Upscale an image

No. Toolpit does not upload your file for processing — the tool runs as code in this browser tab, works on the copy your browser opened, and forgets it when you close the tab. You can check this yourself: open your browser’s network panel and run the tool. You will see the page, its scripts and the ads on the site, and — when you save, if you are signed in — one small request that charges 10 tokens to your account. None of them carries your file, its name or anything read from it, because file processing stays on your device.

The work is happening on your own hardware rather than a rack of GPUs. A small image is quick; a 4000-pixel photo at 4x is a genuinely large amount of computation. Closing other heavy tabs helps.

Not really, and anyone claiming otherwise is selling something. The model has learned what edges, textures and letters usually look like, so it reconstructs a larger image that is plausibly sharp rather than a blurry enlargement — which genuinely looks far better. But detail that was never captured cannot be brought back: a face that is twelve pixels across does not become a recognisable person, it becomes a confident guess.

Because a neural network is being run over every patch of the image, on your own device rather than on a rented server. Time scales with the number of pixels, so upscaling a small image is quick and a large one takes a while. Cropping to the part you actually need before upscaling is the most effective thing you can do; a desktop will also be considerably faster than a phone.

Good: old photographs, small logos and artwork, product shots that need to print larger, screenshots with text. Less good: pictures that are heavily compressed to begin with, since it will sharpen the compression artefacts along with everything else, and anything where accuracy matters more than appearance — it makes images that look right, not images that are evidence.