Free Arabic OCR, in your browser

Arabic is a connected script: letters join to their neighbours and change shape depending on whether they sit at the start, middle or end of a word — printed Arabic is, in a sense, always cursive. A recognition model trained on separated Latin letterforms has nothing to hold on to, which is why generic OCR turns Arabic into noise. This page runs the Tesseract engine with its dedicated Arabic model:

• Connected letterforms are recognised as such — the model was trained on the joined shapes, not isolated letters.
• Right-to-left comes out right. The text is returned in normal logical order, so it pastes into a chat, editor or Markdown file and reads correctly.
• Diacritics are read when printed. In the example below, صباحًا came back with its tanwīn intact. Most everyday text omits the marks anyway, and that reads fine too.
• Everything stays on your device. The model is downloaded to your browser on first use (a few extra seconds); your image never leaves your machine — this converter's rule for every format.

Printed text is the sweet spot. Screenshots, PDFs exported as images, book scans and photographed documents read well when the shot is sharp and straight-on.

Handwriting and calligraphy are out of scope. Handwritten Arabic and decorative calligraphic styles vary far more than any print model covers — expect fragments at best, and proofread anything that matters.

Mixed Arabic-and-Latin text mostly works. Embedded Latin names, numbers and dates come through, though the reading order around the direction change can need a quick check.

Quality in, quality out. Low-contrast photos, small type and busy backgrounds cost accuracy in any language. The built-in preprocessing (grayscale, contrast boost, upscaling) helps, but a clean capture beats any cleanup.