About this tool
It uses U²-Net, an open-source object segmentation model, run with ONNX Runtime in WebAssembly. Everything happens on your device.
The first time, it downloads the model (about 20 MB in total) and then keeps it saved, so it can work offline. Each photo takes a few seconds, depending on your device.
It works best with a well-defined subject, such as a person, a product, or a pet. Hair, fur, and very fine edges can come out rough, and in photos with several objects it may pick only one. For plain backgrounds, “Remove Solid Color Background” gives cleaner edges.