Image to text
Copy the text out of a screenshot, a photo of a document or a scanned PDF. Recognition runs in your browser, in eight languages, with nothing uploaded.
The first run downloads the recognition engine and the language, about 5MB, which your browser then keeps. Choosing the right language matters: it is how accents and characters are recognised.
Screenshots, photos of documents and scans. Or paste a screenshot with Ctrl+V.
Text is recognised on this device. Your images aren’t uploaded.
What it is good at
Screenshots are the ideal case: sharp, high-contrast text with nothing in the way. Pull the text out of an error message, a slide, a photo of a menu, a page of a book, a receipt or a document someone sent as a picture. Scanned letters and forms work well too, as long as the scan is straight and at a reasonable resolution.
Small images are enlarged before recognition, because the engine reads best when letters are around 20 to 30 pixels tall and a cropped screenshot of small print is often less than half that. You don’t need to do anything; it happens automatically.
Getting better results from a photo
Photograph the page straight on rather than at an angle, in even light without the shadow of your phone across it and fill the frame with the text. If the text is only part of the picture, crop to it first, since everything else in the frame is something the engine has to decide isn’t text. A slightly skewed photo is usually fine; one taken at a steep angle isn’t.
Languages
English, Spanish, French, German, Portuguese, Italian, Simplified Chinese and Japanese are available. Choose the language the text is written in: the engine uses it to recognise accented letters and whole words and reading French with the English model loses the accents. Each language downloads once, when first chosen.
How it works
The recognition engine reads whole lines at a time rather than letter by letter, which is what lets it cope with ordinary fonts, screenshots and phone photos of pages. It runs inside your browser and is served from this site rather than a third party, so the image never leaves your device.
Common questions
Is my image uploaded to read the text?
No. The recognition engine is downloaded to your browser and runs there, so the image never leaves your device. That is unusual for OCR tools, most of which send your picture to a server and it matters for the invoices, IDs, letters and screenshots of private messages that people most often need text from.
How accurate is it?
On a clear screenshot or a clean scan of printed text, close to perfect. Accuracy drops with blur, low resolution, strong shadows, text photographed at an angle, decorative fonts and busy backgrounds. The confidence figure under each image is the engine’s own estimate; below about 60 percent, check the result carefully.
Can it read handwriting?
Only very neat block capitals and not reliably. The engine is trained on printed text. Handwriting recognition needs much larger models than can sensibly run in a browser tab.
Can it read a scanned PDF?
Yes. Drop the PDF and each page is rendered at about 200 DPI and read in turn, with the text marked by page. If a PDF already contains real text, PDF to text is faster and exact, since there is nothing to recognise.
Why is the first run slower?
It downloads the recognition engine and the language data, about 5MB for English. Your browser keeps them, so later runs start straight away, even across visits.
Guides
The other tools
All of them work the same way: the file is read in your browser and never uploaded.