OCR PDF
Make a scanned PDF searchable: recognise the text and add it as an invisible layer. Runs on your device; the first run downloads the language pack. It is completely free, needs no sign-up and adds no watermark — and it runs in your browser, so your files are processed on your device and never uploaded.
Scanned PDF → searchable PDFHow to use OCR PDF
- Add your scanned PDF — or skip the PDF entirely and pick a JPG or PNG photo in the “read a photo directly” field.
- Choose the language printed on the page: English is pre-selected, and Lithuanian, Hindi, Russian, German, Spanish and French are also available. This choice matters — it tells the reader which alphabet and dictionary to expect.
- If the page was scanned at an angle, tick the deskew option so a straightened copy is read.
- Tap Run OCR. On the very first run the language pack (a few MB) downloads, so stay online for that one run; later runs start immediately. Up to 10 pages are read per run.
- Read the recognised text in the panel, then download it as a .txt file, or download the searchable PDF — your original pages with the recognised text added as an invisible layer.
Good to know
- Price: free. There is no paid plan, no credits and no trial.
- Sign-up: none. The tool opens and works straight away.
- Watermark: none is added to anything you download.
- Privacy: Your files are processed on your own device, in your browser, and are never uploaded.
- Part of: the Convert tools in PdfLuxe — 151 free tools in one place.
Questions people ask
Is OCR PDF really free?
Yes. Every tool in PdfLuxe is free with no premium tier, and the default on-device OCR engine is free and unlimited: no sign-up, no credits that run out, and no watermark on what you download.
Are my files uploaded anywhere?
Not with the default on-device engine — recognition happens in your browser and your pages never leave your device. The tool also offers optional cloud engines that work with your own API key; only if you choose one of those are page images sent to that provider, under your key.
Which languages can OCR read?
English, Lithuanian, Hindi, Russian, German, Spanish and French. Choose the language printed on the page before running — it tells the engine which alphabet and dictionary to expect, and it makes a real difference for letters with accents and special marks.
Why is some of the recognised text wrong?
OCR reads shapes, so anything that blurs the shapes costs accuracy: low-resolution scans, tilted or shadowed pages, faded print, or the wrong language selected. A straight, clean 300 dpi scan with the correct language set reads far better than a casual photo. Always proofread names, dates and numbers.
Can OCR read handwriting?
Not reliably. The engine is built for printed text — books, letters, forms, receipts. Handwriting, especially cursive, comes back fragmentary, and no setting here changes that.
What is the difference between the .txt download and the searchable PDF?
The .txt is just the recognised words — best for copying into a document or form. The searchable PDF keeps your original scanned pages exactly as they look and adds the recognised text as an invisible layer behind them, so the file looks unchanged but can be searched, selected and copied.
Why did the first OCR run need the internet and take longer?
The engine's language pack — a few MB — is downloaded the first time you run OCR, so that run needs a connection and a little patience. It is cached afterwards, so later runs start straight away and work offline.
How many pages can I OCR at once?
Up to 10 pages per run. For a longer document, split it into chunks with Split PDF and run OCR on each part, or simply run the tool again on the next pages.
Step-by-step guide
What OCR does to a scanned PDF
A scanned page looks like text to you, but to a computer it is only a photograph: thousands of pixels, no letters, nothing to search, select or copy. OCR — optical character recognition — is the process of reading that photograph and working out which letter each shape is. This tool does the reading with the Tesseract engine, running entirely inside your browser. It can give you two results. The plain text itself, shown in the panel and downloadable as a .txt file, is what you want for copying into a document, a form or a translation tool. The searchable PDF is subtler and often more valuable: it keeps your original page images exactly as they are and lays the recognised text invisibly behind them, aligned word by word. The document looks identical — same scan, same stains and stamps — but Find now works, text can be selected and copied, and search tools and screen readers can read it. A scan archive converted this way becomes a library you can actually search.
Getting the best read
OCR accuracy is decided mostly before the tool ever sees the page — by the quality of the scan or photo. These are the things that genuinely move the result:
- Resolution. Scan at 300 dpi if you can. Letters need pixels to be recognised; a small, soft image gives the reader shapes, not characters.
- Straight and flat. Photograph pages from directly above, with the page flat and evenly lit. The deskew option corrects a tilted scan, but it cannot undo the distortion of a curling page or a shadow across the text.
- The right language. Picking the page's language matters more than people expect, especially for letters with marks and accents — Lithuanian ą, č, ę or German ä, ö, ü are only recognised reliably when the reader is expecting them.
- Clean contrast. Black print on white paper reads best. Faded, stained, highlighted or reversed-out text (white on a dark background) will produce more errors wherever it appears.
When OCR struggles — honestly
Handwriting is the big one: this reader is built for printed text, and cursive or handwritten notes will come back fragmentary at best. Very small print, decorative display fonts, crumpled or coffee-stained pages, and print over photographs all raise the error rate. Tables come back as lines of text in reading order, not as a grid of cells — if you need the numbers as a spreadsheet, OCR the scan first and then use PDF to Excel on the searchable result. And whatever the source, proofread anything that matters: names, dates, account numbers and totals. OCR errors are honest mistakes of shape — a 1 read as an l, a 0 as an O — and they cluster exactly where accuracy counts.
Privacy, engines and limits
The default engine is the on-device one, and under it your pages never leave your computer or phone: the only thing ever downloaded is the OCR language pack itself, once. The tool also offers optional cloud engines for people who have their own API key with OCR.space, Google Cloud Vision or Azure — under those, page images are sent to that provider under your key, and the choice is yours; nothing is sent anywhere unless you pick one. Practical limits: a run reads up to 10 pages, so for a longer scan, split it into chunks with Split PDF and run each part, or process the whole book in a few passes. Heavy documents are bounded by your device's memory, and a corrupt or password-locked file needs Repair PDF or Unlock PDF first. No copy of your pages or the recognised text is kept — close the tab and they are gone.
Related tools
Questions about this tool? Email support@pdfluxe.com.