Proof of Use. An independent field guide · No. 01
Back to the collection
Images & Video

SAM 2

A promptable segmentation model that produces object masks in images and tracks them through video.

segmentationvideo maskscomputer visionimage editing
Source-reviewedNot hands-on testedSources checked

Start with an idea

What you can make

01

Create a moving-subject mask for a video-editing or annotation pipeline, then review and refine it.

02

Mark a product in a photograph to get a starting mask for compositing or background editing.

Creator demonstration

See the documented workflow

The repository links this official interactive demo; its page resolved, but interaction and inference were not tested.

Meta creator demo

Before you begin

What it takes to run

Python + CUDA/checkpoints

  • 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.
Runs through / on

Linux · Windows via WSL recommended

The real tradeoffs

Cost & licensing

Running costs

No model or core code license fee; local compute costs apply.

The documented installation targets a GPU machine. GPU rental, storage and serving infrastructure are separate expenses.

Code license

Apache-2.0; BSD-3-Clause for the optional cc_torch component

Read the code license

Models & other terms

Apache-2.0 for the SAM 2 checkpoints

Demo fonts have separate SIL Open Font License terms; dataset terms are separate from model licensing.

Know the 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.

Follow the evidence

Sources & verification

Official sources were reviewed on October 14, 2026. We did not install, run or benchmark this project. This is a source-check date, not a release date.

Links and project terms can change. These sources document the claims above; they are not a safety certification. Read our full method.

Keep exploring

A few other possibilities

All tools

09Images & Video

ComfyUI

A visual node editor for building repeatable AI image, video and other media workflows.

Try it for

Build a reusable image-generation workflow, then repaint selected areas with an inpainting model.

Read entry Official examples
Setup · Moderate
Desktop app + model workflow
Cost to consider
Local compute; optional paid cloud/API

10Images & Video

Diffusers

A Python library for wiring diffusion models into image, video and audio applications.

Try it for

Prototype a text-to-image feature in your own app with a selected model and repeatable parameters.

Read entry Official examples
Setup · Advanced
Python + model pipeline
Cost to consider
Local or rented compute