Image to Text (OCR)
Drop images, or browse
PNG, JPG, HEIC, WebP, GIF, BMP, TIFF — several images become a ZIP of text files
Drop a photo, screenshot or scan onto the page — PNG, JPG, WebP, GIF, BMP and TIFF all work.
The first run downloads the recognition engine (about 37 MB); after that it is cached and starts instantly, even offline.
Text appears in the box below, laid out in reading order — left to right, top to bottom.
Copy it in one click, or download it as a .txt file. Several images at once download together as a ZIP of text files.
Every other free OCR site works the same way: you upload your image to their server, their server reads it, and you hope they delete it afterwards. That is a bad trade for the things people actually want to OCR — a payslip, a passport page, a contract, a screenshot of a private conversation. This page does the reading on your own machine instead. The recognition model is downloaded to your browser once and then runs there, so the picture never leaves your device.
It uses PP-OCRv6, an open-source recognition model from the PaddleOCR project, running through your browser’s own machine-learning acceleration. Detection finds each block of text in the picture — including text that is rotated or photographed at an angle — and recognition reads it out, in any of 50 supported languages, from English and the Latin-script European languages through to Simplified Chinese, Traditional Chinese and Japanese.
100% private“The images people run through OCR are rarely trivial — invoices, IDs, medical letters, screenshots of private chats. Nothing here is uploaded: the model runs inside your browser tab, so the only copy of your image stays on your own disk. Once the engine is cached, the tool works with your network switched off entirely, which is the simplest proof that nothing is being sent anywhere.”
The recognition engine is about 37 MB and has to be downloaded before anything can be read — that is the price of doing the work on your device instead of a server. Your browser caches it, so every visit after the first starts immediately and works offline.
50, sharing a single model: English and 46 Latin-script languages (French, German, Spanish, Italian, Portuguese, Dutch, Polish, Czech, Hungarian, Romanian, Turkish and more), plus Simplified Chinese, Traditional Chinese and Japanese. You do not pick a language — the same model reads them all, and it handles mixed-language pages. Korean, Russian and the other Cyrillic languages, Arabic, Hebrew, Thai, Vietnamese and the Indic scripts are not covered by this model.
Not reliably. The model is trained on printed and typeset text — documents, signs, screenshots, receipts, book pages. Neat block capitals sometimes come through, but cursive handwriting generally will not.
Mostly resolution and focus. Text needs to be a reasonable size in the picture — if a line is only a few pixels tall, there is nothing to read. Heavy motion blur, glare, very low contrast and extreme angles also hurt. Photographing the page straight-on in even light fixes most problems.
Not directly — this tool takes images. Our PDF to Text tool extracts the text layer from PDFs that have one; for a scanned PDF, export the pages as images first and drop them here.
It preserves reading order — lines come out top to bottom, and words within a line left to right — but not visual layout. Columns, tables and exact spacing are flattened into plain text lines.