
+        txt = [template.format(e) for e in prompt]
+        txt_tokens = self.tokenizer(
+            txt, max_length=self.tokenizer_max_length + drop_idx, padding=True, truncation=True, return_tensors="pt"
+        ).to(self.device)
+        encoder_hidden_states = self.text_encoder(
+            input_ids=txt_tokens.input_ids,
+            attention_mask=txt_tokens.attention_mask,
+            output_hidden_states=True,
+        )
+        hidden_states = encoder_hidden_states.hidden_states[-1]
+        split_hidden_states = self._extract_masked_hidden(hidden_states, txt_tokens.attention_mask)
+        split_hidden_states = [e[drop_idx:] for e in split_hidden_states]
+        attn_mask_list = [torch.ones(e.size(0), dtype=torch.long, device=e.device) for e in split_hidden_states]
+        max_seq_len = max([e.size(0) for e in split_hidden_states])
+        prompt_embeds = torch.stack(
+            [torch.cat([u, u.new_zeros(max_seq_len - u.size(0), u.size(1))]) for u in split_hidden_states]
+        )
+        encoder_attention_mask = torch.stack(
+            [torch.cat([u, u.new_zeros(max_seq_len - u.size(0))]) for u in attn_mask_list]
+        )
+
+        prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+        return prompt_embeds, encoder_attention_mask
+
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        device: Optional[torch.device] = None,
+        num_images_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_embeds_mask: Optional[torch.Tensor] = None,
+        max_sequence_length: int = 1024,
+    ):
+        r"""
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+        """
+        device = device or self._execution_device
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt) if prompt_embeds is None else prompt_embeds.shape[0]
+
+        if prompt_embeds is None:
+            prompt_embeds, prompt_embeds_mask = self._get_qwen_prompt_embeds(prompt, device)
+
+        _, seq_len, _ = prompt_embeds.shape
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+        prompt_embeds_mask = prompt_embeds_mask.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds_mask = prompt_embeds_mask.view(batch_size * num_images_per_prompt, seq_len)
+
+        return prompt_embeds, prompt_embeds_mask
+
+    def check_inputs(
+        self,
+        prompt,
+        height,
+        width,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        prompt_embeds_mask=None,
+        negative_prompt_embeds_mask=None,
+        callback_on_step_end_tensor_inputs=None,
+        max_sequence_length=None,
+    ):
+        if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
+            logger.warning(
+                f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}. Dimensions will be resized accordingly"
+            )
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and prompt_embeds_mask is None:
+            raise ValueError(
+                "If `prompt_embeds` are provided, `prompt_embeds_mask` also have to be passed. Make sure to generate `prompt_embeds_mask` from the same text encoder that was used to generate `prompt_embeds`."
+            )
+        if negative_prompt_embeds is not None and negative_prompt_embeds_mask is None:
+            raise ValueError(
+                "If `negative_prompt_embeds` are provided, `negative_prompt_embeds_mask` also have to be passed. Make sure to generate `negative_prompt_embeds_mask` from the same text encoder that was used to generate `negative_prompt_embeds`."
+            )
+
+        if max_sequence_length is not None and max_sequence_length > 1024:
+            raise ValueError(f"`max_sequence_length` cannot be greater than 1024 but is {max_sequence_length}")
+
+    @staticmethod
+    def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
+        latent_image_ids = torch.zeros(height, width, 3)
+        latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height)[:, None]
+        latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width)[None, :]
+
+        latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
+
+        latent_image_ids = latent_image_ids.reshape(
+            latent_image_id_height * latent_image_id_width, latent_image_id_channels
+        )
+
+        return latent_image_ids.to(device=device, dtype=dtype)
+
+    @staticmethod
+    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
+        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
+        latents = latents.permute(0, 2, 4, 1, 3, 5)
+        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)
+
+        return latents
+
+    @staticmethod
+    def _unpack_latents(latents, height, width, vae_scale_factor):
+        batch_size, num_patches, channels = latents.shape
+
+        # VAE applies 8x compression on images but we must also account for packing which requires
+        # latent height and width to be divisible by 2.
+        height = 2 * (int(height) // (vae_scale_factor * 2))
+        width = 2 * (int(width) // (vae_scale_factor * 2))
+
+        latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
+        latents = latents.permute(0, 3, 1, 4, 2, 5)
+
+        latents = latents.reshape(batch_size, channels // (2 * 2), 1, height, width)
+
+        return latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    def prepare_latents(
+        self,
+        batch_size,
+        num_channels_latents,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+        latents=None,
+    ):
+        # VAE applies 8x compression on images but we must also account for packing which requires
+        # latent height and width to be divisible by 2.
+        height = 2 * (int(height) // (self.vae_scale_factor * 2))
+        width = 2 * (int(width) // (self.vae_scale_factor * 2))
+
+        shape = (batch_size, 1, num_channels_latents, height, width)
+
+        if latents is not None:
+            latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+            return latents.to(device=device, dtype=dtype), latent_image_ids
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+
+        latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+
+        return latents, latent_image_ids
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def current_timestep(self):
+        return self._current_timestep
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        negative_prompt: Union[str, List[str]] = None,
+        true_cfg_scale: float = 4.0,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        sigmas: Optional[List[float]] = None,
+        guidance_scale: float = 1.0,
+        num_images_per_prompt: int = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        prompt_embeds_mask: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds_mask: Optional[torch.Tensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        max_sequence_length: int = 512,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
+                not greater than `1`).
+            true_cfg_scale (`float`, *optional*, defaults to 1.0):
+                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image. This is set to 1024 by default for the best results.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
+            guidance_scale (`float`, *optional*, defaults to 3.5):
+                Guidance scale as defined in [Classifier-Free Diffusion
+                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+                the text `prompt`, usually at the expense of lower image quality.
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.Tensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will be generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.qwenimage.QwenImagePipelineOutput`] or `tuple`:
+            [`~pipelines.qwenimage.QwenImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When
+            returning a tuple, the first element is a list with the generated images.
+        """
+
+        height = height or self.default_sample_size * self.vae_scale_factor
+        width = width or self.default_sample_size * self.vae_scale_factor
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            height,
+            width,
+            negative_prompt=negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            prompt_embeds_mask=prompt_embeds_mask,
+            negative_prompt_embeds_mask=negative_prompt_embeds_mask,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+            max_sequence_length=max_sequence_length,
+        )
+
+        self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
+        self._current_timestep = None
+        self._interrupt = False
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+
+        has_neg_prompt = negative_prompt is not None or (
+            negative_prompt_embeds is not None and negative_prompt_embeds_mask is not None
+        )
+        do_true_cfg = true_cfg_scale > 1 and has_neg_prompt
+        prompt_embeds, prompt_embeds_mask = self.encode_prompt(
+            prompt=prompt,
+            prompt_embeds=prompt_embeds,
+            prompt_embeds_mask=prompt_embeds_mask,
+            device=device,
+            num_images_per_prompt=num_images_per_prompt,
+            max_sequence_length=max_sequence_length,
+        )
+        if do_true_cfg:
+            negative_prompt_embeds, negative_prompt_embeds_mask = self.encode_prompt(
+                prompt=negative_prompt,
+                prompt_embeds=negative_prompt_embeds,
+                prompt_embeds_mask=negative_prompt_embeds_mask,
+                device=device,
+                num_images_per_prompt=num_images_per_prompt,
+                max_sequence_length=max_sequence_length,
+            )
+
+        # 4. Prepare latent variables
+        num_channels_latents = self.transformer.config.in_channels // 4
+        latents, latent_image_ids = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+        img_shapes = [(1, height // self.vae_scale_factor // 2, width // self.vae_scale_factor // 2)] * batch_size
+
+        # 5. Prepare timesteps
+        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
+        image_seq_len = latents.shape[1]
+        mu = calculate_shift(
+            image_seq_len,
+            self.scheduler.config.get("base_image_seq_len", 256),
+            self.scheduler.config.get("max_image_seq_len", 4096),
+            self.scheduler.config.get("base_shift", 0.5),
+            self.scheduler.config.get("max_shift", 1.15),
+        )
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler,
+            num_inference_steps,
+            device,
+            sigmas=sigmas,
+            mu=mu,
+        )
+        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+        self._num_timesteps = len(timesteps)
+
+        # handle guidance
+        if self.transformer.config.guidance_embeds:
+            guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
+            guidance = guidance.expand(latents.shape[0])
+        else:
+            guidance = None
+
+        if self.attention_kwargs is None:
+            self._attention_kwargs = {}
+
+        # 6. Denoising loop
+        self.scheduler.set_begin_index(0)
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                self._current_timestep = t
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+                with self.transformer.cache_context("cond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latents,
+                        timestep=timestep / 1000,
+                        guidance=guidance,
+                        encoder_hidden_states_mask=prompt_embeds_mask,
+                        encoder_hidden_states=prompt_embeds,
+                        img_shapes=img_shapes,
+                        txt_seq_lens=prompt_embeds_mask.sum(dim=1).tolist(),
+                        attention_kwargs=self.attention_kwargs,
+                        return_dict=False,
+                    )[0]
+
+                if do_true_cfg:
+                    with self.transformer.cache_context("uncond"):
+                        neg_noise_pred = self.transformer(
+                            hidden_states=latents,
+                            timestep=timestep / 1000,
+                            guidance=guidance,
+                            encoder_hidden_states_mask=negative_prompt_embeds_mask,
+                            encoder_hidden_states=negative_prompt_embeds,
+                            img_shapes=img_shapes,
+                            txt_seq_lens=negative_prompt_embeds_mask.sum(dim=1).tolist(),
+                            attention_kwargs=self.attention_kwargs,
+                            return_dict=False,
+                        )[0]
+                    comb_pred = neg_noise_pred + true_cfg_scale * (noise_pred - neg_noise_pred)
+
+                    cond_norm = torch.norm(noise_pred, dim=-1, keepdim=True)
+                    noise_norm = torch.norm(comb_pred, dim=-1, keepdim=True)
+                    noise_pred = comb_pred * (cond_norm / noise_norm)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        self._current_timestep = None
+        if output_type == "latent":
+            image = latents
+        else:
+            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+            latents = latents.to(self.vae.dtype)
+            latents_mean = (
+                torch.tensor(self.vae.config.latents_mean)
+                .view(1, self.vae.config.z_dim, 1, 1, 1)
+                .to(latents.device, latents.dtype)
+            )
+            latents_std = 1.0 / torch.tensor(self.vae.config.latents_std).view(1, self.vae.config.z_dim, 1, 1, 1).to(
+                latents.device, latents.dtype
+            )
+            latents = latents / latents_std + latents_mean
+            image = self.vae.decode(latents, return_dict=False)[0][:, :, 0]
+            image = self.image_processor.postprocess(image, output_type=output_type)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return QwenImagePipelineOutput(images=image)

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -1034,7 +1034,8 @@ class StableDiffusionPipeline(
 
                 # expand the latents if we are doing classifier free guidance
                 latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
-                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+                if hasattr(self.scheduler, "scale_model_input"):
+                    latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
 
                 # predict the noise residual
                 noise_pred = self.unet(

src/diffusers/pipelines/wan/pipeline_wan.py

@@ -125,15 +125,15 @@ class WanPipeline(DiffusionPipeline, WanLoraLoaderMixin):
 
     model_cpu_offload_seq = "text_encoder->transformer->transformer_2->vae"
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
-    _optional_components = ["transformer_2"]
+    _optional_components = ["transformer", "transformer_2"]
 
     def __init__(
         self,
         tokenizer: AutoTokenizer,
         text_encoder: UMT5EncoderModel,
-        transformer: WanTransformer3DModel,
         vae: AutoencoderKLWan,
         scheduler: FlowMatchEulerDiscreteScheduler,
+        transformer: Optional[WanTransformer3DModel] = None,
         transformer_2: Optional[WanTransformer3DModel] = None,
         boundary_ratio: Optional[float] = None,
         expand_timesteps: bool = False,  # Wan2.2 ti2v
@@ -526,7 +526,7 @@ class WanPipeline(DiffusionPipeline, WanLoraLoaderMixin):
             device=device,
         )
 
-        transformer_dtype = self.transformer.dtype
+        transformer_dtype = self.transformer.dtype if self.transformer is not None else self.transformer_2.dtype
         prompt_embeds = prompt_embeds.to(transformer_dtype)
         if negative_prompt_embeds is not None:
             negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
@@ -536,7 +536,11 @@ class WanPipeline(DiffusionPipeline, WanLoraLoaderMixin):
         timesteps = self.scheduler.timesteps
 
         # 5. Prepare latent variables
-        num_channels_latents = self.transformer.config.in_channels
+        num_channels_latents = (
+            self.transformer.config.in_channels
+            if self.transformer is not None
+            else self.transformer_2.config.in_channels
+        )
         latents = self.prepare_latents(
             batch_size * num_videos_per_prompt,
             num_channels_latents,

src/diffusers/pipelines/wan/pipeline_wan_i2v.py

@@ -162,17 +162,17 @@ class WanImageToVideoPipeline(DiffusionPipeline, WanLoraLoaderMixin):
 
     model_cpu_offload_seq = "text_encoder->image_encoder->transformer->transformer_2->vae"
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
-    _optional_components = ["transformer_2", "image_encoder", "image_processor"]
+    _optional_components = ["transformer", "transformer_2", "image_encoder", "image_processor"]
 
     def __init__(
         self,
         tokenizer: AutoTokenizer,
         text_encoder: UMT5EncoderModel,
-        transformer: WanTransformer3DModel,
         vae: AutoencoderKLWan,
         scheduler: FlowMatchEulerDiscreteScheduler,
         image_processor: CLIPImageProcessor = None,
         image_encoder: CLIPVisionModel = None,
+        transformer: WanTransformer3DModel = None,
         transformer_2: WanTransformer3DModel = None,
         boundary_ratio: Optional[float] = None,
         expand_timesteps: bool = False,
@@ -669,12 +669,13 @@ class WanImageToVideoPipeline(DiffusionPipeline, WanLoraLoaderMixin):
         )
 
         # Encode image embedding
-        transformer_dtype = self.transformer.dtype
+        transformer_dtype = self.transformer.dtype if self.transformer is not None else self.transformer_2.dtype
         prompt_embeds = prompt_embeds.to(transformer_dtype)
         if negative_prompt_embeds is not None:
             negative_prompt_embeds = negative_prompt_embeds.to(transformer_dtype)
 
-        if self.config.boundary_ratio is None and not self.config.expand_timesteps:
+        # only wan 2.1 i2v transformer accepts image_embeds
+        if self.transformer is not None and self.transformer.config.image_dim is not None:
             if image_embeds is None:
                 if last_image is None:
                     image_embeds = self.encode_image(image, device)
@@ -709,6 +710,7 @@ class WanImageToVideoPipeline(DiffusionPipeline, WanLoraLoaderMixin):
             last_image,
         )
         if self.config.expand_timesteps:
+            # wan 2.2 5b i2v use firt_frame_mask to mask timesteps
             latents, condition, first_frame_mask = latents_outputs
         else:
             latents, condition = latents_outputs

src/diffusers/utils/dummy_pt_objects.py

@@ -423,6 +423,21 @@ class AutoencoderKLMochi(metaclass=DummyObject):
         requires_backends(cls, ["torch"])
 
 
+class AutoencoderKLQwenImage(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
 class AutoencoderKLTemporalDecoder(metaclass=DummyObject):
     _backends = ["torch"]
 
@@ -1038,6 +1053,21 @@ class PriorTransformer(metaclass=DummyObject):
         requires_backends(cls, ["torch"])
 
 
+class QwenImageTransformer2DModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch"])
+
+
 class SanaControlNetModel(metaclass=DummyObject):
     _backends = ["torch"]
 

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -1742,6 +1742,21 @@ class PixArtSigmaPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class QwenImagePipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class ReduxImageEncoder(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

tests/pipelines/qwenimage/test_qwenimage.py

@@ -0,0 +1,236 @@
+# Copyright 2025 The HuggingFace Team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import numpy as np
+import torch
+from transformers import Qwen2_5_VLConfig, Qwen2_5_VLForConditionalGeneration, Qwen2Tokenizer
+
+from diffusers import (
+    AutoencoderKLQwenImage,
+    FlowMatchEulerDiscreteScheduler,
+    QwenImagePipeline,
+    QwenImageTransformer2DModel,
+)
+from diffusers.utils.testing_utils import enable_full_determinism, torch_device
+
+from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
+from ..test_pipelines_common import PipelineTesterMixin, to_np
+
+
+enable_full_determinism()
+
+
+class QwenImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = QwenImagePipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    supports_dduf = False
+    test_xformers_attention = False
+    test_layerwise_casting = True
+    test_group_offloading = True
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        transformer = QwenImageTransformer2DModel(
+            patch_size=2,
+            in_channels=16,
+            out_channels=4,
+            num_layers=2,
+            attention_head_dim=16,
+            num_attention_heads=3,
+            joint_attention_dim=16,
+            guidance_embeds=False,
+            axes_dims_rope=(8, 4, 4),
+        )
+
+        torch.manual_seed(0)
+        z_dim = 4
+        vae = AutoencoderKLQwenImage(
+            base_dim=z_dim * 6,
+            z_dim=z_dim,
+            dim_mult=[1, 2, 4],
+            num_res_blocks=1,
+            temperal_downsample=[False, True],
+            # fmt: off
+            latents_mean=[0.0] * 4,
+            latents_std=[1.0] * 4,
+            # fmt: on
+        )
+
+        torch.manual_seed(0)
+        scheduler = FlowMatchEulerDiscreteScheduler()
+
+        torch.manual_seed(0)
+        config = Qwen2_5_VLConfig(
+            text_config={
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_hidden_layers": 2,
+                "num_attention_heads": 2,
+                "num_key_value_heads": 2,
+                "rope_scaling": {
+                    "mrope_section": [1, 1, 2],
+                    "rope_type": "default",
+                    "type": "default",
+                },
+                "rope_theta": 1000000.0,
+            },
+            vision_config={
+                "depth": 2,
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_heads": 2,
+                "out_hidden_size": 16,
+            },
+            hidden_size=16,
+            vocab_size=152064,
+            vision_end_token_id=151653,
+            vision_start_token_id=151652,
+            vision_token_id=151654,
+        )
+        text_encoder = Qwen2_5_VLForConditionalGeneration(config)
+        tokenizer = Qwen2Tokenizer.from_pretrained("hf-internal-testing/tiny-random-Qwen2VLForConditionalGeneration")
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+
+        inputs = {
+            "prompt": "dance monkey",
+            "negative_prompt": "bad quality",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 3.0,
+            "true_cfg_scale": 1.0,
+            "height": 32,
+            "width": 32,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = pipe(**inputs).images
+        generated_image = image[0]
+        self.assertEqual(generated_image.shape, (3, 32, 32))
+
+        # fmt: off
+        expected_slice = torch.tensor([0.563, 0.6358, 0.6028, 0.5656, 0.5806, 0.5512, 0.5712, 0.6331, 0.4147, 0.3558, 0.5625, 0.4831, 0.4957, 0.5258, 0.4075, 0.5018])
+        # fmt: on
+
+        generated_slice = generated_image.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(torch.allclose(generated_slice, expected_slice, atol=1e-3))
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(batch_size=3, expected_max_diff=1e-1)
+
+    def test_attention_slicing_forward_pass(
+        self, test_max_difference=True, test_mean_pixel_difference=True, expected_max_diff=1e-3
+    ):
+        if not self.test_attention_slicing:
+            return
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        output_without_slicing = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=1)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing1 = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=2)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing2 = pipe(**inputs)[0]
+
+        if test_max_difference:
+            max_diff1 = np.abs(to_np(output_with_slicing1) - to_np(output_without_slicing)).max()
+            max_diff2 = np.abs(to_np(output_with_slicing2) - to_np(output_without_slicing)).max()
+            self.assertLess(
+                max(max_diff1, max_diff2),
+                expected_max_diff,
+                "Attention slicing should not affect the inference results",
+            )
+
+    def test_vae_tiling(self, expected_diff_max: float = 0.2):
+        generator_device = "cpu"
+        components = self.get_dummy_components()
+
+        pipe = self.pipeline_class(**components)
+        pipe.to("cpu")
+        pipe.set_progress_bar_config(disable=None)
+
+        # Without tiling
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_without_tiling = pipe(**inputs)[0]
+
+        # With tiling
+        pipe.vae.enable_tiling(
+            tile_sample_min_height=96,
+            tile_sample_min_width=96,
+            tile_sample_stride_height=64,
+            tile_sample_stride_width=64,
+        )
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_with_tiling = pipe(**inputs)[0]
+
+        self.assertLess(
+            (to_np(output_without_tiling) - to_np(output_with_tiling)).max(),
+            expected_diff_max,
+            "VAE tiling should not affect the inference results",
+        )

tests/pipelines/wan/test_wan.py

@@ -13,8 +13,10 @@
 # limitations under the License.
 
 import gc
+import tempfile
 import unittest
 
+import numpy as np
 import torch
 from transformers import AutoTokenizer, T5EncoderModel
 
@@ -85,29 +87,13 @@ class WanPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             rope_max_seq_len=32,
         )
 
-        torch.manual_seed(0)
-        transformer_2 = WanTransformer3DModel(
-            patch_size=(1, 2, 2),
-            num_attention_heads=2,
-            attention_head_dim=12,
-            in_channels=16,
-            out_channels=16,
-            text_dim=32,
-            freq_dim=256,
-            ffn_dim=32,
-            num_layers=2,
-            cross_attn_norm=True,
-            qk_norm="rms_norm_across_heads",
-            rope_max_seq_len=32,
-        )
-
         components = {
             "transformer": transformer,
             "vae": vae,
             "scheduler": scheduler,
             "text_encoder": text_encoder,
             "tokenizer": tokenizer,
-            "transformer_2": transformer_2,
+            "transformer_2": None,
         }
         return components
 
@@ -155,6 +141,45 @@ class WanPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
     def test_attention_slicing_forward_pass(self):
         pass
 
+    # _optional_components include transformer, transformer_2, but only transformer_2 is optional for this wan2.1 t2v pipeline
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = "transformer_2"
+
+        components = self.get_dummy_components()
+        components[optional_component] = None
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        self.assertTrue(
+            getattr(pipe_loaded, optional_component) is None,
+            f"`{optional_component}` did not stay set to None after loading.",
+        )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
+
 
 @slow
 @require_torch_accelerator

tests/pipelines/wan/test_wan_22.py

@@ -0,0 +1,367 @@
+# Copyright 2025 The HuggingFace Team.
+#
+# 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 tempfile
+import unittest
+
+import numpy as np
+import torch
+from transformers import AutoTokenizer, T5EncoderModel
+
+from diffusers import AutoencoderKLWan, UniPCMultistepScheduler, WanPipeline, WanTransformer3DModel
+from diffusers.utils.testing_utils import (
+    enable_full_determinism,
+    torch_device,
+)
+
+from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
+from ..test_pipelines_common import PipelineTesterMixin
+
+
+enable_full_determinism()
+
+
+class Wan22PipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = WanPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    test_xformers_attention = False
+    supports_dduf = False
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        vae = AutoencoderKLWan(
+            base_dim=3,
+            z_dim=16,
+            dim_mult=[1, 1, 1, 1],
+            num_res_blocks=1,
+            temperal_downsample=[False, True, True],
+        )
+
+        torch.manual_seed(0)
+        scheduler = UniPCMultistepScheduler(prediction_type="flow_prediction", use_flow_sigmas=True, flow_shift=3.0)
+        text_encoder = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")
+        tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        torch.manual_seed(0)
+        transformer = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=16,
+            out_channels=16,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        torch.manual_seed(0)
+        transformer_2 = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=16,
+            out_channels=16,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "transformer_2": transformer_2,
+            "boundary_ratio": 0.875,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        inputs = {
+            "prompt": "dance monkey",
+            "negative_prompt": "negative",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "height": 16,
+            "width": 16,
+            "num_frames": 9,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(
+            **components,
+        )
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+        self.assertEqual(generated_video.shape, (9, 3, 16, 16))
+
+        # fmt: off
+        expected_slice = torch.tensor([0.4525, 0.452, 0.4485, 0.4534, 0.4524, 0.4529, 0.454, 0.453, 0.5127, 0.5326, 0.5204, 0.5253, 0.5439, 0.5424, 0.5133, 0.5078])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(torch.allclose(generated_slice, expected_slice, atol=1e-3))
+
+    @unittest.skip("Test not supported")
+    def test_attention_slicing_forward_pass(self):
+        pass
+
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = "transformer"
+
+        components = self.get_dummy_components()
+        components[optional_component] = None
+        components["boundary_ratio"] = 1.0  # for wan 2.2 14B, transformer is not used when boundary_ratio is 1.0
+
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        self.assertTrue(
+            getattr(pipe_loaded, "transformer") is None,
+            "`transformer` did not stay set to None after loading.",
+        )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
+
+
+class Wan225BPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = WanPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    test_xformers_attention = False
+    supports_dduf = False
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        vae = AutoencoderKLWan(
+            base_dim=3,
+            z_dim=48,
+            in_channels=12,
+            out_channels=12,
+            is_residual=True,
+            patch_size=2,
+            latents_mean=[0.0] * 48,
+            latents_std=[1.0] * 48,
+            dim_mult=[1, 1, 1, 1],
+            num_res_blocks=1,
+            scale_factor_spatial=16,
+            scale_factor_temporal=4,
+            temperal_downsample=[False, True, True],
+        )
+
+        torch.manual_seed(0)
+        scheduler = UniPCMultistepScheduler(prediction_type="flow_prediction", use_flow_sigmas=True, flow_shift=3.0)
+        text_encoder = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")
+        tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        torch.manual_seed(0)
+        transformer = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=48,
+            out_channels=48,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "transformer_2": None,
+            "boundary_ratio": None,
+            "expand_timesteps": True,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        inputs = {
+            "prompt": "dance monkey",
+            "negative_prompt": "negative",  # TODO
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "height": 32,
+            "width": 32,
+            "num_frames": 9,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(
+            **components,
+        )
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+        self.assertEqual(generated_video.shape, (9, 3, 32, 32))
+
+        # fmt: off
+        expected_slice = torch.tensor([[[0.4814, 0.4298, 0.5094, 0.4289, 0.5061, 0.4301, 0.5043, 0.4284, 0.5375,
+                                        0.5965, 0.5527, 0.6014, 0.5228, 0.6076, 0.6644, 0.5651]]])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(
+            torch.allclose(generated_slice, expected_slice, atol=1e-3),
+            f"generated_slice: {generated_slice}, expected_slice: {expected_slice}",
+        )
+
+    @unittest.skip("Test not supported")
+    def test_attention_slicing_forward_pass(self):
+        pass
+
+    def test_components_function(self):
+        init_components = self.get_dummy_components()
+        init_components.pop("boundary_ratio")
+        init_components.pop("expand_timesteps")
+        pipe = self.pipeline_class(**init_components)
+
+        self.assertTrue(hasattr(pipe, "components"))
+        self.assertTrue(set(pipe.components.keys()) == set(init_components.keys()))
+
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = "transformer_2"
+
+        components = self.get_dummy_components()
+        components[optional_component] = None
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        self.assertTrue(
+            getattr(pipe_loaded, optional_component) is None,
+            f"`{optional_component}` did not stay set to None after loading.",
+        )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(expected_max_diff=2e-3)

tests/pipelines/wan/test_wan_22_image_to_video.py

@@ -0,0 +1,392 @@
+# Copyright 2025 The HuggingFace Team.
+#
+# 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 tempfile
+import unittest
+
+import numpy as np
+import torch
+from PIL import Image
+from transformers import AutoTokenizer, T5EncoderModel
+
+from diffusers import AutoencoderKLWan, UniPCMultistepScheduler, WanImageToVideoPipeline, WanTransformer3DModel
+from diffusers.utils.testing_utils import (
+    enable_full_determinism,
+    torch_device,
+)
+
+from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
+from ..test_pipelines_common import PipelineTesterMixin
+
+
+enable_full_determinism()
+
+
+class Wan22ImageToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = WanImageToVideoPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    test_xformers_attention = False
+    supports_dduf = False
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        vae = AutoencoderKLWan(
+            base_dim=3,
+            z_dim=16,
+            dim_mult=[1, 1, 1, 1],
+            num_res_blocks=1,
+            temperal_downsample=[False, True, True],
+        )
+
+        torch.manual_seed(0)
+        scheduler = UniPCMultistepScheduler(prediction_type="flow_prediction", use_flow_sigmas=True, flow_shift=3.0)
+        text_encoder = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")
+        tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        torch.manual_seed(0)
+        transformer = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=36,
+            out_channels=16,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        torch.manual_seed(0)
+        transformer_2 = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=36,
+            out_channels=16,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "transformer_2": transformer_2,
+            "image_encoder": None,
+            "image_processor": None,
+            "boundary_ratio": 0.875,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        image_height = 16
+        image_width = 16
+        image = Image.new("RGB", (image_width, image_height))
+        inputs = {
+            "image": image,
+            "prompt": "dance monkey",
+            "negative_prompt": "negative",  # TODO
+            "height": image_height,
+            "width": image_width,
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "num_frames": 9,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(
+            **components,
+        )
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+        self.assertEqual(generated_video.shape, (9, 3, 16, 16))
+
+        # fmt: off
+        expected_slice = torch.tensor([0.4527, 0.4526, 0.4498, 0.4539, 0.4521, 0.4524, 0.4533, 0.4535, 0.5154,
+                                       0.5353, 0.5200, 0.5174, 0.5434, 0.5301, 0.5199, 0.5216])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(
+            torch.allclose(generated_slice, expected_slice, atol=1e-3),
+            f"generated_slice: {generated_slice}, expected_slice: {expected_slice}",
+        )
+
+    @unittest.skip("Test not supported")
+    def test_attention_slicing_forward_pass(self):
+        pass
+
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = ["transformer", "image_encoder", "image_processor"]
+
+        components = self.get_dummy_components()
+        for component in optional_component:
+            components[component] = None
+        components["boundary_ratio"] = 1.0  # for wan 2.2 14B, transformer is not used when boundary_ratio is 1.0
+
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        for component in optional_component:
+            self.assertTrue(
+                getattr(pipe_loaded, component) is None,
+                f"`{component}` did not stay set to None after loading.",
+            )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
+
+
+class Wan225BImageToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = WanImageToVideoPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    test_xformers_attention = False
+    supports_dduf = False
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        vae = AutoencoderKLWan(
+            base_dim=3,
+            z_dim=48,
+            in_channels=12,
+            out_channels=12,
+            is_residual=True,
+            patch_size=2,
+            latents_mean=[0.0] * 48,
+            latents_std=[1.0] * 48,
+            dim_mult=[1, 1, 1, 1],
+            num_res_blocks=1,
+            scale_factor_spatial=16,
+            scale_factor_temporal=4,
+            temperal_downsample=[False, True, True],
+        )
+
+        torch.manual_seed(0)
+        scheduler = UniPCMultistepScheduler(prediction_type="flow_prediction", use_flow_sigmas=True, flow_shift=3.0)
+        text_encoder = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")
+        tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+        torch.manual_seed(0)
+        transformer = WanTransformer3DModel(
+            patch_size=(1, 2, 2),
+            num_attention_heads=2,
+            attention_head_dim=12,
+            in_channels=48,
+            out_channels=48,
+            text_dim=32,
+            freq_dim=256,
+            ffn_dim=32,
+            num_layers=2,
+            cross_attn_norm=True,
+            qk_norm="rms_norm_across_heads",
+            rope_max_seq_len=32,
+        )
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "transformer_2": None,
+            "image_encoder": None,
+            "image_processor": None,
+            "boundary_ratio": None,
+            "expand_timesteps": True,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        image_height = 32
+        image_width = 32
+        image = Image.new("RGB", (image_width, image_height))
+        inputs = {
+            "image": image,
+            "prompt": "dance monkey",
+            "negative_prompt": "negative",  # TODO
+            "height": image_height,
+            "width": image_width,
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "num_frames": 9,
+            "max_sequence_length": 16,
+            "output_type": "pt",
+        }
+        return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(
+            **components,
+        )
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+        self.assertEqual(generated_video.shape, (9, 3, 32, 32))
+
+        # fmt: off
+        expected_slice = torch.tensor([[0.4833, 0.4305, 0.5100, 0.4299, 0.5056, 0.4298, 0.5052, 0.4332, 0.5550,
+                                       0.6092, 0.5536, 0.5928, 0.5199, 0.5864, 0.6705, 0.5493]])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(
+            torch.allclose(generated_slice, expected_slice, atol=1e-3),
+            f"generated_slice: {generated_slice}, expected_slice: {expected_slice}",
+        )
+
+    @unittest.skip("Test not supported")
+    def test_attention_slicing_forward_pass(self):
+        pass
+
+    def test_components_function(self):
+        init_components = self.get_dummy_components()
+        init_components.pop("boundary_ratio")
+        init_components.pop("expand_timesteps")
+        pipe = self.pipeline_class(**init_components)
+
+        self.assertTrue(hasattr(pipe, "components"))
+        self.assertTrue(set(pipe.components.keys()) == set(init_components.keys()))
+
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = ["transformer_2", "image_encoder", "image_processor"]
+
+        components = self.get_dummy_components()
+        for component in optional_component:
+            components[component] = None
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        for component in optional_component:
+            self.assertTrue(
+                getattr(pipe_loaded, component) is None,
+                f"`{component}` did not stay set to None after loading.",
+            )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(expected_max_diff=2e-3)
+
+    @unittest.skip("Test not supported")
+    def test_callback_inputs(self):
+        pass

tests/pipelines/wan/test_wan_image_to_video.py

@@ -12,8 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import tempfile
 import unittest
 
+import numpy as np
 import torch
 from PIL import Image
 from transformers import (
@@ -25,7 +27,7 @@ from transformers import (
 )
 
 from diffusers import AutoencoderKLWan, FlowMatchEulerDiscreteScheduler, WanImageToVideoPipeline, WanTransformer3DModel
-from diffusers.utils.testing_utils import enable_full_determinism
+from diffusers.utils.testing_utils import enable_full_determinism, torch_device
 
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
@@ -86,23 +88,6 @@ class WanImageToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             image_dim=4,
         )
 
-        torch.manual_seed(0)
-        transformer_2 = WanTransformer3DModel(
-            patch_size=(1, 2, 2),
-            num_attention_heads=2,
-            attention_head_dim=12,
-            in_channels=36,
-            out_channels=16,
-            text_dim=32,
-            freq_dim=256,
-            ffn_dim=32,
-            num_layers=2,
-            cross_attn_norm=True,
-            qk_norm="rms_norm_across_heads",
-            rope_max_seq_len=32,
-            image_dim=4,
-        )
-
         torch.manual_seed(0)
         image_encoder_config = CLIPVisionConfig(
             hidden_size=4,
@@ -126,7 +111,7 @@ class WanImageToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "tokenizer": tokenizer,
             "image_encoder": image_encoder,
             "image_processor": image_processor,
-            "transformer_2": transformer_2,
+            "transformer_2": None,
         }
         return components
 
@@ -182,11 +167,44 @@ class WanImageToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
     def test_inference_batch_single_identical(self):
         pass
 
-    @unittest.skip(
-        "TODO: refactor this test: one component can be optional for certain checkpoints but not for others"
-    )
-    def test_save_load_optional_components(self):
-        pass
+    # _optional_components include transformer, transformer_2 and image_encoder, image_processor, but only transformer_2 is optional for wan2.1 i2v pipeline
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = "transformer_2"
+
+        components = self.get_dummy_components()
+        components[optional_component] = None
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        self.assertTrue(
+            getattr(pipe_loaded, optional_component) is None,
+            f"`{optional_component}` did not stay set to None after loading.",
+        )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)
 
 
 class WanFLFToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
@@ -242,24 +260,6 @@ class WanFLFToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             pos_embed_seq_len=2 * (4 * 4 + 1),
         )
 
-        torch.manual_seed(0)
-        transformer_2 = WanTransformer3DModel(
-            patch_size=(1, 2, 2),
-            num_attention_heads=2,
-            attention_head_dim=12,
-            in_channels=36,
-            out_channels=16,
-            text_dim=32,
-            freq_dim=256,
-            ffn_dim=32,
-            num_layers=2,
-            cross_attn_norm=True,
-            qk_norm="rms_norm_across_heads",
-            rope_max_seq_len=32,
-            image_dim=4,
-            pos_embed_seq_len=2 * (4 * 4 + 1),
-        )
-
         torch.manual_seed(0)
         image_encoder_config = CLIPVisionConfig(
             hidden_size=4,
@@ -283,7 +283,7 @@ class WanFLFToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
             "tokenizer": tokenizer,
             "image_encoder": image_encoder,
             "image_processor": image_processor,
-            "transformer_2": transformer_2,
+            "transformer_2": None,
         }
         return components
 
@@ -341,8 +341,41 @@ class WanFLFToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
     def test_inference_batch_single_identical(self):
         pass
 
-    @unittest.skip(
-        "TODO: refactor this test: one component can be optional for certain checkpoints but not for others"
-    )
-    def test_save_load_optional_components(self):
-        pass
+    # _optional_components include transformer, transformer_2 and image_encoder, image_processor, but only transformer_2 is optional for wan2.1 FLFT2V pipeline
+    def test_save_load_optional_components(self, expected_max_difference=1e-4):
+        optional_component = "transformer_2"
+
+        components = self.get_dummy_components()
+        components[optional_component] = None
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output = pipe(**inputs)[0]
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            pipe.save_pretrained(tmpdir, safe_serialization=False)
+            pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
+            for component in pipe_loaded.components.values():
+                if hasattr(component, "set_default_attn_processor"):
+                    component.set_default_attn_processor()
+            pipe_loaded.to(torch_device)
+            pipe_loaded.set_progress_bar_config(disable=None)
+
+        self.assertTrue(
+            getattr(pipe_loaded, optional_component) is None,
+            f"`{optional_component}` did not stay set to None after loading.",
+        )
+
+        inputs = self.get_dummy_inputs(generator_device)
+        torch.manual_seed(0)
+        output_loaded = pipe_loaded(**inputs)[0]
+
+        max_diff = np.abs(output.detach().cpu().numpy() - output_loaded.detach().cpu().numpy()).max()
+        self.assertLess(max_diff, expected_max_difference)