Skip to main content
Motion Track

Motion Track · screenshot coming soon

​
What this node does

Motion Track follows one or more points across your clip. You click points directly on the node’s video preview; on ▶ Run it does pattern-matching from frame to frame and returns the tracked paths as text data. Optionally it also solves a single motion transform (translation, similarity, or perspective corners) from those points so downstream nodes can pin, stabilize, or corner-pin to the motion. Unlike the other masking nodes, its output is COMFYTV_TEXT (track data as JSON), not a video. Video in is COMFYTV_VIDEO. To use track text with native ComfyUI nodes, bridge via ComfyTV/Bridge — see https://github.com/jtydhr88/ComfyTV/blob/main/docs/bridges.md.

​
The on-node editor

The card overlays a canvas on the video:
  • Click empty space to add a track point (a green cross with a numbered pattern box).
  • Double-click a point to remove it.
  • The first point also shows a dashed yellow search box.
  • The overlay draws each point’s pattern box (the template being matched) and, for point 1, the search box (how far it hunts each frame).
  • A Clear action removes all points.
You must place at least one point before running, or the node raises “click at least one point on the node first.”

​
Parameters

​
points

The clicked track points, stored as JSON [{x,y},...]. Written by the on-node editor as you click. Default empty. (If points is empty but point_x/point_y are set, a single point is used.)

​
point_x / point_y

The first point’s coordinates, 0–8192, default 0. These mirror point 1 from the editor.

​
solve

What transform to fit from the tracked points. Options: none, translation, similarity, perspective. Default none. similarity needs 2+ points and feeds Transform/Composite; perspective needs 4+ points and feeds Corner Pin. none returns raw tracks only.

​
t_start / t_end

The time window (seconds) to track. t_start 0–3600 default 0; t_end default -1 (meaning end of clip), range -1–3600, step 0.05.

​
pattern

Pattern half-size in pixels — the template box matched each frame. Range 4–64, default 12. Larger patterns are more distinctive but slower and less tolerant of deformation. Search radius in pixels — how far the tracker looks for the pattern each frame. Range 8–128, default 24. Increase for fast motion.

​
Outputs

​
Tips

  • Pick a high-contrast, non-repeating feature to track; flat or repetitive areas cause jitter (watch the confidence in the result).
  • Match search to the motion speed — too small and the tracker loses fast movement, too large and it slows down and may lock onto the wrong feature.
  • For perspective corner-pin, place points on the four corners of the planar surface.
  • Mask Propagate — carries a mask across motion (uses a similar solve internally)
  • Roto Mask / Shape Mask — produce mattes you might pin using this track