Click an object. The real SAM masks it.
This is not a video or an API call. The picture's embedding was computed by Meta's Segment Anything ViT-B encoder on a Raspberry Pi, and the actual SAM mask decoder — 4 million of the model's real weights, quantized to 8-bit — is running in your browser right now via WebAssembly. model runs locally
Keyboard: focus the picture, then ←↑↓→ move the crosshair (hold Shift for fine steps), Enter places a point, Shift+Enter removes from the mask, Backspace undoes the last point and Esc clears. Tab leaves the picture; the page scrolls again once you do.
Tap or click the thing you want. Switch to － remove and tap what you don't want to refine the mask (shift-click does the same) · move the mouse to preview before clicking. Then cut it out and save the sticker.
Want this on your own photo? EfficientSAM runs the whole model in your browser →
About this demo
masks taught it
SAM was trained on SA-1B: 11 million images, 1.1 billion masks — ~400× more masks than any prior dataset. That's why it segments objects it has never seen.
halves of one model
A heavy image encoder (ran once per photo, on a Raspberry Pi 5, offline) and a featherweight prompt decoder — the part answering your taps in tens of milliseconds, locally.
in your tab
What you downloaded: the WebAssembly runtime (~3 MB compressed) and the genuine sam_vit_b decoder exported to ONNX and 8-bit quantized (~3 MB) — the exact architecture from Meta's own browser demo — plus ~2 MB of embedding per photo.
fully open
Code and weights are open source. Read the paper, browse the repo, or meet the whole family on the Universe map.