OCR Text Recognition

About this tool

Quoting text trapped in a screenshot, archiving a photographed contract, digitising receipts - that is what OCR is for. This tool recognises text in images inside your browser with the WASM build of the Tesseract engine from Mozilla's ecosystem. Simplified Chinese mixed with English works out of the box, and English-only, Arabic-only, Arabic + English and Thai modes are one click away.

Results appear as editable text you can copy with one click or download as a TXT file. The first run downloads the recognition model (1-31 MB depending on the language, cached by the browser afterwards). Recognition is fully local - images are never uploaded, so ID cards, contracts and receipts are safe to process. Sharper images, larger type and straighter pages mean better accuracy; handwriting and decorative fonts remain hard.

Features

• Simplified Chinese + English mixed recognition, plus English-only, Arabic-only, Arabic + English and Thai modes
• One-click copy of results, or download as a TXT file
• Overall confidence score so you know how well it went
• Model cached after the first download - later runs start instantly
• Tesseract WASM runs locally - no upload, no signup, no limits

How to use

1. Drop or select the JPG / PNG / WebP image to recognise
2. Pick the language (Simplified Chinese + English by default, Arabic and Thai available)
3. The first run downloads the model - please wait
4. Read or copy the text, or download it as TXT

Frequently asked questions

How do I improve OCR accuracy?

Practical tips: use the original resolution rather than a compressed thumbnail; let the text dominate the frame; keep photos focused, evenly lit and square to the page; screenshots beat photos. Printed text is recognised far better than handwriting.

Why is there a model download on first use?

The OCR engine needs language model files (~31 MB for Chinese + English, about 1 MB for Arabic or Thai). They run locally inside the browser as WASM + model data, which is exactly what keeps your images private. The model is cached after the first download and later runs start immediately.

How good is Arabic recognition?

Printed Arabic is recognised very accurately - clean screenshots come back essentially error-free. Fully vowelled text, handwriting and low-resolution photos still lose accuracy. When Latin words or numbers sit inside Arabic, choose the Arabic + English mode; for pure Arabic pick Arabic only, whose model is about 1 MB and downloads in a moment.

Are my files uploaded?

No. Recognition runs inside your browser via Tesseract WASM, and the image only exists in your device memory - it works offline too.

Can it recognise scanned PDFs?

This tool currently accepts JPG / PNG / WebP images. Export the pages of a scanned PDF with the PDF to JPG tool first, then recognise them one by one; combined with the editable-text PDF to Word export, scanned documents become editable again.

Related tools