Masked video inpainting on Wan 2.2 VACE. You paint the subject on a single frame
and MatAnyone propagates that mask across the whole clip, with matting quality
edges on fine detail like hair.

How it works

This runs twice. The first run stops at a preview so you paint the mask on frame
zero. The second run propagates it and renders. If you want text prompted masking
with no painting and a single pass, use the SAM3 version instead.

Painted masking is slower to set up, but it gives you exact control over what is
covered, and MatAnyone handles soft edges better than a detector can.

Features

INT8 ConvRot models throughout, native on Ampere and newer, no FP8 emulation
Chunked sampling loop, so clip length is not bounded by VRAM
Optional 4x upscale with per source type model recommendations
Runs on 4 step Lightning distills, a few seconds per chunk rather than minutes
Full on canvas documentation covering every knob and all download links