Image to Text (OCR)
Get the text out of a screenshot, a scan or a photo. It is recognised on your own machine in 16 languages, and the image never leaves it.
The image stays on your device. Recognition runs in your browser, and nothing is uploaded.
Choose an image and press Read the image. The text lands here.
The 16 languages it reads
One at a time. Each language is a separate model, downloaded the first time you pick it and cached by your browser after that.
- Englisheng
- Spanishspa
- Frenchfra
- Germandeu
- Portuguesepor
- Italianita
- Dutchnld
- Polishpol
- Swedishswe
- Turkishtur
- Russianrus
- Ukrainianukr
- Japanesejpn
- Koreankor
- Chinese (Simplified)chi_sim
- Arabicara
How it works
How to use it
- Drop an image on the panel, click it to pick a file, or press CtrlV to paste a screenshot straight from the clipboard.
- Pick the language the text is written in, then press Read the image. The first run downloads the recogniser and the language, which takes a few seconds; later runs do not.
- Fix anything it misread, then copy the text or save it as a .txt file.
Getting a clean read
Recognition works on letter shapes, so the sharper and larger those shapes are, the better it does. Four things help more than anything else:
- Use the original. A screenshot at full size beats a photo of a screen, and a photo beats a photo of a photo.
- Crop to the text. Menus, icons and window furniture around the words are read as characters too, and they are what fills the output with stray symbols.
- Keep it straight. Text tilted more than a few degrees loses accuracy quickly.
- Pick the right language. Reading Spanish with the English model costs accents and whole words.
The confidence figure next to the result is the recogniser's own average across every word it read. Anything under 70 is worth a second look before you trust the numbers and the names in it.
Where the work happens
Everything runs on your machine. The page loads the recogniser only when you press the button, hands it the file directly, and runs it in a background worker so the page stays responsive. Your image is never sent to a server, ours or anyone else's, which is also why this works on a screenshot of something you would not paste into a website.
Questions
Does my image get uploaded?
No. The recognition runs inside your browser, in a background worker, and the file is read straight off your disk. The only thing fetched from the network is the recognition model itself, which is a public file and carries nothing of yours. Nothing is stored and nothing is logged.
Why is some of the text wrong?
Recognition reads letter shapes, so small, blurry, low contrast, angled or heavily styled text is where it slips. Use the largest copy of the image you have, crop to the text you want, and pick the right language. The text box is editable, so a stray character is a quick fix before you copy.
Which languages can it read?
16: English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Swedish, Turkish, Russian, Ukrainian, Japanese, Korean, Chinese (Simplified), Arabic. Pick one before you run it, since the recogniser reads one language at a time and the wrong choice is what turns clean text into nonsense.
Can it read handwriting or a PDF?
Handwriting, no. This reads printed and rendered text, the kind a screenshot or a scan contains. PDFs are not images, so open the page and screenshot it, or export the page as a PNG first.