Recognize printed Arabic in photos, scans and PDFs and paste it right-to-left into Word, Google Docs or a chat, without retyping. Nothing is uploaded, and there’s no account or watermark.
Drop your images or PDFs onto the drop zone, or tap “or tap to choose files”.
Arabic is already checked under “Text language”. Add French or English only if the document mixes scripts, for example an Arabic and French form.
For a PDF, type a range in “PDF pages” if you only need some pages, then press “Extract text”.
Open “Show text”, check numbers and dates, then “Copy text” or “Download .txt” (UTF-8, so Arabic displays correctly).
Arabic is harder to recognize than Latin script: letters join, change shape according to their position in the word, and many differ only by dots. The Tesseract Arabic model handles clear printed text well. Typed letters, official documents, books and newspapers set in Naskh-style fonts give the best results; in our tests on clean printed text about 96% of characters were right, with whole words almost always correct.
Some styles are beyond it. Handwriting gives poor results, as do calligraphic styles such as Nastaliq, Diwani and Thuluth, and decorative headlines on posters. Full vowel marks (tashkeel or harakat) are often dropped or misread, so fully vocalized religious or educational texts come out mostly without them. Arabic-Indic digits (٠١٢٣) are the weakest point: always check numbers, dates and amounts.
The text is stored in logical reading order, the way it was written, not the visual order of the picture. That is why it pastes correctly right-to-left into Word, Google Docs, WhatsApp or an email. Tesseract adds invisible direction marks around numbers; filetidy removes them so that searching and sorting the pasted text behave as expected.
iPhone & iPad
Arabic support in Live Text depends on your iOS version. If the button doesn't appear on an Arabic page, your version doesn't read it yet.
Android
Windows
If the direction buttons are missing, add Arabic as an editing language in Word's language options.
For a page written entirely in Arabic, keep Arabic alone. Adding a Latin language makes the engine hesitate between alphabets, and it may read Arabic digits or short words as Latin letters, which lowers accuracy.
Add a second language when the document really mixes scripts. Administrative papers in Morocco, Algeria and Tunisia often put Arabic and French side by side: choose Arabic and French. For a technical text with English product names or email addresses inside Arabic sentences, choose Arabic and English. The process is slower with two models, but Latin words come out properly instead of as nonsense.
Compared with retyping on an Arabic keyboard, OCR saves the most time on long passages; you only correct a few letters and numbers. Phone tools like Live Text and Google Lens handle one picture at a time, while this page takes a whole batch of photos or a multi-page PDF and saves the text to files.
The engine often drops harakat or confuses them with dots and noise, especially in small print. For a fully vocalized text, expect to add the tashkeel back by hand, or work with the unvocalized text if that is enough.
Arabic-Indic digits are the hardest part for the model, and they are easy to misread when a second Latin language is enabled. Run the page again with Arabic alone and compare every amount, date and reference number with the original.
Not dependably. The model is trained on printed fonts, so handwritten notes and most calligraphy produce a Low confidence rating and many wrong letters. Typed or printed documents are where it works well.
The characters are in the right order; the paragraph direction of the destination is wrong. In Word or Google Docs, set the paragraph to right-to-left. Plain-text editors without RTL support may display mixed Arabic and Latin text oddly.
Yes. Drop the PDF here; scanned pages are recognized with the Arabic model and pages that already contain text are copied directly. Each page gets its own heading in the result.
It works well on typed certificates, contracts and letters, and nothing is uploaded, so the documents stay on your device. Still proofread names, ID numbers and dates before using the text anywhere official.
Last updated October 28, 2026
Enlarges small text, turns dark backgrounds light and evens out shadows. Try it off if letters come out broken.
Up to 50 scanned pages are recognized per PDF.
Pages with real text are copied exactly and instantly, without OCR. Tick this if that text comes out garbled.