Gray Paint Tracker

Wan VACE inpainting needs the thing-you-want-replaced painted over in gray, on every frame, with a matching mask. For a moving subject that means the gray region has to follow the subject - a static blob is useless the moment anything moves. Gray Paint Tracker is the node that solves both problems with one interaction: you paint a gray region on the first frame, it tracks a point through the clip, and the gray rides along with it. Outputs: the gray-composited video and a clean binary mask, both VACE-ready.

This is the kind of node that only makes sense if you've actually built a VACE inpaint pipeline and felt how much of the work is in the prep - compositing gray-over-original, tracking it, re-doing the mask every frame. The author of this pack clearly lives in that world; this node and the keyframe builder around it are the heart of why people know the pack's Wan tooling.

How it works

The interaction is deliberately two-run:

1. Run once with your video connected. The first frame loads onto a paint canvas on the node.
2. Paint a gray region over the subject, set your tracking anchor, and run again.

The painted mask (a base64 PNG in the hidden mask_data widget) is then carried through the whole clip. Tracking uses pyramidal Lucas-Kanade optical flow - the same robust point-tracking family as the pack's Point Tracker - with a template-match fallback for larger motion. The defaults (window_size 31, pyramid_levels 4, iterations 10) handle most footage; raise window_size toward 200+ and pyramid_levels toward 6–8 for fast-moving subjects. smoothing (0–0.9) damps jitter in the tracked path, which is often the difference between a clean mask and a wobbling one.

The knobs that actually matter:

• anchor_mode - centroid (default) tracks the center of the painted region; manual tracks a point you clicked with "Set Anchor" on the canvas. Manual is better when the subject's meaningful center isn't the paint blob's centroid.
• gray_value (default 0.5) - the fill gray. Leave it: 0.5 is exactly what VACE's filler expects.
• expand_pixels - grows the mask outward with a hard edge before compositing. A little expansion keeps VACE from repainting right on the subject's boundary.
• track_enabled - off gives you a static mask held across all frames, for the rare case the subject doesn't move.

Outputs: images (video with gray composited over the moving region), masks (binary, white where gray was applied - VACE-ready), and tracking_preview (a debug overlay with crosshairs; 1×1 placeholder unless you turn draw_preview on, which is slower and meant for diagnosis).

Install

Search "Trent Nodes" in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/TrentHunter82/TrentNodes.git
cd TrentNodes
pip install -r requirements.txt

No models - the tracking is OpenCV optical flow (the pack's opencv-python requirement) and the compositing is GPU tensor math. Restart ComfyUI after installing. (The author has noted Manager install flakiness from an early repo rename; manual clone is the reliable path.)

Common issues

• The gray drifts off the subject. That's tracking loss. Raise window_size and pyramid_levels for bigger motion, or switch anchor_mode to manual and anchor on a high-contrast feature (an eye, a corner).
• Nothing shows until you run twice. Yes - the paint step happens on the node's canvas between runs. That's by design, not a bug.
• The mask flickers. Increase smoothing, and check expand_pixels isn't so small that tracking noise shows at the edges.
• Forgetting draw_preview is a speed trap. Leave it off in production; it's a debug aid.

For "erase this thing from my video with minimal fuss," this is the node that makes VACE inpainting feel like an edit rather than a pipeline. Paint once, track automatically, feed VACE, done.