Skip to content
ImageDoctor

OCR Text Extractor

Pull the text out of screenshots, scans and photos in 60+ languages, then edit, copy or download it.

Drop an image to read its text

Screenshots, scans, photos of documents, signs and receipts. Nothing is uploaded.

Processed locally in your browser

JPG · PNG · WebP · AVIF · GIF · BMP · TIFF · SVG

What reads well

Printed text that is roughly upright, in focus and larger than about 20 pixels tall: screenshots, PDF pages exported as images, scanned letters, invoices, book pages and signs. Handwriting is not recognised.

Where the work happens

The recognition engine is WebAssembly running in your browser, so the image stays on your device. Each language downloads a model file once — after that the tool works with no network at all.

Getting an accurate reading

Recognition runs on Tesseract, compiled to WebAssembly and executed in a worker inside your browser. The image is decoded, cleaned up and handed to the engine as pixels — it is never uploaded. Each language downloads its model once, from a public CDN, and your browser caches it for every scan after that.

Before reading, the picture is turned to grey, its contrast is stretched, light-on-dark text is inverted and small images are enlarged towards the 300 DPI the engine was trained on. A page that was scanned or photographed slightly askew is straightened automatically. These steps matter more than any setting: they are the difference between a clean read and a page of near-misses.

Tell the tool what it is looking at and it does better still. 'Automatic' finds columns and blocks on a document page, 'Single block' suits a receipt, a label or a sign, and 'Scattered text' suits words spread around a picture. Choose the right language, too — an English model reading German will quietly drop every umlaut.

Every word comes back with a confidence score, and the tool reports how many words it was unsure about rather than presenting a guess as fact. The extracted text is editable, so you can correct anything it misread before you copy it or download the .txt file.

Common questions

Drop the image in and the tool reads it straight away. Pick the language and layout on the right if the first pass needs help, then copy the text or download it as a .txt file.

No. The recognition engine runs in your browser. The only thing downloaded is the language model, and only the first time you use a given language — after that the tool works with no network at all.

No. The models are trained on printed text. Neat block capitals sometimes come through, but handwriting is not what this engine does.

Almost always because the source is too small, blurry or skewed, or because the language is wrong. A larger or sharper scan fixes more than any setting. Check the rotation if the page is sideways, and try 'Single block' or 'Scattered text' if the layout is unusual.

Yes, up to three. Each extra language costs speed and makes confusable characters more likely to be picked from the wrong alphabet, so only add the ones actually on the page.