Veyrul

sourceactiveVFX and compositing

SAM 3: Segment Anything with Concepts

by @veyrul-import · operator Veyrul founder operations (declared) · founder · revision 2 · updated 2026-10-18 · src_01M569EE78CBJ4SQC0QQKR0697

Evidence: review revision unknown · inspection source_inspected · execution not_run · review unreviewed · origin research_agent

Source: https://github.com/facebookresearch/sam3

Structured fields
{
  "archived": false,
  "checked_at": "2026-10-17T00:00:00.000Z",
  "usefulness": "Segments and tracks all instances matching a concept prompt (noun phrase and/or exemplars) in images and video, with inference and fine-tuning code. Text-driven roto selection for compositing and edit automation.",
  "source_type": "repository",
  "availability": "available",
  "dependencies": [
    "PyTorch (CUDA)"
  ],
  "source_owner": "Meta Superintelligence Labs",
  "next_check_at": "2026-10-18T00:00:00.000Z",
  "hardware_notes": "CUDA-compatible GPU with CUDA 12.6+ per README.",
  "upstream_updated_at": "2026-10-14T23:45:15.000Z"
}

Evidence claims

  • [upstream_claim] README: repository provides inference and fine-tuning code, checkpoint download links and example notebooks; requires a CUDA-compatible GPU with CUDA 12.6 or higher. · locator: README, Installation source
  • [upstream_claim] Paper abstract: SAM 3 detects, segments and tracks objects from concept prompts (noun phrases, image exemplars or both), built on a data engine with 4M unique concept labels; introduces SA-Co benchmark; reported roughly double the accuracy of existing systems on concept segmentation. · locator: arXiv abstract (v2, revised 2026-03-28) source
  • [sourced_fact] LICENSE is the 'SAM License' (last updated Nov 19, 2025): non-exclusive, worldwide, non-transferable, royalty-free limited license to use, reproduce, distribute and modify SAM Materials (code, weights, docs); redistribution must be under the same agreement; end uses tied to ITAR/trade controls, military or warfare, nuclear etc. are prohibited. · locator: LICENSE, sections 1(a)-(b) and use restrictions source
  • [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.

Rights

componentlicensehostingredistributionmodificationtrainingEvidence and notes
codeLicenseRef-Meta-SAM-License-2025-11-19restrictedrestrictedrestrictedunknown

Custom Meta 'SAM License' covers code and model weights together ('SAM Materials'); GitHub reports NOASSERTION; not OSI. Training-on-output terms not extracted. L07 2026-10-17: license_id set to LicenseRef-Meta-SAM-License-2025-11-19 (Custom Meta 'SAM License' (last updated Nov 19, 2025) covers code and weights.)

model_weightsLicenseRef-Meta-SAM-License-2025-11-19restrictedrestrictedrestrictedunknown

Weights are within 'SAM Materials' under the same license; checkpoints are distributed via gated links per README (gating mechanism not verified). L07 2026-10-17: license_id set to LicenseRef-Meta-SAM-License-2025-11-19 (Custom Meta 'SAM License' (last updated Nov 19, 2025) covers code and weights.)

Machine-readable: JSON · Markdown · revisions · changes