SD3 Image-to-Image and Inpainting

[RyanJDick]

Summary

Add support for SD3 image-to-image and inpainting. Similar to FLUX, the implementation supports fractional denoise_start/denoise_end for more fine-grained denoise strength control, and a gradient mask adjustment schedule for smoother inpainting seams.

Example

Workflow

Result

QA Instructions

• Regression test of text-to-image
• Test image-to-image without mask
• Test that adjusting denoising_start allows fine-grained control of amount of change in image-to-image
• Test inpainting with mask
• Smoke test SD1, SDXL, FLUX image-to-image to make sure there was no regression with the frontend changes.

Merge Plan

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[RyanJDick]

I ran through the tests listed in the PR description. Everything works, although inpainting/outpainting quality with SD3 is clearly worse than other model architectures (SD1, SDXL, FLUX). (The example in the PR description is cherry-picked.) I think it's ok to merge, but we may want to loop back and spend more time iterating on the inpainting implementation at some point.