Diffusion Editing with Soft Mask: Pixel Level Redo of Image and Video with Adjustable Strength

arXiv cs.GR details SoftPaint, a new zero-shot sampling method for diffusion editing that aims to move beyond coarse mask-based workflows. The paper frames the problem clearly: prompt- and reference-guided editors can change too much at once, while training for fine-grained control usually needs expensive pixel-level annotations.

SoftPaint uses soft masks to specify per-pixel edit strength, creating a continuous range from untouched source pixels to fully regenerated regions. The authors say the sampler is based on Langevin iterations, works without gradients, and is designed to plug into both image and video diffusion backbones. That makes the approach interesting not just for still-image touchups, but for temporal editing where consistency is usually the hard part.

For game teams, the practical angle is obvious: more controllable concept iteration, marketing art cleanup, and video asset editing without having to rebuild a pipeline around supervised mask data. The method is still research-stage, but the combination of zero-shot operation and memory efficiency is the kind of thing that could matter if diffusion tools keep moving into production-facing content workflows.