Is my image uploaded anywhere?
No. Almost every free background remover sends your photo to a server, runs the model there and sends a cutout back, which means your image sits on someone else's disk for as long as their retention policy says. This one does the opposite: the model is downloaded to your browser and your photo is scored on your own machine. Open your browser's network tab and watch — the only requests are for the model files, and none of them carry your picture. That also means it works with the wifi off once the model has arrived.
Why does the first cutout take a while?
Because the engine has to arrive before it can run. That is about 17 MB: a WebAssembly build of the ONNX runtime plus the segmentation weights. It is fetched once, the first time you pick an image, and then reused for every photo in that visit. Opening this page costs you nothing at all — no model, no runtime, nothing heavy loads until you actually hand it a picture. On a slow connection the first run is a real wait, so the bar counts real megabytes rather than pretending.
What kind of photo does a background remover work best on?
One clear subject in front of a background it contrasts with. A person, a product or a pet shot against a wall, a sky or a plain surface cuts cleanly. What struggles: a busy scene with no obvious subject, a subject the same brightness as whatever is behind it, and heavy motion blur. If nothing stands out at all the tool tells you so instead of handing back an empty PNG, which is the failure most cutout tools hide.
Does it handle hair and fur?
Reasonably. The mask is enlarged from the model's own output before its edge is hardened, and that order is what keeps a three to four pixel feather just inside the cut, so hair and fur end up soft rather than scissored. What it will not do is pick out every individual strand against a busy background. For that you still want a manual mask in a full editor — but for a thumbnail subject at 1280 pixels, this is the edge you want.
What to do with the cutout
A transparent subject is the first move in almost every thumbnail: subject on one layer, background on another, text on top. Once you have the PNG, check how the finished frame holds up at real sizes with the thumbnail tester, keep the subject clear of YouTube's own furniture with the safe zone checker, and pull colours for the background out of the photo itself with the colour palette extractor. If you are rebuilding someone else's layout, the walkthrough is in how to recreate a thumbnail style.