Add UniPic3Pipeline for multi-image editing and composition

docs/source/en/_toctree.yml

@@ -645,6 +645,8 @@
         title: Stable Diffusion
       - local: api/pipelines/stable_unclip
         title: Stable unCLIP
+      - local: api/pipelines/unipic3
+        title: UniPic-3
       - local: api/pipelines/value_guided_sampling
         title: Value-guided sampling
       - local: api/pipelines/visualcloze

docs/source/en/api/pipelines/unipic3.md

@@ -0,0 +1,56 @@
+<!--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.
+-->
+
+# UniPic-3
+
+[UniPic-3](https://arxiv.org/abs/2601.15664) is a unified framework for **single-image editing** and **multi-image composition** by Skywork, built on the Qwen-Image-Edit architecture. It supports 1–6 input images with arbitrary output resolutions.
+
+| Model | HuggingFace | Inference Steps |
+|-------|-------------|-----------------|
+| Base Model | [Skywork/Unipic3](https://huggingface.co/Skywork/Unipic3) | 50 steps |
+| Consistency Model | [Skywork/Unipic3-Consistency-Model](https://huggingface.co/Skywork/Unipic3-Consistency-Model) | 8 steps |
+| DMD Model | [Skywork/Unipic3-DMD](https://huggingface.co/Skywork/Unipic3-DMD) | 8 steps |
+
+> [!TIP]
+> Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.md) to learn how to explore the tradeoff between scheduler speed and quality.
+
+## Usage example
+
+```python
+import torch
+from PIL import Image
+from diffusers import UniPic3Pipeline
+from diffusers.utils import load_image
+
+pipe = UniPic3Pipeline.from_pretrained("Skywork/Unipic3", torch_dtype=torch.bfloat16)
+pipe.to("cuda")
+
+image1 = load_image("https://example.com/pig.png").convert("RGB")
+image2 = load_image("https://example.com/sunglasses.png").convert("RGB")
+
+image = pipe(
+    images=[image1, image2],
+    prompt="A pig wearing sunglasses.",
+    negative_prompt=" ",
+    num_inference_steps=50,
+    true_cfg_scale=4.0,
+    generator=torch.manual_seed(0),
+).images[0]
+
+image.save("unipic3_output.png")
+```
+
+## UniPic3Pipeline
+
+[[autodoc]] UniPic3Pipeline
+  - all
+  - __call__

src/diffusers/__init__.py

@@ -813,6 +813,7 @@
             "UniDiffuserModel",
             "UniDiffuserPipeline",
             "UniDiffuserTextDecoder",
+            "UniPic3Pipeline",
             "VersatileDiffusionDualGuidedPipeline",
             "VersatileDiffusionImageVariationPipeline",
             "VersatileDiffusionPipeline",
@@ -1667,6 +1668,7 @@
             UniDiffuserModel,
             UniDiffuserPipeline,
             UniDiffuserTextDecoder,
+            UniPic3Pipeline,
             VersatileDiffusionDualGuidedPipeline,
             VersatileDiffusionImageVariationPipeline,
             VersatileDiffusionPipeline,

src/diffusers/pipelines/__init__.py

@@ -469,6 +469,7 @@
         "QwenImageControlNetPipeline",
         "QwenImageLayeredPipeline",
     ]
+    _import_structure["unipic3"] = ["UniPic3Pipeline"]
     _import_structure["chronoedit"] = ["ChronoEditPipeline"]
     _import_structure["glm_image"] = ["GlmImagePipeline"]
 
@@ -925,6 +926,7 @@
             StableDiffusionAdapterPipeline,
             StableDiffusionXLAdapterPipeline,
         )
+        from .unipic3 import UniPic3Pipeline
         from .visualcloze import VisualClozeGenerationPipeline, VisualClozePipeline
         from .wan import (
             WanAnimatePipeline,

src/diffusers/pipelines/unipic3/__init__.py

@@ -0,0 +1,48 @@
+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 = {}
+
+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["pipeline_unipic3"] = ["UniPic3Pipeline"]
+
+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_unipic3 import UniPic3Pipeline
+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)