Refactor OmniGen (huggingface#10771)

* OmniGen model.py

* update OmniGenTransformerModel

* omnigen pipeline

* omnigen pipeline

* update omnigen_pipeline

* test case for omnigen

* update omnigenpipeline

* update docs

* update docs

* offload_transformer

* enable_transformer_block_cpu_offload

* update docs

* reformat

* reformat

* reformat

* update docs

* update docs

* make style

* make style

* Update docs/source/en/api/models/omnigen_transformer.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* update docs

* revert changes to examples/

* update OmniGen2DModel

* make style

* update test cases

* Update docs/source/en/api/pipelines/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* Update docs/source/en/using-diffusers/omnigen.md

Co-authored-by: Steven Liu <[email protected]>

* update docs

* typo

* Update src/diffusers/models/embeddings.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/models/attention.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/models/transformers/transformer_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/models/transformers/transformer_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/models/transformers/transformer_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update tests/pipelines/omnigen/test_pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update tests/pipelines/omnigen/test_pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* Update src/diffusers/pipelines/omnigen/pipeline_omnigen.py

Co-authored-by: hlky <[email protected]>

* consistent attention processor

* updata

* update

* check_inputs

* make style

* update testpipeline

* update testpipeline

* refactor omnigen

* more updates

* apply review suggestion

---------

Co-authored-by: shitao <[email protected]>
Co-authored-by: Steven Liu <[email protected]>
Co-authored-by: hlky <[email protected]>

Author: 4 people

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

@@ -14,6 +14,17 @@ specific language governing permissions and limitations under the License.
 
 A Transformer model that accepts multimodal instructions to generate images for [OmniGen](https://github.com/VectorSpaceLab/OmniGen/).
 
+The abstract from the paper is:
+
+*The emergence of Large Language Models (LLMs) has unified language  generation tasks and revolutionized human-machine interaction.  However, in the realm of image generation, a unified model capable of handling various tasks within a single framework remains largely unexplored. In this work, we introduce OmniGen, a new diffusion model for unified image generation. OmniGen is characterized by the following features: 1) Unification: OmniGen not only demonstrates text-to-image generation capabilities but also inherently supports various downstream tasks, such as image editing, subject-driven generation, and visual conditional generation. 2) Simplicity: The architecture of OmniGen is highly simplified, eliminating the need for additional plugins. Moreover, compared to existing diffusion models, it is more user-friendly and can complete complex tasks end-to-end through instructions without the need for extra intermediate steps, greatly simplifying the image generation workflow. 3) Knowledge Transfer: Benefit from learning in a unified format, OmniGen effectively transfers knowledge across different tasks, manages unseen tasks and domains, and exhibits novel capabilities. We also explore the model’s reasoning capabilities and potential applications of the chain-of-thought mechanism.  This work represents the first attempt at a general-purpose image generation model,  and we will release our resources at https://github.com/VectorSpaceLab/OmniGen to foster future advancements.*
+
+```python
+import torch
+from diffusers import OmniGenTransformer2DModel
+
+transformer = OmniGenTransformer2DModel.from_pretrained("Shitao/OmniGen-v1-diffusers", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
 ## OmniGenTransformer2DModel
 
 [[autodoc]] OmniGenTransformer2DModel

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

@@ -19,27 +19,7 @@
 
 The abstract from the paper is:
 
-*The emergence of Large Language Models (LLMs) has unified language 
-generation tasks and revolutionized human-machine interaction. 
-However, in the realm of image generation, a unified model capable of handling various tasks
-within a single framework remains largely unexplored. In
-this work, we introduce OmniGen, a new diffusion model
-for unified image generation. OmniGen is characterized
-by the following features: 1) Unification: OmniGen not
-only demonstrates text-to-image generation capabilities but
-also inherently supports various downstream tasks, such
-as image editing, subject-driven generation, and visual conditional generation. 2) Simplicity: The architecture of
-OmniGen is highly simplified, eliminating the need for additional plugins. Moreover, compared to existing diffusion
-models, it is more user-friendly and can complete complex
-tasks end-to-end through instructions without the need for
-extra intermediate steps, greatly simplifying the image generation workflow. 3) Knowledge Transfer: Benefit from
-learning in a unified format, OmniGen effectively transfers
-knowledge across different tasks, manages unseen tasks and
-domains, and exhibits novel capabilities. We also explore
-the model’s reasoning capabilities and potential applications of the chain-of-thought mechanism. 
-This work represents the first attempt at a general-purpose image generation model, 
-and we will release our resources at https:
-//github.com/VectorSpaceLab/OmniGen to foster future advancements.*
+*The emergence of Large Language Models (LLMs) has unified language  generation tasks and revolutionized human-machine interaction.  However, in the realm of image generation, a unified model capable of handling various tasks within a single framework remains largely unexplored. In this work, we introduce OmniGen, a new diffusion model for unified image generation. OmniGen is characterized by the following features: 1) Unification: OmniGen not only demonstrates text-to-image generation capabilities but also inherently supports various downstream tasks, such as image editing, subject-driven generation, and visual conditional generation. 2) Simplicity: The architecture of OmniGen is highly simplified, eliminating the need for additional plugins. Moreover, compared to existing diffusion models, it is more user-friendly and can complete complex tasks end-to-end through instructions without the need for extra intermediate steps, greatly simplifying the image generation workflow. 3) Knowledge Transfer: Benefit from learning in a unified format, OmniGen effectively transfers knowledge across different tasks, manages unseen tasks and domains, and exhibits novel capabilities. We also explore the model’s reasoning capabilities and potential applications of the chain-of-thought mechanism.  This work represents the first attempt at a general-purpose image generation model,  and we will release our resources at https://github.com/VectorSpaceLab/OmniGen to foster future advancements.*
 
 <Tip>
 
@@ -49,25 +29,22 @@ Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.m
 
 This pipeline was contributed by [staoxiao](https://github.com/staoxiao). The original codebase can be found [here](https://github.com/VectorSpaceLab/OmniGen). The original weights can be found under [hf.co/shitao](https://huggingface.co/Shitao/OmniGen-v1).
 
-
 ## Inference
 
 First, load the pipeline:
 
 ```python
 import torch
 from diffusers import OmniGenPipeline
-pipe = OmniGenPipeline.from_pretrained(
-    "Shitao/OmniGen-v1-diffusers",
-    torch_dtype=torch.bfloat16
-)
+
+pipe = OmniGenPipeline.from_pretrained("Shitao/OmniGen-v1-diffusers", torch_dtype=torch.bfloat16)
 pipe.to("cuda")
 ```
 
 For text-to-image, pass a text prompt. By default, OmniGen generates a 1024x1024 image. 
 You can try setting the `height` and `width` parameters to generate images with different size.
 
-```py
+```python
 prompt = "Realistic photo. A young woman sits on a sofa, holding a book and facing the camera. She wears delicate silver hoop earrings adorned with tiny, sparkling diamonds that catch the light, with her long chestnut hair cascading over her shoulders. Her eyes are focused and gentle, framed by long, dark lashes. She is dressed in a cozy cream sweater, which complements her warm, inviting smile. Behind her, there is a table with a cup of water in a sleek, minimalist blue mug. The background is a serene indoor setting with soft natural light filtering through a window, adorned with tasteful art and flowers, creating a cozy and peaceful ambiance. 4K, HD."
 image = pipe(
     prompt=prompt,
@@ -76,14 +53,14 @@ image = pipe(
     guidance_scale=3,
     generator=torch.Generator(device="cpu").manual_seed(111),
 ).images[0]
-image
+image.save("output.png")
 ```
 
 OmniGen supports multimodal inputs. 
 When the input includes an image, you need to add a placeholder `<img><|image_1|></img>` in the text prompt to represent the image. 
 It is recommended to enable `use_input_image_size_as_output` to keep the edited image the same size as the original image.
 
-```py
+```python
 prompt="<img><|image_1|></img> Remove the woman's earrings. Replace the mug with a clear glass filled with sparkling iced cola."
 input_images=[load_image("https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/t2i_woman_with_book.png")]
 image = pipe(
@@ -93,14 +70,11 @@ image = pipe(
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
     generator=torch.Generator(device="cpu").manual_seed(222)).images[0]
-image
+image.save("output.png")
 ```
 
-
 ## OmniGenPipeline
 
 [[autodoc]] OmniGenPipeline
   - all
   - __call__
-
-

docs/source/en/using-diffusers/omnigen.md

@@ -19,25 +19,22 @@ For more information, please refer to the [paper](https://arxiv.org/pdf/2409.113
 This guide will walk you through using OmniGen for various tasks and use cases.
 
 ## Load model checkpoints
+
 Model weights may be stored in separate subfolders on the Hub or locally, in which case, you should use the [`~DiffusionPipeline.from_pretrained`] method.
 
-```py
+```python
 import torch
 from diffusers import OmniGenPipeline
-pipe = OmniGenPipeline.from_pretrained(
-    "Shitao/OmniGen-v1-diffusers",
-    torch_dtype=torch.bfloat16
-)
-```
-
 
+pipe = OmniGenPipeline.from_pretrained("Shitao/OmniGen-v1-diffusers", torch_dtype=torch.bfloat16)
+```
 
 ## Text-to-image
 
 For text-to-image, pass a text prompt. By default, OmniGen generates a 1024x1024 image. 
 You can try setting the `height` and `width` parameters to generate images with different size.
 
-```py
+```python
 import torch
 from diffusers import OmniGenPipeline
 
@@ -55,8 +52,9 @@ image = pipe(
     guidance_scale=3,
     generator=torch.Generator(device="cpu").manual_seed(111),
 ).images[0]
-image
+image.save("output.png")
 ```
+
 <div class="flex justify-center">
     <img src="https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/t2i_woman_with_book.png" alt="generated image"/>
 </div>
@@ -67,7 +65,7 @@ OmniGen supports multimodal inputs.
 When the input includes an image, you need to add a placeholder `<img><|image_1|></img>` in the text prompt to represent the image. 
 It is recommended to enable `use_input_image_size_as_output` to keep the edited image the same size as the original image.
 
-```py
+```python
 import torch
 from diffusers import OmniGenPipeline
 from diffusers.utils import load_image 
@@ -86,9 +84,11 @@ image = pipe(
     guidance_scale=2, 
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
-    generator=torch.Generator(device="cpu").manual_seed(222)).images[0]
-image
+    generator=torch.Generator(device="cpu").manual_seed(222)
+).images[0]
+image.save("output.png")
 ```
+
 <div class="flex flex-row gap-4">
   <div class="flex-1">
     <img class="rounded-xl" src="https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/t2i_woman_with_book.png"/>
@@ -101,7 +101,8 @@ image
 </div>
 
 OmniGen has some interesting features, such as visual reasoning, as shown in the example below.
-```py
+
+```python
 prompt="If the woman is thirsty, what should she take? Find it in the image and highlight it in blue. <img><|image_1|></img>"
 input_images=[load_image("https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/edit.png")]
 image = pipe(
@@ -110,20 +111,20 @@ image = pipe(
     guidance_scale=2, 
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
-    generator=torch.Generator(device="cpu").manual_seed(0)).images[0]
-image
+    generator=torch.Generator(device="cpu").manual_seed(0)
+).images[0]
+image.save("output.png")
 ```
+
 <div class="flex justify-center">
     <img src="https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/reasoning.png" alt="generated image"/>
 </div>
 
-
 ## Controllable generation
 
- OmniGen can handle several classic computer vision tasks. 
- As shown below, OmniGen can detect human skeletons in input images, which can be used as control conditions to generate new images.
+OmniGen can handle several classic computer vision tasks. As shown below, OmniGen can detect human skeletons in input images, which can be used as control conditions to generate new images.
 
-```py
+```python
 import torch
 from diffusers import OmniGenPipeline
 from diffusers.utils import load_image 
@@ -142,8 +143,9 @@ image1 = pipe(
     guidance_scale=2, 
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
-    generator=torch.Generator(device="cpu").manual_seed(333)).images[0]
-image1
+    generator=torch.Generator(device="cpu").manual_seed(333)
+).images[0]
+image1.save("image1.png")
 
 prompt="Generate a new photo using the following picture and text as conditions: <img><|image_1|></img>\n A young boy is sitting on a sofa in the library, holding a book. His hair is neatly combed, and a faint smile plays on his lips, with a few freckles scattered across his cheeks. The library is quiet, with rows of shelves filled with books stretching out behind him."
 input_images=[load_image("https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/skeletal.png")]
@@ -153,8 +155,9 @@ image2 = pipe(
     guidance_scale=2, 
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
-    generator=torch.Generator(device="cpu").manual_seed(333)).images[0]
-image2
+    generator=torch.Generator(device="cpu").manual_seed(333)
+).images[0]
+image2.save("image2.png")
 ```
 
 <div class="flex flex-row gap-4">
@@ -174,7 +177,8 @@ image2
 
 
 OmniGen can also directly use relevant information from input images to generate new images.
-```py
+
+```python
 import torch
 from diffusers import OmniGenPipeline
 from diffusers.utils import load_image 
@@ -193,23 +197,24 @@ image = pipe(
     guidance_scale=2, 
     img_guidance_scale=1.6,
     use_input_image_size_as_output=True,
-    generator=torch.Generator(device="cpu").manual_seed(0)).images[0]
-image
+    generator=torch.Generator(device="cpu").manual_seed(0)
+).images[0]
+image.save("output.png")
 ```
+
 <div class="flex flex-row gap-4">
   <div class="flex-1">
     <img class="rounded-xl" src="https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/same_pose.png"/>
     <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
   </div>
 </div>
 
-
 ## ID and object preserving
 
 OmniGen can generate multiple images based on the people and objects in the input image and supports inputting multiple images simultaneously. 
 Additionally, OmniGen can extract desired objects from an image containing multiple objects based on instructions.
 
-```py
+```python
 import torch
 from diffusers import OmniGenPipeline
 from diffusers.utils import load_image 
@@ -231,9 +236,11 @@ image = pipe(
     width=1024,
     guidance_scale=2.5, 
     img_guidance_scale=1.6,
-    generator=torch.Generator(device="cpu").manual_seed(666)).images[0]
-image
+    generator=torch.Generator(device="cpu").manual_seed(666)
+).images[0]
+image.save("output.png")
 ```
+
 <div class="flex flex-row gap-4">
   <div class="flex-1">
     <img class="rounded-xl" src="https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/3.png"/>
@@ -249,7 +256,6 @@ image
   </div>
 </div>
 
-
 ```py
 import torch
 from diffusers import OmniGenPipeline
@@ -261,7 +267,6 @@ pipe = OmniGenPipeline.from_pretrained(
 )
 pipe.to("cuda")
 
-
 prompt="A woman is walking down the street, wearing a white long-sleeve blouse with lace details on the sleeves, paired with a blue pleated skirt. The woman is <img><|image_1|></img>. The long-sleeve blouse and a pleated skirt are <img><|image_2|></img>."
 input_image_1 = load_image("https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/emma.jpeg")
 input_image_2 = load_image("https://raw.githubusercontent.com/VectorSpaceLab/OmniGen/main/imgs/docs_img/dress.jpg")
@@ -273,8 +278,9 @@ image = pipe(
     width=1024,
     guidance_scale=2.5, 
     img_guidance_scale=1.6,
-    generator=torch.Generator(device="cpu").manual_seed(666)).images[0]
-image
+    generator=torch.Generator(device="cpu").manual_seed(666)
+).images[0]
+image.save("output.png")
 ```
 
 <div class="flex flex-row gap-4">
@@ -292,13 +298,12 @@ image
   </div>
 </div>
 
-
-## Optimization when inputting multiple images 
+## Optimization when using multiple images 
 
 For text-to-image task, OmniGen requires minimal memory and time costs (9GB memory and 31s for a 1024x1024 image on A800 GPU). 
 However, when using input images, the computational cost increases. 
 
-Here are some guidelines to help you reduce computational costs when inputting multiple images. The experiments are conducted on an A800 GPU with two input images.
+Here are some guidelines to help you reduce computational costs when using multiple images. The experiments are conducted on an A800 GPU with two input images.
 
 Like other pipelines, you can reduce memory usage by offloading the model: `pipe.enable_model_cpu_offload()` or `pipe.enable_sequential_cpu_offload() `. 
 In OmniGen, you can also decrease computational overhead by reducing the `max_input_image_size`. 
@@ -310,5 +315,3 @@ The memory consumption for different image sizes is shown in the table below:
 | max_input_image_size=512  | 17GB         |
 | max_input_image_size=256  | 14GB         |
 
-
-

src/diffusers/models/embeddings.py

@@ -1199,7 +1199,7 @@ def apply_rotary_emb(
             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
+            # Used for Stable Audio and OmniGen
             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: