Background removal typically means uploading your image to remove.bg or a similar service, which then applies a paid AI model. This tool runs the same class of model — U-Net-based semantic segmentation — locally in your browser using ONNX Runtime for Web and the open-source u2netp model. Quality is very close to remove.bg for typical subjects (people, products, animals); your image never leaves the tab.
How the AI runs locally
The u2netp model (~4 MB) is downloaded once and cached in your browser. Inference runs on your device's CPU or GPU (WebGL/WebGPU) using ONNX Runtime Web. Processing a 1080p photo takes 3–8 seconds on modern hardware. No server, no upload, no queue.
What works and what doesn't
Works well: people, products against solid backgrounds, animals, food, single objects. Works okay: fine details like hair, complex backgrounds. Doesn't work: multi-subject scenes, semi-transparent objects (glass, smoke), very low contrast subjects.
Post-processing
After removal, the tool offers edge refinement (smooth the segmentation edge for better anti-aliasing), replace background with a color/image, and add a shadow. Great for product photography where you need consistent backgrounds.