Every tool asks the same question first: can this run on the machine the user is already holding?
If it can, it runs there. That makes the tool free to use, keeps the user’s files on their own disk, and — the part that matters most to me — means it costs nothing to serve. A tool that costs nothing to serve does not switch off in a month when there is no money to pay for idle hardware.
Most AI products upload your file to someone else’s GPU and bill by the second. That is two problems wearing one coat: your files leave your machine, and the product only exists while somebody keeps paying. So the free tiers here use models small enough to ship to the browser — a 5 MB upscaler, a 4 MB cutout model, Whisper, my own text detector. Only the heavy jobs reach a GPU, and only when you ask for one.
I work end to end: the model choice, the queueing and the deployment as much as the interface on top. Available for freelance and contract work.