Cutout

How it works: the old way of doing this was picking a background color and deleting everything close to it, which falls apart the moment a subject has hair, fur, glass, smoke or anything semi-transparent, and falls apart completely when the background is not one flat color. This uses a segmentation model that was trained to understand what the subject of a photo is, so it produces a soft mask with real partial transparency at the edges rather than a hard cut. Strands of hair survive. The whole thing runs inside this tab: the model downloads once from a public host and then your photos never leave your machine. Best results come from a subject that is reasonably in focus and clearly separated from what is behind it \u2014 a busy photo where the subject blends into the background is hard for a person to trace too.