How to get the text out of a picture or a scanned PDF
Drop the picture or PDF on the page, pick the language of the text and press Read text. Copy the text it finds, save it as a .txt file, or save a searchable PDF that keeps the original pages.
Open the file
A photo, a screenshot (paste it with Ctrl+V) or a PDF of up to 500 pages. Choose several pictures at once to read them as the pages of one document.
Pick the language
Your browser's language is picked to start with. Add a second one for mixed text. With script detection on, a page in another script, such as Cyrillic or Arabic, is read in that script's language, and sideways or upside-down pages are turned upright.
Read the text
Two readers: Tesseract for documents and scans, PaddleOCR for photos and screenshots, picked for your file and yours to switch. The first time, the reader downloads once and stays on your device: Tesseract 3.5 MB plus 0.3 to 3.8 MB per language, PaddleOCR 28 to 39 MB with its runtime. A page takes a few seconds; long PDFs are read several pages at a time.
Copy or save
Edit the text if you like, copy it, save it as a .txt file, or save the searchable PDF. Switch to the page view to see every word it found; doubtful words are marked.
Documents or photos: two readers
Tesseract reads clean pages very well: scans, printed letters, PDFs, in all 38 languages. Text in a photo, on a busy background, at an angle, or the small type of a screenshot is where PaddleOCR is much better: it finds each line of text first, straightens it, and reads it with a network trained on photos. In our test of white lettering straight on a busy picture, PaddleOCR read every character and Tesseract found nothing.
| Reader | Best for | Languages |
|---|---|---|
| Tesseract | Scans, documents, PDFs, vertical Japanese | All 38 |
| PaddleOCR | Photos, signs, screenshots, skewed text | Latin-script languages, Chinese, Japanese, Korean, Cyrillic, Arabic, Persian, Urdu |
Languages and scripts
It reads 38 languages with Tesseract's fast models. Right-to-left text comes out in reading order, and Japanese written top to bottom has its own choice, Japanese, vertical.
| Script | Languages |
|---|---|
| Latin | English, Catalan, Czech, Danish, Dutch, Finnish, French, German, Hungarian, Indonesian, Italian, Malay, Norwegian, Polish, Portuguese, Romanian, Spanish, Swedish, Turkish, Vietnamese |
| Cyrillic | Russian, Ukrainian, Bulgarian, Serbian |
| Right to left | Arabic, Persian, Urdu, Hebrew |
| Asian | Chinese (simplified, traditional), Japanese (horizontal, vertical), Korean, Thai |
| Others | Greek, Hindi, Bengali, Tamil |
What the searchable PDF is
Each page of your PDF stays exactly as it was. The text read from it goes over the page as invisible text, every word at the place it was found, so a PDF reader can search it, and you can select and copy it. Pages that already had text are kept as they are. Pictures become pages of their own, at their own quality.
- Any script in the text layer, right to left and vertical included
- Sideways and upside-down scans are turned upright
- A PDF with a password to open must be unlocked first; one with only a permissions password is rebuilt from page pictures
Where does the file go?
It stays on your device. The text reader and its language files download from this site once and are kept in your browser for next time.
Specs and browser support
| Input | JPG, PNG, WebP, GIF (first frame), BMP, AVIF, several at once, up to 100 megapixels each; PDF up to 500 pages |
|---|---|
| Output | Plain text (copy, or .txt in UTF-8, with page headings for several pages); searchable PDF |
| Engine | Tesseract 5 LSTM (tesseract.js-core, Apache-2.0) with tessdata_fast models (Apache-2.0); PaddleOCR PP-OCRv5 mobile detection, recognition and line orientation (Apache-2.0) on onnxruntime-web (MIT); pdf.js (Apache-2.0) draws PDF pages; pdf-lib (MIT) writes the PDF |
| Download | Tesseract 3.5 MB, each language 0.3 to 3.8 MB, script detection 4.3 MB; PaddleOCR 28 to 39 MB (runtime 14 or 17 MB, line finder 4.8 MB, one recogniser 8 to 17 MB); once, then kept on your device |
| Browsers | Current Chrome, Edge, Firefox and Safari on a computer; phones work for a few pages |
| Price | Free, no account, no watermark |
Questions
No. The text reader and its language files download from this site to your browser and run there. Your pictures and PDFs stay on your device.
38: most European languages, Russian, Ukrainian and other Cyrillic ones, Greek, Arabic, Persian, Urdu, Hebrew, Hindi, Bengali, Tamil, Thai, Chinese (simplified and traditional), Japanese (also written top to bottom) and Korean.
The original pages, unchanged, with the text read from them laid over each page as invisible text. You can search, select and copy it in any PDF reader, while the pages look exactly as before.
Clean scans and screenshots read almost word for word. Photos taken at an angle, handwriting, very small or blurry text read worse; words read with low confidence are marked on the page view.