Add QwenImage Inpainting and Img2Img pipeline

[Trgtuan10]

What does this PR do?

This PR introduces support for the QwenImage model in both Inpainting and Img2Img tasks, expanding the model’s creative capabilities and integration within the Diffusers library.

1. Image to Image

Example code

import torch
from diffusers import QwenImageImg2ImgPipeline
from diffusers.utils import load_image

pipe = QwenImageImg2ImgPipeline.from_pretrained("Qwen/Qwen-Image", torch_dtype=torch.bfloat16,)
pipe = pipe.to("cuda")

init_image = load_image("https://github.com/Trgtuan10/Image_storage/blob/main/cute_cat.png?raw=true")

prompt = "wizard dog, Gandalf-inspired, Lord of the Rings aesthetic, majestic yet cute, Studio Ghibli style"
negative_prompt = ""

strengths = [0.6, 0.7, 0.8, 0.9, 1.0]

for s in strengths:
    out = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        image=init_image,
        height=init_image.size[1],
        width=init_image.size[0],
        strength=s,
        num_inference_steps=35,
        true_cfg_scale=4.0,
        generator=torch.Generator(device="cuda").manual_seed(42),
    )
    out.images[0].save(f"example_img2img_{s}.png")

Results at different strengths

Init image

0.6

0.7

0.8

0.9

1.0

2. Inpainting

Example code

import torch
from PIL import Image
from diffusers import QwenImageInpaintPipeline
from diffusers.utils import load_image

pipe = QwenImageInpaintPipeline.from_pretrained("Qwen/Qwen-Image", torch_dtype=torch.bfloat16)
pipe.to("cuda")
prompt = "cat wizard with red hat, gandalf, lord of the rings, detailed, fantasy, cute, adorable, Pixar, Disney"
negative_prompt = " "
source = load_image("https://github.com/Trgtuan10/Image_storage/blob/main/cute_cat.png?raw=true")
mask = load_image("https://github.com/Trgtuan10/Image_storage/blob/main/mask_cat.png?raw=true")

strengths = [0.6, 0.7, 0.8, 0.9, 1.0]

for strength in strengths:
    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        height=source.size[1],
        width=source.size[0],
        image=source,
        mask_image=mask,
        strength=strength,
        num_inference_steps=35,
        true_cfg_scale=4.0,
        generator=torch.Generator(device="cuda").manual_seed(42)
    ).images[0]
    image.save(f"qwen_inpainting_{strength}.png")

Results at different strengths

Init image

Mask

0.6

0.7

0.8

0.9

1.0

Before submitting

• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

• Pipelines and pipeline callbacks: @yiyixuxu and @asomoza

[a-r-r-o-w]

Thanks! The PR looks super clean :) Just have some nits about # Copied from convention we follow. Other than that, looks good to me

Could you also update the qwenimage.md in docs/ folder to reflect these pipelines:

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

Line 27 in 03c3f69

## QwenImagePipeline

[Trgtuan10]

@a-r-r-o-w I’ve finished updating qwenimage.md to reflect the pipelines and fixed the # Copied from convention. Please take a look at the changes.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[DN6]

@Trgtuan10 You just have to change from diffusers.pipelines.qwenimage.pipeline_qwenimage in the copied from statement on the pipeline methods to from diffusers.pipelines.qwenimage.pipeline_qwenimage.QwenImagePipeline

[Trgtuan10]

Thanks @DN6

[DN6]

@Trgtuan10 Pushed up minor changes to your branch for style and copy checks to pass. FYI you don't need to merge into main on your fork before opening the PR. It's better to open from your working branch.