[core] support QwenImage Edit Plus in modular

[sayakpaul]

What does this PR do?

Test code:

import torch
from diffusers import ModularPipeline
from diffusers.utils import load_image

# repo_id = "Qwen/Qwen-Image-Edit"
repo_id = "Qwen/Qwen-Image-Edit-2509"

pipeline = ModularPipeline.from_pretrained(repo_id)
pipeline.load_components(torch_dtype=torch.bfloat16)
pipeline.to("cuda")

guider_spec = pipeline.get_component_spec("guider")
guider = guider_spec.create(guidance_scale=4.5)
pipeline.update_components(guider=guider)

image = load_image(
    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/yarn-art-pikachu.png"
).convert("RGB")
prompt = (
    "Make Pikachu hold a sign that says 'Qwen is awesome', yarn art style, detailed, vibrant colors"
)
image = pipeline(
    image=image, 
    prompt=prompt, 
    negative_prompt=" ",
    num_inference_steps=40,
    generator=torch.manual_seed(0),
).images[0]
image.save("qwenimage_edit_plus_modular.png")

Result:

[sayakpaul]

Hopefully this is okay?

[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.

[sayakpaul]

This is where this step differs from the existing RoPE input step.

[asomoza]

thanks!, this part is just for this model and when doing the denoise step alone and passing the image latents, it gives an error since the images_sizes input is missing.

ValueError: Required input 'image_sizes' is missing

[sayakpaul]

Can you test again?

[asomoza]

fixed, it works!

[sayakpaul]

@asomoza if you tests are looking good, maybe we could merge it? If so, could you review and approve as you go?

[asomoza]

sure, I wanted to test and review with more scenarios but that will take more time and probably they're going to work, everything else looks good to me, if I find something after we can fix it in a following PR.

[sayakpaul]

Sure! Plese hit me up if any issue comes up.