ComfyUI Node
IAMCCS SCAIL Identity Seeder
One mask per person from point or box prompts -> SAM3_VideoTrack.initial_mask. Guarantees one tracked object (one colour) per subject, bypassing the auto-detector's overlap-merging. Leave SAM3_VideoTrack conditioning disconnected.
IAMCCS SCAIL Identity Seeder
- model
- image
- bboxes
- initial_masks
◄modepoints►
◄refine_iterations2►
◄points►
CategoryIAMCCS/video/SCAIL-2
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | SAM3 model (same one feeding SAM3_VideoTrack). | |
| image | IMAGE | Frame to segment, at the resolution the tracker runs on (resized reference image, or pose video's first frame). | |
| mode | COMBO | points | points: one positive click per person (from a PointsEditor). boxes: one bounding box per person. |
| refine_iterations | INT | 20–5 | SAM decoder refinement passes per object (0 = raw prompt mask). |
| pointsopt | STRING | points mode: JSON list, one positive point per person, e.g. [{"x":120,"y":210},{"x":480,"y":205}]. Order sets identity order. | |
| bboxesopt | BOUNDING_BOX | boxes mode: one bounding box per person. Order sets identity order. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| initial_masks | MASK | — |