Turn a photo of a page into a scan

Because doing it badly is worse than not doing it. Page detection works well on a dark desk and poorly on a light one, on a document with clear edges and not on a page over a printed tablecloth - and when it is wrong it crops away part of your document and hands you a result that looks deliberate. You would not notice until someone else read it. Four handles take a few seconds, and you can see exactly what will be kept.

Because the problem is not that the page is too dark - it is that one side is much darker than the other. There is no single brightness or contrast value that is right for both ends of a gradient: raise it enough to fix the shadow and the lit side blows out. This estimates the paper's brightness at every point and divides by it, so the paper comes out even and the ink stays dark, because ink is dark *relative to the paper right around it* rather than dark in absolute terms.

Because the lighting estimate assumes that anything large and smooth is paper. A photograph pasted into the page, a solid stamp or a big block of colour is large and smooth, so it gets treated as background and divided away. Turn the flatten control down - at around 30-40% you still get most of the lighting correction and keep the picture. This is a known limitation of the approach rather than a bug.

For plain printed text, yes - it produces a far smaller file and looks closest to a real scan. For anything else, no: it destroys photographs, signatures in blue ink, highlighter, coloured forms and stamps, replacing each with pure black or pure white. It is off by default for that reason.

Not from this page - the output is an image of the page, so you can read it but not select the words. The text extractor on this site does the recognition part, and cleaning a photo here first usually improves what it can read, because the recognition works much better on a straightened page with even lighting.

No, and on a page about documents that is worth being precise about: the photo is decoded, straightened, cleaned and re-encoded by your own browser on your own device. No request carrying it is made, and a content security policy the browser enforces stops this page reaching another domain at all. Contracts, invoices and medical letters are exactly the kind of thing people scan, and none of them leave the tab.