PDF to EPUB conversion examples
Each case is a real PDF from our licensed or public-domain corpus, converted with the same engine you use, with the results of every engine we measured, including where ours loses.
Updated:
Every case on this page is a real PDF from our test corpus, converted with the same engine and the same default settings you get when you upload a file. The documents are public domain or openly licensed (CC BY, CC BY-SA), and each case names its source and licence, so you can download the original and check the result yourself.
For each one you get a page of the original next to the same passage in the EPUB, the sample EPUB to download, the EPUBCheck result and the score of every engine we measured on that document. Some cases show documents where another engine does better than ours; we publish those too.
The cases cover the layouts that break most converters: columns, repeated headers, footnotes, chapters and non-Latin scripts. Your own files are never used here: they stay private and are deleted automatically.
-
Two-column PDF to EPUB: an academic paper
A two-column arXiv paper converted to EPUB: reading order, per-engine results, EPUBCheck and observed limits. Our engine is not the best here.
OLMo paper (arXiv 2402.00838)
-
Three-column PDF to EPUB: the Federal Register
A three-column Federal Register document converted to EPUB: column order, rotated margin text, repeated headers, per-engine results and EPUBCheck.
Federal Register document 2026-22443
-
PDF footnotes to EPUB notes: an academic paper
Footnotes in a long arXiv paper converted to linked EPUB notes: what was detected, per-engine text results, EPUBCheck, and where other engines do better.
QLoRA paper (arXiv 2305.14314)
-
Removing repeated headers: the 9/11 Commission Report
The 9/11 Commission Report (public domain) converted to EPUB with running headers and page numbers removed: margin leak per engine and EPUBCheck.
The 9/11 Commission Report (GPO)
-
Chapters and table of contents from a PDF book
A LaTeX-typeset public-domain textbook converted to EPUB with chapters from the PDF outline: results per engine, EPUBCheck and the limits with formulas.
Calculus textbook typeset in LaTeX (Project Gutenberg #33283)
-
Arabic PDF to EPUB: right-to-left text
An Arabic Wikipedia article converted to a right-to-left EPUB: word order, per-engine results, EPUBCheck and limitations.
Wikipedia (Arabic): One Thousand and One Nights
-
Japanese PDF to EPUB
A Japanese Wikipedia article converted to EPUB: text without spaces between words, per-engine results, EPUBCheck and limitations.
Wikipedia (Japanese): The Tale of Genji
-
Spanish PDF to EPUB: accents, «» and ¿?
A Spanish Wikipedia article (50 pages) converted to EPUB: accents, «» quotes and ¿? kept, language tagged es, EPUBCheck clean; pdf_oxide scores higher.
Wikipedia (Spanish): Don Quijote de la Mancha
-
French PDF to EPUB: guillemets and apostrophes
A French Wikipedia article converted to EPUB: « » with spaces, apostrophes and œ kept, tagged fr. Our 8-gram recall is 0.819; pdf_oxide reaches 0.903.
Wikipedia (French): Les Misérables
-
German PDF to EPUB: umlauts, ß and „quotes“
An 86-page German Wikipedia article converted to EPUB: umlauts, ß and „…“ quotes kept, tagged de. 8-gram recall 0.845 against 0.877 for pdf_oxide.
Wikipedia (German): Johann Wolfgang von Goethe
-
PDF charts in EPUB: what is lost
Vector charts in the 487-page Economic Report of the President do not survive our conversion: lines are lost, axis labels become text. Real evidence.
Economic Report of the President 2024
-
What happens with a scanned PDF
A scanned 1867 book with an old OCR layer: EpubCC has no OCR, so it reuses that layer, errors and all. 8-gram recall 0.513, below PDFium's 0.651.
Alice's Adventures in Wonderland, 1867 printing (scan, Internet Archive)
-
Text wrapped around images: a known limit
When text flows around a floating image, our engine splits the paragraph and can mix in the caption. A real sample from an English Wikipedia PDF.
Wikipedia (English): Alice's Adventures in Wonderland
-
Government report PDF to EPUB: footnotes kept
A 58-page GAO report to EPUB: 50 footnotes found, 47 linked to their calls, running footers removed, EPUBCheck clean. Tables and figures stay limited.
gao-24-106219
-
EU regulation PDF to EPUB, in Spanish
The 144-page Spanish AI Act from the EU Official Journal to EPUB: 125 chapters, headers removed, precision 0.982. Numbered points can come loose.
eurlex-2024-1689-es
-
PDF with equations to EPUB: the honest result
A CC BY arXiv paper with display equations to EPUB: prose and inline symbols survive, display equations come out garbled. No MathML yet. Real sample.
arxiv-dpo
-
Line-numbered PDF to EPUB: not handled yet
A line-numbered U.S. Senate bill to EPUB: the margin numbers become a stack of stray paragraphs and words split. Every engine struggles; ours too.
govinfo-bill-118s1409