Why do I have to choose the language myself?

A single matching language pack reads more accurately than a mix, because the engine does not have to decide between two alphabets and dictionaries on every word. The list starts on the site's language, but you should set it to the document's.

Can I OCR photos taken with my phone?

Yes. Upload the JPG or PNG files together and they become one document in upload order. Photograph the page flat and well lit for the best result.

Why did the tool return my PDF unchanged?

Every page you selected already had its own selectable text, so there was nothing to recognize. That happens with PDFs exported from Word or a browser.

Can I upload a PDF and images together?

No. Choose either one PDF or one or more images in a single run. To combine both, turn the images into a PDF with JPG to PDF and merge the files first.

Is my scanned document sent to an AI service?

No. OCR runs on vupdf's own servers, files are used only for your task and deleted after the result is delivered, and nothing is used to train AI models.

All Tools

OCR - Searchable PDF

Turn scanned pages into a searchable, copyable PDF that keeps the original look.

Choose files

or drop it here

Upload a PDF, or one or more images (JPG, PNG) - images are combined into a single document in upload order.

Up to 50 MB per file

Completely free
No sign-up
Files are deleted right after processing

What OCR - Convert to Text does

OCR (optical character recognition) reads the letters in a scanned page or a photo and turns them into real text. Use it when a PDF from a scanner or a phone app will not let you search for a word, select a sentence or copy a number, because each page is really just a picture. Offices digitizing paper archives, students working from photographed notes and anyone filling in data from an old contract will recognize the problem.

vupdf gives you two kinds of result. A searchable PDF keeps every page looking exactly as scanned and adds an invisible text layer beneath it, so search, selection and copying work. Plain text gives you just the words in a .txt file, which you can correct right on the result screen before copying or downloading. Recognition runs on vupdf's own servers using an open-source OCR engine, not an outside AI service.

How to use OCR - Convert to Text

Open OCR and upload one PDF, or one or more JPG or PNG images, which are combined into one document in the order you add them.
Pick the Document Language that matches the text; the list covers 40 languages.
Under Part to read, keep All pages, or choose Select areas by hand and draw rectangles over the parts you want read.
For a PDF, optionally switch to Only these pages and type a range such as 1-3, 5.
Choose the Output: Searchable PDF or Plain text (.txt).
Start the run with the OCR button and download the PDF, or edit, copy and download the text.

Features and options

Searchable PDF that looks unchanged: The original page images stay untouched and the recognized words are laid invisibly on top of the matching spots. You can then press Ctrl+F, highlight lines or copy passages, and the file still looks like the scan.
One language per document, 40 to choose from: Telling the engine the exact language, such as Turkish, German, Greek or Japanese, gives noticeably better results than guessing. Chinese, Japanese and Korean are read together with English, since such documents often contain Latin words and numbers.
Hand-drawn reading areas: In manual mode you draw boxes over only the parts that matter, such as an address block or a total, and apply one box to every page if the layout repeats. In a searchable PDF, an area can also be given a web link that becomes clickable.
Crooked and sideways scans corrected: A page fed into the scanner at a slight angle is straightened before reading, and a page that is sideways or upside down is turned upright. The text layer is then mapped back onto the page as it actually appears.
Pages that already have text are skipped: If a page already contains selectable text, it is left as it is. When every chosen page already has text, the tool says so and returns the file unchanged.

When it helps

A property manager scans a box of signed tenancy agreements and needs to find every contract that mentions a particular street name.
A student photographs pages of a library book that cannot be borrowed and wants searchable notes to quote from.
An accounts assistant only needs the invoice number and total from each scanned invoice, so they draw two areas and export plain text.

Tips and limits

Each upload can be up to 50 MB, and up to 300 pages are recognized in one run. For longer scans, use Only these pages and process the document in parts.
Clear, evenly lit scans read best. Handwriting, very small print and heavily stylized fonts are recognized poorly or not at all.
OCR is never perfect: check names, figures and dates in the result before relying on them, especially in plain text output.
To edit the recognized document in Word or Excel afterwards, feed the searchable PDF into PDF to Word or PDF to Excel.

Frequently asked questions

Why do I have to choose the language myself?: A single matching language pack reads more accurately than a mix, because the engine does not have to decide between two alphabets and dictionaries on every word. The list starts on the site's language, but you should set it to the document's.
Can I OCR photos taken with my phone?: Yes. Upload the JPG or PNG files together and they become one document in upload order. Photograph the page flat and well lit for the best result.
Why did the tool return my PDF unchanged?: Every page you selected already had its own selectable text, so there was nothing to recognize. That happens with PDFs exported from Word or a browser.
Can I upload a PDF and images together?: No. Choose either one PDF or one or more images in a single run. To combine both, turn the images into a PDF with JPG to PDF and merge the files first.
Is my scanned document sent to an AI service?: No. OCR runs on vupdf's own servers, files are used only for your task and deleted after the result is delivered, and nothing is used to train AI models.