Why this one does not upload your photo
Almost every background remover on the web works the same way: you upload the image, a server runs a segmentation model on it, and the result comes back. That means a copy of your photo exists on someone else's hardware, subject to their retention policy and their breaches.
This tool inverts that. The model is downloaded to your browser and runs there, on your machine, using your processor. Your photo is read into memory, segmented and composited without a single byte crossing the network. That is why there is a one-time download — and why there is no upload at all.
What the model is actually doing
It is a segmentation network: it looks at the image and predicts, for every pixel, how likely that pixel is to belong to the foreground subject. Those predictions become an alpha channel — fully opaque where it is confident about the subject, fully transparent where it is confident about the background, and partially transparent at the edges, which is what makes hair and fur look reasonable rather than cut out with scissors.
Where it struggles
Three situations defeat every automatic tool. Fine strands of hair against a busy background, because the model has to guess at the sub-pixel level. Subjects the same colour as what is behind them, because there is genuinely little to separate. And transparent or reflective objects — glass, water, a windscreen — where the correct answer is ambiguous even to a person.
If the result disappoints, the fix is usually in the photograph rather than the settings: more separation between subject and background, and more even light on both.
Choosing an output background
Transparent PNG is right when the cut-out is going into a design, a slide or a listing where something else sits behind it. A solid colour is what you want for identity and visa photographs, which almost always specify plain white or light blue, and for marketplace product images, where a pure white background is commonly mandatory.