Image to text

How it works

This tool runs Tesseract, an open source text recognition engine first developed at HP and now maintained by its community, compiled to WebAssembly by tesseract.js. Your browser opens the image, turns it upright if the camera saved it sideways, and puts it on a white background. Small images, like most screenshots, are read at twice their size, because Tesseract misses fewer letters when they are bigger. Then Tesseract finds the lines of text and reads each one with a small neural network trained on printed text.

We use Tesseract’s “fast” language models for English and Italian. They are much smaller than Tesseract’s most accurate models and read faster in a browser, at the cost of a few more mistakes on hard images. On clean screenshots and scans the difference is small.

What to expect

Print is what Tesseract does well: documents, letters, receipts, screenshots of chats and web pages, book pages, slides. Columns and tables come out as text in reading order, without the layout. Handwriting, text on curved or crumpled surfaces, very small or blurry text and artistic fonts are where it struggles. If a page matters, read the result once against the original.

Privacy, limits and licenses

Your images never leave your device. You can add up to 20 files of up to 50 MB each, and they are read one at a time, so the page stays responsive and you can stop whenever you like. The engine and the language data come from this site; nothing is sent to a server. For scanned PDFs you want to keep as PDFs, OCR PDF adds an invisible text layer so the file becomes searchable. Tesseract, tesseract.js and the tessdata_fast language models are released under the Apache 2.0 license.