Image to Text (OCR)
Text recognition runs in your browser with the Tesseract engine and reports a confidence percentage. Sharp, straight, printed text works best. Only the language data (1.5 to 3 MB) is downloaded, and your image is never uploaded.
How to use
- Drop one or more images onto the box, or press it to choose files (JPG, PNG, WebP and similar, up to 50 MB each). The images are listed with their size.
- Choose the language of the text. It starts on the language of this page. If the text also contains English words or numbers, keep Also recognize English ticked.
- Press Extract text. The first time, the engine and the language data are downloaded, which takes a few seconds. The text appears in the box, where you can edit it, copy it or save it as .txt.
How it works
Text is read with Tesseract, a text recognition engine that runs inside your browser as WebAssembly. It finds the lines of text in the picture, then recognizes each character with a small neural network, and it reports how confident it is, as a percentage.
Accuracy depends on the picture. Sharp, straight, well lit text with dark letters on a plain background gives the best result. Blurry photos, very small text, handwriting and busy backgrounds give a lower confidence and more mistakes, so check the result.
The engine files come from this site. The language data, about 1.5 to 3 MB per language, is downloaded once from a public file network and then kept by your browser. Your images are never sent to a server.
FAQ
Is my image uploaded?
No. The image is read and recognized in this browser tab. The only download is the language data. There is no upload, no account and no copy kept anywhere else.
Why is the first run slow?
The first time you use a language, the browser downloads the engine and the language data. After that they are kept by your browser, so the next runs start much faster.
Why are there mistakes in the text?
Recognition is never perfect. Letters that look alike, such as O and 0 or l and 1, are easy to mix up, especially in small or blurry text. Check important numbers and names against the image.
Does it read handwriting?
Not reliably. The engine is built for printed text. Neat block letters sometimes work, but cursive handwriting usually gives poor results.
Which languages are supported?
English, Korean, Japanese, Chinese (simplified and traditional), Spanish, Portuguese, French, German, Italian, Russian, Turkish, Arabic, Hindi, Vietnamese and Indonesian. Pick the language that matches the text in the picture.