Meta’s vision model, repackaged for the clipboard

What it does

Magic Copy is a browser extension for Chrome and Firefox, plus a Figma plugin, that uses Meta’s Segment Anything Model to isolate a foreground object in an image and copy it to your clipboard. It follows the same segmentation procedure as the official SAM demo.

The interesting bit

Rather than displaying the result on a canvas, the extension intercepts the mask and places it directly on the system clipboard. The author also documents how to self-host the vision transformer backend if you prefer not to use Meta’s default embedding service.

Key highlights

• Available as a browser extension and a Figma plugin.
• Offloads embedding generation to a server-side vision transformer; defaults to the same third-party service as the SAM demo.
• Provides a server-example directory with a proof-of-concept backend and Dockerfile for self-hosting.
• Expects a tightly specified API response: a JSON array of length 1 containing a base64-encoded embedding tensor of shape (1, 256, 64, 64).

Caveats

• Chrome Web Store and Firefox Add-ons listings may still be pending review at any given moment.
• The provided self-hosting example is explicitly not production-ready and serves mainly to document the input/output format.

Verdict

Worth a look if you frequently grab image assets from reference photos and want a faster path than manual masking. If you need guaranteed offline operation or a hardened backend, you’ll have to build beyond the included proof of concept.