Segment videos with Meta SAM 3.1. Text prompt initializes detection; optional per-frame points, box, or mask refine tracking.
