Point it at a screenshot, a photographed page or a scan and it gives you the words back as text you can paste. The reading is done by an OCR engine that runs inside this page, so the picture stays on your machine. It also tells you how confident it was, because OCR misreads quietly and a number you cannot check is worse than no number.
This one is the heavy exception on this site: the OCR engine is about 6.7 MB and it downloads the first time you use the tool, then your browser keeps it. It is served from this site, not from anyone else’s. Your file is read on your own machine and never sent anywhere — open your browser's network tab and watch while you use it.
No. The engine itself is downloaded to your browser and the reading happens there. Open your network tab while it works: the engine comes down once, and your image never goes up.
Because a real OCR engine has to arrive first — about 6.7 MB of it, which is the heavy exception on this site. It is served from here rather than from a third party, and your browser keeps it, so the second image starts straight away.
Good on clean, straight, reasonably large text; worse on small print, handwriting, low contrast or a page photographed at an angle. You are given the engine’s own confidence score with the result, and told to check the numbers when it is low. It is a reading aid, not a guarantee.
English. The language data is a separate file for each language and each one is a few megabytes, so shipping every language would make the tool heavy for everyone to serve the few who need it. If you need another language, say so and it can be added.
Everything here is free and works the same way — in your browser.