ComfyUI Node
kubakub tracker
Finds what moves in a video and follows it, each thing with its own id. Pick the finders on the node (several at once): motion, a colour (eyedropper on the video), lights, people and things (YOLO; a pose file adds body points, a seg file outlines), anything you name in words (Grounding DINO), body / hands / face points (MediaPipe), a motif you click on a frame (SAM 2), and exact outlines inside the boxes (SAM 2). Everything comes from models already on your disk; nothing is downloaded. After a run the node shows the video with what was found; 'try this frame' checks a change on one frame in a second. Nothing connected: a built-in sample clip. Draw the result with kubakub track draw.
kubakub tracker
- video
- depth
- tracks
- audio
- sound
- tracks
- found
- near
- frames
- fps
- report
- depth
- audio
◄settings►
◄fps25►
Categorykubakub/2d/motion
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| settings | STRING | Set on the node's pills and sliders; never typed. | |
| fps | FLOAT | 251–120 | Speed of a batch of frames (a video brings its own). |
| videoopt | IMAGE,VIDEO | The video (or a batch of frames). Nothing connected: a sample clip. | |
| depthopt | IMAGE | The depth of the same video from any depth node (Depth Anything 3 / DA3 Streaming, Depth Anything 2, Marigold, Video Depth Anything ...): every tracked thing gets its depth (near / far) and track draw can cut the picture into depth layers. Which side is near: the pill on the node. | |
| tracksopt | KUBA_TRACKS | The tracks of another kubakub tracker on the same video: this tracker adds its finds to them (other finders, another colour, another sensitivity), so track draw gets one set and can join everything with lines. A kind found twice gets a number (colour, colour 2). | |
| audioopt | AUDIO | The video's sound (or music): the tracks carry how loud it is and the beats, frame by frame, so track draw's marks pulse and its lines flash on the beat (how much: 'sound' in track draw's look). It plays with the video on the node and goes on to 'audio'. | |
| soundopt | KUBA_SOUND | From kubakub sound: the sound and how it is heard (tempo, gain, release), set once for every node that listens. Connected, it comes before 'audio'. |
Outputs (8)
| Name | Type | Description |
|---|---|---|
| tracks | KUBA_TRACKS | Everything found, with ids: for kubakub track draw. |
| found | MASK | White on what was found, every frame (outline if there is one, else the box). |
| near | MASK | A soft glow around everything found, every frame (1 on it, 0 further than 1.5 times its size): the 'field' of kubakub shader, a mask for mask animate, the 'change' of the region sampler. |
| frames | IMAGE | The video's frames (for track draw's video input). |
| fps | FLOAT | The video's speed (for Create Video after track draw). |
| report | STRING | — |
| depth | IMAGE | The depth on 'depth', one scale for the whole clip, near = white (for track draw's depth). Black without one. |
| audio | AUDIO | The sound, cut to the video (silence without one): for Create Video. |