PDF OCR — read a scanned PDF
Drop scanned PDFs, or browse
Every page is read on your device — several PDFs become a ZIP of text files
Drop a scanned PDF onto the page, or click to browse — several at once is fine.
The first run downloads the recognition engine (about 37 MB); after that it is cached and starts instantly, even offline.
Pages are read one at a time, with progress shown — expect a few seconds per page.
Copy the text in one click, or download it as .txt. Several PDFs arrive together as a ZIP of text files.
A scanned PDF is a stack of photographs wearing a PDF’s clothes. There is no text inside it — try to select a word and you get a rectangle of nothing — which is why ordinary “PDF to text” tools return an empty file for it. Reading a scan needs OCR: software that looks at the pixels and works out which letters they are. This page does that, and does it on your own machine.
That last part is the whole point. The documents people need OCR for are the ones they least want to hand to a stranger’s server — scanned contracts, bank statements, medical letters, ID pages, old family paperwork. Here the recognition model is downloaded to your browser once and runs there. Each page is rendered by pdf.js, read by PP-OCRv6 (an open-source model from the PaddleOCR project), and the text appears below in reading order. Nothing is uploaded, and there is no page limit beyond your own patience.
If you drop a PDF that already has a proper text layer, the tool notices and simply reads it directly — that is faster and exact, and there is no sense recognising text a document already knows.
100% private“Scanned documents are usually the sensitive ones — signatures, account numbers, addresses, dates of birth. Every page here is rendered and recognised inside your browser tab, so the only copy stays on your disk. Once the engine is cached the tool works with your network off entirely, which is the simplest proof that nothing is leaving your machine.”
PDF to Text extracts a text layer the document already contains — instant and exact, but it returns nothing for a scan. This tool reads the pixels with OCR, which is the only way to get text out of an image-only PDF. If you drop a PDF that does have a text layer here, it uses that instead.
Roughly two to four seconds per page on a typical laptop, plus the one-off engine download. A 20-page scan is about a minute. Machines with WebGPU support are meaningfully faster, and the page stays responsive throughout because the work runs in a background thread.
50 share a single model: English and 46 Latin-script languages, plus Simplified Chinese, Traditional Chinese and Japanese. There is nothing to select — the same model reads them all, including mixed-language pages. Korean, Cyrillic, Arabic, Hebrew, Thai, Vietnamese and the Indic scripts are not covered.
OCR quality follows scan quality. Text needs enough pixels to be legible, so 300 dpi scans do far better than 72 dpi ones. Heavy skew, speckled backgrounds, faint photocopies, unusual display fonts and handwriting are the usual culprits. Rescanning straight and brighter fixes most problems.
Not directly — remove the password first with our Unlock PDF tool, then run the result through here.
Not yet. This tool produces plain text you can copy or download as .txt; it does not write a new PDF with an invisible text layer behind the scan.