Scan to text (OCR)
Read the words out of a scan or photo
Drop files here
or — you can also paste with ⌘V
One scanned PDF, or photos and images · up to 100 MB each
This tool runs entirely inside your browser. Your files are never uploaded, never stored, and never seen by anyone but you.
How to get text out of a scan
- Drop in a scanned PDF, or photos of pages.
- Choose the language the text is written in — this matters a lot for accuracy.
- For a long document, try a page or two first. Recognition takes a second or two per page.
- Copy the text, or download it as a .txt file.
Does this send my document anywhere?
No. The recognition engine is downloaded to your browser and runs there, the same as every other tool here. What is downloaded is the engine and the language model — several megabytes, once, kept in your browser's cache so later visits start immediately. Your pages are never uploaded and are not part of what gets stored.
Why is there a download at all?
Recognising text is genuinely hard, and the program that does it is large. Every site that offers OCR without a download is running it on a server — which means your document is uploaded to one. The download is what buys the privacy.
When you should not use this
If your PDF was made by a word processor or a printer driver, its text is already inside the file. PDF to text pulls that out instantly and exactly — no download, no recognition errors. This tool is for pages that are pictures: scans, photographs, faxes. It tells you if you have brought the wrong kind.
Getting a better result
Accuracy depends almost entirely on the image, not on settings. Scan at 300 DPI or higher, keep the page square to the glass, use good even light for photographs, and pick the right language. Small skew is handled; heavy skew, shadows, creases and handwriting are not. Handwriting in particular will not work — this reads print.
Which languages?
English, Spanish, German and French. Each has its own model, downloaded only if you choose it. A page in a language you have not selected will come out as nonsense rather than as an error, so it is worth checking before a long run.
Limits
Up to 30 files, 100 MB each and 200 MB in total. There is no page limit, but recognition is the slowest thing on this site — around a second or two per page, and it runs on your own machine, so a hundred-page scan is a coffee break rather than a click.