Are my files uploaded to a server?

No. This conversion runs inside your browser using your own device’s processing power, and the file is not uploaded. Once a tool has been used once (video, text recognition and background removal download their engine from this site the first time), it keeps working when your connection drops.

How do I know if my PDF needs OCR?

Try selecting text in it. If nothing highlights, it is a scan and needs OCR; if text selects normally, the plain PDF-to-Word converter will do a better job.

Does the layout survive?

You get the text in reading order, not a reconstruction of the page design. Columns and tables come through as text rather than as formatted structures.

PDF OCR: extract text from a scanned PDF

Runs on your device · Your file never leaves your device.

A scanned PDF is a stack of pictures — that is why selecting or searching its text does nothing, and why converting it to Word comes back empty. This reads the characters out of each page image and gives you real text.

Every page is rendered at a high resolution first and then recognised, with progress shown page by page. Long documents take a while: recognition is genuinely doing work on your device rather than queueing on a server.

Arabic, English and mixed Arabic/English documents are all supported, and Arabic output keeps its right-to-left direction in both the preview and the Word export.

How to make a PDF OCR

Add a scanned PDF — one where selecting text does nothing.
Choose the language the text is written in; for Arabic and English on one page, choose Arabic + English.
Press Extract text. The recognition model downloads once on the first run and is cached afterwards.
Copy the text, or download it as TXT or Word.

Supported formats and limits

Input: Scanned PDFs and photographed pages
Output: Text — copy, TXT or Word
Languages: 15: English, Arabic, Arabic + English mixed, German, French, Spanish, Russian, Portuguese, Chinese (Simplified and Traditional), Hindi, Turkish, Hebrew, Persian, Urdu, Indonesian, Bangla
First run: One-time model download, then cached offline
Where it runs: In your browser; the document is never uploaded

Not for: Handwriting — recognition is reliable on printed text and much less so on handwritten notes.

How each page is prepared for reading

Every page is first drawn as a picture at 216 dots per inch, the way a PDF reader shows it, so a page that carries a rotation setting is read upright. A page whose scan itself lies on its side is read as it is and comes out as nonsense; turn those pages with /rotate-pdf first. Very large pages, around A2 and bigger, are drawn at a lower resolution to stay within what a browser can hold as one picture.

Every page is read from its picture, including pages that already contain real text. For a document that is partly typed and partly scanned, the typed pages come out exactly through /pdf-to-text, while recognition can misread a character.

Long documents

Each page is drawn, read and let go before the next one is drawn, so a long scan needs about as much memory as a single page. The bar fills a little while the recognition engine loads, then page by page, with the page being read named underneath.

Cancel stops at the page being read and keeps the text of the pages already finished: a line says how many of how many were done, and the copy, TXT and Word downloads hold those pages. Several PDFs, and photos of pages, can be added together and are read one after another in a single job.

What you get back

The result is text, not a new PDF: the scanned file stays a set of pictures. When more than one page is read, each page's text starts with a line such as --- scan.pdf — page 3 ---, written in the language of the page you are using, so you can find where a passage came from. The Word file is titled with the PDF's name, and each page sits under its own heading, one after another rather than on separate pages.

On this page the language list starts on English. Set it to the language the document is written in before you press Extract.

Frequently asked questions

Are my files uploaded to a server?: No. This conversion runs inside your browser using your own device’s processing power, and the file is not uploaded. Once a tool has been used once (video, text recognition and background removal download their engine from this site the first time), it keeps working when your connection drops.
How do I know if my PDF needs OCR?: Try selecting text in it. If nothing highlights, it is a scan and needs OCR; if text selects normally, the plain PDF-to-Word converter will do a better job.
Does the layout survive?: You get the text in reading order, not a reconstruction of the page design. Columns and tables come through as text rather than as formatted structures.