Text layers are used directly
Every page is checked for embedded text first. Anything exported from Word, InDesign or LaTeX converts in seconds with perfect accuracy — no recognition guesswork at all.
Books that reflow to any screen
Most PDFs already carry their text, and those convert in seconds. Only genuinely scanned pages go through OCR — in parallel, across every core your machine has.
Skips OCR where it can
Parallel recognition
Tamil & English
Nothing is uploaded
1 Open a PDF
Read on your device. Nothing is uploaded anywhere.
Drop a book PDF here, or click to browse
Up to 200 MB · encrypted files will ask for the password
2 Settings
These only affect pages that actually need OCR.
3 Convert
Leave the tab open — the work happens here, not on a server.
0From text
0OCR'd
0:00Elapsed
—Remaining
Text layer OCR Waiting
Why it's quicker
The fastest OCR is the one you don't run
A converter that reads every page as a picture spends minutes doing work the PDF had already done for it.
OCR runs in parallel
Scanned pages are shared across several recognition workers at once instead of queuing behind a single one, which is roughly a four-fold difference on a typical laptop.
Rendering overlaps recognition
The next page is drawn while the current one is being read, so the processor is never sitting idle waiting for the other half of the job.
A real EPUB, not a zip with a new name
Proper table of contents, unique identifier, split chapters and an uncompressed mimetype entry — so Calibre, Apple Books and Kindle Previewer all open it without complaint.
Seven language pairs
Tamil, Hindi, Malayalam, Telugu and Kannada alongside English. Picking English only when that is all you have roughly halves the recognition time.
Stop without losing the work
Change your mind on page 300 and the pages already read are still assembled into a usable EPUB, rather than the whole run being thrown away.
Questions
Frequently asked
How long does a book take?
If the PDF already contains text — most do — a three hundred page book takes a few seconds. A genuinely scanned book needs OCR on every page, which is roughly one to three seconds per page spread across your processor cores, so expect a couple of minutes for two hundred pages.
How accurate is the Tamil OCR?
Good enough to read and search, not good enough to publish unproofed. Tamil is a complex script and recognition accuracy on a clean scan is typically in the nineties, which still means several errors per page. Straight, high-contrast scans do much better than photographs of pages. Always read through the result before sharing it.
Are my books uploaded anywhere?
No. The PDF is read and the EPUB assembled inside your browser tab. The only network traffic is downloading the recognition language data the first time you use it, which is then cached for later runs.
Why does the first run take longer to start?
The recognition engine downloads a language model, which for Indian scripts can be several megabytes. It is stored in your browser afterwards, so the second conversion in the same browser starts almost immediately.
Will the EPUB keep the original layout?
No, and that is the point of the format. EPUB reflows text to fit whatever screen it is on, so fixed columns, page numbers and image placement do not carry over. If you need the layout preserved exactly, keep the PDF.
Can I convert only part of a book?
Yes. Put a range like 1-40 in the pages box to test the settings on a chapter before committing to the whole thing — a sensible habit with a scanned book, since it tells you within a minute whether the OCR quality is worth the wait.