Diffusion Editing with Soft Mask: Pixel Level Redo of Image and Video with Adjustable Strength
SoftPaint uses soft masks to vary edit strength pixel by pixel in diffusion image and video editing.
The paper frames the problem as existing prompt- and reference-guided diffusion edits being too coarse for fine spatial control. Its method is zero-shot, avoiding the pixel-wise annotations that supervised soft-mask control would require. The authors describe a Langevin-iteration-based sampler that follows per-pixel mask strengths, from preservation to full re-synthesis. They say it is gradient-free, memory-efficient, and works across multiple image and video diffusion backbones. HF Daily Papers' note
The paper frames the problem as existing prompt- and reference-guided diffusion edits being too coarse for fine spatial control. Its method is zero-shot, avoiding the pixel-wise annotations that supervised soft-mask control would require. The authors describe a Langevin-iteration-based sampler that follows per-pixel mask strengths, from preservation to full re-synthesis. They say it is gradient-free, memory-efficient, and works across multiple image and video diffusion backbones. HF Daily Papers' note
score 4