Sam 3
/v1/runSam 3 Video Rle is Meta's video-to-video AI model. Transform, enhance, and edit video content using text prompts - from style changes to object manipulation and scene modification.
Request body
Submit an async generation request. The model field selects the model; other fields are model-specific input parameters.
Model identifier. Set to meta/sam-3/video-rle.
Default: meta/sam-3/video-rle
Text prompt for segmentation. Use commas to track multiple objects (e.g., 'person, cloth').
Default:
The URL of the mask to be applied initially.
The URL of the video to be segmented.
Apply the mask on the video.
Default: false
Return per-frame bounding box overlays as a zip archive.
Default: false
List of box prompts with optional frame_index.
Default:
List of point prompts with frame indices.
Default:
Detection confidence threshold (0.0-1.0). Lower = more detections but less precise. Defaults: 0.5 for existing, 0.7 for new objects. Try 0.2-0.3 if text prompts fail.
Range: 0.01 to 1
Default: 0.5
Frame index used for initial interaction when mask_url is provided.
Default: 0
Response Schema
The submit endpoint returns a run response. If its status is pending or running, poll GET /v1/run/{id} with the returned opaque ID until it reaches a terminal state.
Opaque SandBase run identifier. Use it exactly as returned; no prefix is guaranteed.
Current public run status.
Allowed values: pending, running, completed, failed, timeout
Public SandBase model name used for this run.
Present only for completed runs. Each object is capability-specific; inspect the selected model schema for its fields.
Present only for failed or timeout runs. Contains a public error type and sanitized message.
Stable public error category.
Sanitized error message safe to show to clients.
Usage details when available.
Model capabilities
Capabilities declared by the model registry.
Default: video-to-video
Execution mode declared by the model registry.
Default: async