SAM 3 Video
AI generationMetaAI rotoscoping - segment and track objects across video frames with text prompts.
- Runs
- On our servers
- Typical time
- 30-120s
- Inputs
- 2
- Outputs
- 1
Inputs
| Port | Type | ID |
|---|---|---|
| Prompt | text | prompt-in |
| Video | video | video-in |
Outputs
| Port | Type | ID |
|---|---|---|
| Video | video | video-out |
Settings
| Setting | Type | Default | Options |
|---|---|---|---|
| Detection ThresholdLower finds more objects but may include false matches; higher keeps only confident detections. | slider | 0.3 | 0 to 1 |
| Output ModeCutout returns the subject on a transparent background. Alpha Mask returns a black-and-white matte to use in Merge, Keymix, or Copy. | select | true | Cutout (object only), Alpha Mask (B&W matte) |
Prompting
What to track ("person" - commas track several: "car, road"). All subjects return in ONE video; for separate mattes, run once per subject...
Maximum prompt length: 500 characters.
What this connects into (18)
Computed from the port compatibility rules, so it stays correct as nodes change.
Node ID sam3-video. Open it in Studio with Tab, then type the name.