Type the text, choose where it sits and how strong it is, and it is drawn into the picture itself. This is a genuinely different thing from a PDF watermark: there is no separate layer to delete, because the pixels underneath have been replaced and cannot be recovered from the saved file. What it still cannot do is stop anyone using the picture — a corner mark can be cropped away, so the tiled option exists for when that matters.
This page loads no libraries at all — the work is done by the browser itself. Your file is read on your own machine and never sent anywhere — open your browser's network tab and watch while you use it.
Not by peeling it off. In a PDF the stamp is a separate object that any editor can delete, and the text underneath stays fully readable; here the mark is drawn into the pixel grid, so what was under it is gone from the saved file and cannot be recovered. That is the honest limit of the comparison, though — see the next question.
No, and it is worth being blunt about that. A corner mark can be cropped off in seconds, and modern tools can paint over a mark convincingly. A watermark makes casual reuse awkward and makes the source obvious when the picture travels; it does not prevent anything. Treat it as a signature, not a lock.
It repeats the text diagonally across the whole picture instead of placing it once. Cropping then means cropping away the picture as well, which is the point. It is uglier and much harder to defeat, so it suits proofs and previews rather than finished work.
It is re-encoded, so a JPEG loses a little as it always does when rewritten; save as PNG to avoid that. The metadata is dropped in the process too, which for a photograph you are about to publish is usually a bonus — the GPS position goes with it.
Everything here is free and works the same way — in your browser.