GitHub radar
SAM 3 by Meta: Segment Anything in Video by Text
Facebook (Meta)released SAM 3 — a model that finds, outlines, and tracks any object in photos and videos from a text description or a click, and fits on a regular laptop.
SAM 3 is Meta's unified foundation model for segmenting objects in images and videos. It responds to text prompts — describe what you want to isolate, and the model draws an accurate mask. Visual prompts (points, bounding boxes, drawn masks) also work and can be combined with text. The model tracks objects across video frames without needing per-frame annotation. It achieves 75–80% of human performance on the SA-CO benchmark, which covers 270K unique visual concepts — over 50 times more than previous benchmarks. The model has 0.9B parameters and weighs about 1.7 GB in full precision.
Why a vibe-coder should care
If you work with photos or video content, SAM 3 lets you cut out any object by simply describing it — no manual masking, no per-frame work. For video editors, content creators, and anyone building visual workflows with an AI agent, this is a practical local tool: small enough to run on any laptop, accurate enough to use in production.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up SAM 3 by Facebook (Meta)for me: https://huggingface.co/facebook/sam3 — use the transformers library, load the model and show me how to segment an object in my photo using a text description.
Обычный ноутбук от 8 ГБ оперативки — модель весит около 1.7 ГБ в оригинальном формате.
Open on Hugging Face▌ More finds