Image to Text (OCR) — Extract Text from Images

Drop an image, choose a file or paste a screenshot with Ctrl+V, pick the languages in it, and the tool reads the text with Tesseract, the open-source OCR engine, compiled to WebAssembly and running in this tab. The result appears in an editable box you can copy or save as a .txt file. The picture never leaves your device, so it is safe for receipts, contracts and screenshots of private messages.

Twelve languages are available: English, Korean, Japanese, Chinese (simplified and traditional), Spanish, Portuguese, French, German, Italian, Russian and Vietnamese. Choose every language the image contains — up to three — because Tesseract only recognises the scripts it was told to expect; Korean text read with English alone comes out as nonsense. Each language's model (1–3 MB) and the engine (about 4 MB) download the first time they are needed and are then cached by the browser.

Accuracy depends far more on the image than on the engine. Clean screenshots and scans at 300 dpi usually come back almost perfect. Small text (below about 10 pixels high), blur, strong perspective, curved pages, handwriting, stylised fonts and text over busy photos all lower it, and the confidence figure shown with the result drops with them. Crop to the text and photograph pages straight on for the best result.

The output is plain text: line breaks follow the image, but tables, columns, bold and font sizes are not reproduced, and the reading order of complex layouts can be off. Tesseract puts a space between every Chinese or Japanese character; the tool removes those, since the languages are written without spaces.