Image to Text OCR
Pull the words out of a photo, screenshot or scan and get back editable, copyable text.
How the recognition works
This is OCR — optical character recognition. The tool studies the shapes of the letters in your image and reconstructs the characters, using a model trained on huge amounts of printed text. It's genuinely excellent on clean, typed material and far less sure of itself on anything messy. All of it runs in your browser; the first time, it downloads the language model, then caches it so later scans are quick.
The single biggest factor in accuracy is the picture you feed it. High resolution, good contrast, a straight (not skewed) page, a tight crop, and the correct language selected will all do more for the result than any setting.
Where it struggles
Be realistic about the limits so you're not caught out: handwriting — especially cursive — is unreliable, decorative and script fonts trip it up, and blurry or low-resolution images lose the detail it needs. Text laid over a busy photo is hit and miss. For clean printed text, though, it's a huge time-saver over retyping.
How to extract text from an image
- Add the picture containing the text — a screenshot, a scan or a photo of a page.
- Let it recognise the characters.
- Copy the text out or save it.
What makes OCR succeed or fail
Optical character recognition reads the shapes of letters out of a picture and turns them back into editable text. Its accuracy depends almost entirely on the input: sharp, level, well-lit images of printed text work extremely well, while blurry photos taken at an angle, low-contrast text and handwriting are much harder.
If a result comes back garbled, retake the picture square-on with more light rather than trying to fix the text afterwards — a better source image improves accuracy far more than any correction can.
Common questions
Does OCR work on handwriting?
Printed text is far more reliable. Neat handwriting sometimes works, but expect mistakes.
Why is the extracted text wrong in places?
Usually blur, a slanted angle, low contrast or an unusual font. A sharper, straighter, brighter photo is the fix.
Is my image uploaded to read the text?
No. The recognition runs in your browser, so private documents stay on your device.
Can it read handwriting?
Not reliably. It's built for printed text; neat block capitals sometimes work, but cursive is beyond it.
Why is the first scan slower than the rest?
The first run downloads the language model for the language you picked. After that it's cached in your browser and later scans are much faster.
Is my image uploaded to run the OCR?
No. Once the model is cached, recognition happens on your device and the image never leaves it.