Image to Text – Free OCR for Arabic and English

Turn the writing inside an image into text you can edit, copy and download, free. Arabic and English are supported, and the image is read on your own device inside the browser.

Choose an image containing text

JPG, PNG and WebP – up to 10 MB and 25 million pixels

No image selected

You need a connection so the OCR engine and its language data can load. The image is never uploaded to the site’s server. Reading can take a moment, especially the first time.

Reading text out of an image

Optical character recognition looks at the shapes in a picture and guesses which letters they are. It is guessing, which is the single most useful thing to know about it: the output is a draft, usually a very good one, and always worth reading before you trust it. The engine and its language data run in your browser, so the picture stays on your device, and the first run is slower because those files have to download once.

How to use it

  1. Choose a photo or screenshot containing printed text.
  2. Pick the language. This matters more than anything else you can change.
  3. Press extract, and wait: a dense page can take a while.
  4. Correct the result in the box, then copy it or download it as a TXT file.

Choosing the language properly

The engine compares shapes against the alphabet you select, so telling it the wrong language guarantees nonsense. Choose Arabic for Arabic text, English for Latin text, and both only when the page genuinely mixes the two. Loading both costs accuracy as well as time, because the engine then has twice as many candidate letters for every shape, and that extra doubt is exactly what produces mistakes.

Why Arabic is harder

Arabic letters join, change shape by position and sit under a line of diacritics, so recognition is noticeably weaker than for Latin text. Expect to fix joined words, confused dots, and stray marks where a hamza or a shadda was. Modern printed Arabic in a plain font does reasonably well; decorative fonts, Qur’anic typography and old scanned books do not. Handwriting of any language is out of reach: this engine reads print.

What a good source image looks like

Resolution matters less than sharpness and contrast. Fill the frame with the text, shoot straight on rather than at an angle, use even light, and avoid the shadow of your own hand. A slightly rotated page confuses line detection, so straighten it before extracting if you can. Crop out everything that is not text: margins, a desk, a facing page. Black text on white is the easy case; white on dark, text over a photograph, and heavy JPEG artefacts are the hard ones.

Multi-column layouts are read as if they were single columns, so a newspaper page comes out interleaved. Crop one column at a time and paste the results together.

After extraction

Line breaks follow the image, not your sentences, so a paragraph arrives chopped into lines. Numbers deserve a second look: zero and O, one and l, five and S are the classic confusions, and they matter most in exactly the places you least want an error, such as an invoice total or a reference number. The TXT download is written with a byte-order mark so Arabic opens correctly in Notepad and Excel instead of turning into question marks.

Frequently asked questions

Is my image uploaded?

No. The engine runs locally; only the engine files come from the network.

Can it read handwriting?

No. Printed text only.

Can it read a PDF?

Not directly. Convert the page to an image first, then extract.

Why is it slow?

The first run downloads the engine and language data. Later runs are faster, though a large page is always real work for your device.

Related tools

Scroll to Top