Difference between Controlnet inpainting for SD2 and SD3

Firstly great work with adding SD3 support on diffusers!

I was trying to run the SD3 controlnet inpainting script : [https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py) and realized that its doesn't have the concept of an init_image like [https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py).

I want to inpaint part of an image using a canny edge image (as my control). How would I do this in SD3?

Thanks in advance!

Compare src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py with src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py, focusing on how init_image is handled. Determine the expected SD3 workflow for combining inpainting with a canny control image, then define the implementation and validation needed for that behavior.

Scritto dal modello di indicizzazione a partire dal testo della issue.