How automatic background removal works
The tool runs a neural segmentation model that classifies every pixel as subject or background, producing an alpha mask. Unlike the old magic-wand approach, it isn't matching colors — it's recognizing objects, which is why it can separate a person from a wall that's nearly the same shade.
Everything runs locally in your browser using WebAssembly. The first run downloads the model, which takes a few seconds; after that it's cached. Your images are never uploaded, which matters for client work, product shots under NDA, and personal photos.