The short answer
- Calibre is excellent for PDFs that already contain text. On a scan it packs the page images into an EPUB and stops there.
- OCRmyPDF makes a scan searchable by adding an invisible text layer. The page itself stays fixed, so nothing reflows.
- OCRmyPDF, then Calibre looks like the obvious combination. In our test Calibre ignored the OCR layer with both of its PDF engines.
- k2pdfopt cuts each page into strips and re-renders them for a small screen. You get a better PDF, not an EPUB.
- upub reads the page images with an AI model and writes an EPUB 3 with paragraphs, headings and a table of contents.
How we tested
We used one public-domain scan for every tool: Alice's Adventures in Wonderland, Macmillan 1872, 214 pages from the University of Toronto copy on the Internet Archive. To make it a true scan, we built a PDF from the page images alone, with no text layer. That is what you get from a phone scanner app or a flatbed scanner.
Each EPUB was compared word by word with the proofread text of Project Gutenberg #11, 26,681 words long. We also ran every file through EPUBCheck, the validator Kobo and Apple use when they accept a book. The sample EPUB is linked at the end of this page, so you can open it and check.
| Method | Words recovered | Clean text | Contents entries | EPUBCheck | File size |
|---|---|---|---|---|---|
| Internet Archive's EPUB | 96.8% | 88.5% | 0 | 32 errors | 12.1 MB |
| Calibre on the scan | 0% | – | 0 | valid | 44.0 MB |
| OCRmyPDF, then Calibre | 0% | – | 0 | valid | 43.9 MB |
| upub | 99.2% | 99.2% | 39 | valid | 167 KB |
Measured on 2026-10-02 against the proofread Project Gutenberg #11 text (26,681 words). Words recovered: the share of that text found in the EPUB, in order. Clean text: the share of the EPUB's words that belong to the book, not to running heads, page numbers or OCR noise.
One honest caveat: the Internet Archive EPUB keeps the 47 Tenniel illustrations as pictures, and our sample does not include them. upub's figure extraction does not yet work on full-page scans, so the sample file carries the text only.
Calibre on a scanned PDF
Calibre's own manual says it plainly: complex, multi-column and image-based documents are not supported, and PDF is a bad input format. Our run confirms it. The EPUB Calibre produced is a stack of page pictures with 18 recognisable words, and its table of contents has 0 entries. On an e-reader it behaves like the original PDF, only larger.
None of this is a fault in Calibre. It converts text, and a scan contains no text. For an e-book you bought or a PDF exported from Word, Calibre is still the tool we would recommend first.
OCRmyPDF, then Calibre
OCRmyPDF runs Tesseract over each page and adds the recognised words as an invisible layer, so you can search and copy text while the page still looks like the scan. Its documentation describes exactly that: a searchable PDF, not a new layout.
The text layer itself is reasonable. Tesseract recovered 81.9% of the book's words in order. The problem comes next. When we fed that PDF to Calibre, the EPUB again held 17 words, and switching Calibre to its pdftohtml engine gave 15. The invisible text never reaches the e-book.
You can extract the layer by hand with pdftotext and build an EPUB from it, but then every running head, page number and line-end hyphen arrives in the middle of your sentences, and there are no chapters. That clean-up is the real work, and it is what most guides skip.
k2pdfopt
k2pdfopt is a clever tool with a different goal. Its author describes it as converting each page to a bitmap, finding the readable areas and laying them out again for a small screen. Since version 1.50 it can also embed OCR text so the result is searchable. The output is still a PDF of images: you cannot change the font, the dictionary cannot look up words reliably, and the file grows with every page.
We did not run k2pdfopt ourselves because its download sits behind a form, so the table above has no row for it. If you read on a 6-inch screen and only need a scan to fit, it is a good free option.
What about the Internet Archive's own EPUB?
Many scans already come with an EPUB, so it is worth checking first. For this book the Archive's EPUB recovered 96.8% of the words, but only 88.5% of its text is the book: the rest is running heads, page numbers and OCR noise such as "w^onder". It has 207 paragraphs for 214 pages because each page is one paragraph, an empty table of contents and 32 EPUBCheck errors. Our Internet Archive guide looks at this file in detail.
Which one should you use?
| Your file | Best tool | Why |
|---|---|---|
| PDF exported from Word or a publisher, text selectable | Calibre | Free, fast and good at reflowing real text |
| Scan you only need to search or archive | OCRmyPDF | Keeps the page exactly as scanned and adds search |
| Scan you read on a small screen and don't mind as images | k2pdfopt | Re-cuts the page to fit, no OCR clean-up needed |
| Scanned book you want to read as an e-book | upub | Reads the pages and rebuilds paragraphs, headings and contents |
A quick test tells you which group your file is in: open the PDF and try to select a sentence. If you can, it has text. If the whole page highlights as one picture, it is a scan. The device pages show the same 1872 scan converted for Kindle, Kobo, reMarkable, tolino and others.
Questions
- Can Calibre convert a scanned PDF to EPUB?
- Not into readable text. Calibre has no OCR, so a scanned PDF becomes an EPUB of page images. In our 214-page test it recovered 18 words of the book.
- Does running OCRmyPDF before Calibre fix it?
- Not in our test. OCRmyPDF adds an invisible text layer, and Calibre left it out with both of its PDF engines. The EPUB still held only page images.
- Is k2pdfopt an EPUB converter?
- No. k2pdfopt writes a new PDF with the page re-cut for a small screen. It can embed OCR text for searching, but the pages are still images.
- What is the best free way to read a scanned book on an e-reader?
- If the scan is on the Internet Archive, try its EPUB first and check a few pages. Otherwise k2pdfopt gives a free, readable PDF. For a true reflowable EPUB with chapters you need OCR plus structure recovery, which is what upub does.
Try it on your own scan
Upload a scanned PDF or a folder of page photos. You get an EPUB with real text, chapters and a table of contents.