video3d

Sam 3

Eight-endpoint SAM 3 family for detection, segmentation, tracking, embeddings, and 3D reconstruction.

Studio is in early access; availability resolves per project after sign-in.

Endpoints: 8
Eligible: 8
Input schemas imported: 8 / 8
Delivery: fal.ai
Weights: Not stated
Developer: Not stated in source

Overview

SAM 3 unifies detection, segmentation, and tracking across images and video through multi-prompt inputs covering text, points, and boxes. Video endpoints segment continuously and track objects across frames. The 3D line reconstructs bodies as GLB meshes, objects with texture, and aligned full scenes from single images. An embed endpoint extracts segmentation vectors.

Capabilities

Multi-prompt detection and segmentation
Continuous video segmentation
Cross-frame object tracking
Single-image 3D body, object, and scene

Best for

AR and VR scene composition
Game environment assembly
Digital twin creation

Use cases

Clutter-robust 3D capture
Human pose GLB export
Embedding pipelines

Endpoints

8 eligible. Cataloged is not the same as executable: Studio resolves which endpoints a project can run.

Endpoint	Task	Catalog status	Inputs
`fal-ai/sam-3/3d-align`	3d generation	Eligible	5 fields · requires image_url, body_mesh_url
`fal-ai/sam-3/3d-body`	3d generation	Eligible	5 fields · requires image_url
`fal-ai/sam-3/3d-objects`	3d generation	Eligible	9 fields · requires image_url
`fal-ai/sam-3/image`	image generation	Eligible	12 fields · requires image_url
`fal-ai/sam-3/image-rle`	image generation	Eligible	12 fields · requires image_url
`fal-ai/sam-3/image/embed`	image generation	Eligible	1 fields · requires image_url
`fal-ai/sam-3/video`	video editing	Eligible	8 fields · requires video_url
`fal-ai/sam-3/video-rle`	video editing	Eligible	9 fields · requires video_url

Questions

Which modalities?

Image, video, embeddings, and three 3D endpoints.

Single or multi-view 3D?

Single-image convenience over multi-view rigs.