OCR PDF
Processed in your browser. The file never leaves your device.
Frequent questions
Because each language model is 6 to 10 MB and has to be fetched once. Your browser caches it after that: the second run starts cold and works offline. Tick two languages and both are downloaded.
No. The OCR engine runs inside your browser, compiled to WebAssembly. The only thing that crosses the network is the language model, which comes from the model server and carries nothing of yours.
The searchable PDF keeps the document looking exactly as it does and adds a text layer underneath that is never drawn, so Ctrl+F finds matches over the image. The .txt throws the layout away and keeps only the words, in page order.
Usually, and usually from the original: a straight scan at 300 DPI with good contrast beats a skewed photograph by a mile. Picking the right language matters too — and ticking five languages "just in case" makes it worse, not better.
Yes, and you should: every page takes several seconds. Type the numbers you care about and the rest are neither drawn nor recognised.
It works, but you do not need it. If the PDF already carries embedded text, the extract-text tool gives it to you instantly and with no misreadings. OCR is for scans and photographs.