Is my photo uploaded to remove the background?
No. The segmentation model (a one-time download) runs inside your browser using WebAssembly. The photo never leaves your device - once the model has loaded, the tool works with your internet disconnected.
Why does the first image take longer?
The first use downloads the AI model and warms it up. Your browser caches it, so later visits skip the download and go straight to the few seconds of on-device processing.
How good are the edges?
Clean subject outlines come out well. Very fine structures - individual hairs, fur, semi-transparent objects - are approximated. Cloud services running much larger models can resolve finer detail; the trade-off here is that your photo stays on your device.
What AI model does this actually use?
A neural network from the U²-Net family - the open-source salient-object segmentation architecture behind many popular cutout tools - running on ONNX Runtime compiled to WebAssembly. We use the compact variant so the download stays small and processing stays quick; it's the same technology cloud services run, executed on your device instead of theirs.
Why is the output PNG only?
PNG is the format that supports the transparency the tool produces. If you pick a solid white or black background you can convert the result to JPG afterwards for a smaller file.
Can I remove backgrounds from many photos at once?
Yes - batch processing with ZIP export is part of GenClean Pro. Each photo is processed locally, one after another, with the same settings.