{
  "id": "sam-2",
  "name": "SAM 2",
  "category": "images_video",
  "tags": [
    "segmentation",
    "video masks",
    "computer vision",
    "image editing"
  ],
  "description": "A promptable segmentation model that produces object masks in images and tracks them through video.",
  "useCases": [
    "Create a moving-subject mask for a video-editing or annotation pipeline, then review and refine it.",
    "Mark a product in a photograph to get a starting mask for compositing or background editing."
  ],
  "repoUrl": "https://github.com/facebookresearch/sam2",
  "docsUrl": "https://github.com/facebookresearch/sam2#readme",
  "demo": {
    "url": "https://sam2.metademolab.com/demo",
    "label": "Meta creator demo",
    "evidenceType": "creator_demo",
    "note": "The repository links this official interactive demo; its page resolved, but interaction and inference were not tested."
  },
  "license": {
    "code": "Apache-2.0; BSD-3-Clause for the optional cc_torch component",
    "weights": "Apache-2.0 for the SAM 2 checkpoints",
    "caveat": "Demo fonts have separate SIL Open Font License terms; dataset terms are separate from model licensing.",
    "url": "https://github.com/facebookresearch/sam2/blob/main/LICENSE"
  },
  "pricing": {
    "summary": "No model or core code license fee; local compute costs apply.",
    "caveat": "The documented installation targets a GPU machine. GPU rental, storage and serving infrastructure are separate expenses."
  },
  "platforms": [
    "Linux",
    "Windows via WSL recommended"
  ],
  "setupDifficulty": "advanced",
  "setup": {
    "summary": "The documented setup uses Python 3.10+, PyTorch 2.5.1+, matching TorchVision/CUDA, downloaded checkpoints and optional notebook dependencies.",
    "hardware": "The official setup targets CUDA GPUs; skipping the optional CUDA extension disables a mask-cleanup step."
  },
  "limitations": [
    "Produces masks rather than finished video edits; prompts and output review are still needed.",
    "Compatibility between checkpoint versions and installed code matters; hardware-specific speed claims are not guarantees for your machine."
  ],
  "evidenceCheckedDate": "2026-10-14",
  "testStatus": "not_tested",
  "sources": [
    {
      "title": "Official repository, examples and checkpoint terms",
      "url": "https://github.com/facebookresearch/sam2",
      "claims": [
        "image/video segmentation",
        "checkpoint license",
        "demo link"
      ],
      "evidenceCheckedDate": "2026-10-14"
    },
    {
      "title": "Core license",
      "url": "https://github.com/facebookresearch/sam2/blob/main/LICENSE",
      "claims": [
        "Apache-2.0"
      ],
      "evidenceCheckedDate": "2026-10-14"
    },
    {
      "title": "Optional component license",
      "url": "https://github.com/facebookresearch/sam2/blob/main/LICENSE_cctorch",
      "claims": [
        "BSD-3-Clause"
      ],
      "evidenceCheckedDate": "2026-10-14"
    },
    {
      "title": "Installation requirements",
      "url": "https://github.com/facebookresearch/sam2/blob/main/INSTALL.md",
      "claims": [
        "Linux/WSL",
        "versions",
        "CUDA extension caveat"
      ],
      "evidenceCheckedDate": "2026-10-14"
    },
    {
      "title": "Official video example notebook",
      "url": "https://github.com/facebookresearch/sam2/blob/main/notebooks/video_predictor_example.ipynb",
      "claims": [
        "documented video workflow"
      ],
      "evidenceCheckedDate": "2026-10-14"
    },
    {
      "title": "Official interactive demo",
      "url": "https://sam2.metademolab.com/demo",
      "claims": [
        "creator demo destination"
      ],
      "evidenceCheckedDate": "2026-10-14"
    }
  ],
  "categoryLabel": "Images & Video",
  "card": {
    "cost": "Local or rented GPU compute",
    "setup": "Python + CUDA/checkpoints"
  },
  "setupRequirements": [
    "The documented setup uses Python 3.10+, PyTorch 2.5.1+, matching TorchVision/CUDA, downloaded checkpoints and optional notebook dependencies.",
    "The official setup targets CUDA GPUs; skipping the optional CUDA extension disables a mask-cleanup step."
  ],
  "evidence": {
    "status": "documentation_reviewed",
    "demoType": "creator_demo",
    "independentlyTested": false,
    "note": "Official source documentation reviewed; no repository installed or executed."
  },
  "releaseDate": null,
  "releaseDateNote": "Not researched; the evidence check date does not indicate a recent release."
}
