In-Browser AI Background Removal: How On-Device WebAssembly Protects Photo Privacy

Traditional background removal services send your uploaded image to a remote Python backend running PyTorch. While effective, this architecture exposes your data to server logging, network transit eavesdropping, and vendor storage policies. Forma's 'Cutout Subject' tool leverages optimized neural segmentation models compiled to WebAssembly. The model executes directly inside your browser tab: - Zero bytes are transmitted to any server. - The segmentation mask is mapped directly to the alpha channel of an HTML5 Canvas. - Transparent pixels render instantly against a checkered canvas background.

Evaluating edge details around hair, product outlines, or intricate shapes requires comparing against the original unedited source photo. Forma includes two precision comparison modes: 1. Split Slider: An interactive draggable divider that reveals the original photo on the left and the transparent cutout on the right. Both sides remain locked to identical zoom and crop geometry. 2. Hold for Before: A dedicated keycap button (with an eye icon) that temporarily restores the 100% unedited original when held, then snaps back to the cutout when released.

If you are preparing product catalog photos or team member headshots, repeating background removal manually on 20 individual files is tedious. In Forma, clicking 'Apply to All' after cutting out a subject automatically triggers sequential neural segmentation across every valid image in your queue, applying consistent transparency and canvas settings across the entire set.