Video Inpainting

Routes video inpainting requests to the right RunComfy endpoint via the CLI. Defaults to Wan 2-7 edit-video for prompt-driven region removal (watermarks, passing objects, overhead wires), falls back to Lucy Edit for identity-stable swaps or Seedream for frame-by-frame sequences. The prompting is spatial language, not pixel masks, so "remove the watermark in the bottom-right" works but surgical edits need a ComfyUI workflow. Honest tradeoff: CLI accessibility versus precision. If you're removing a logo from 200 frames of footage and don't want to open a GUI, this is the path. If temporal consistency matters more than speed, you'll end up in a mask propagation workflow anyway.