Turn a PDF into an EPUB you can actually read

A PDF is a fixed rectangle. That is the whole idea of the format and it is exactly the wrong idea for a six-inch screen: you get an A4 page shrunk until the text is too small, or you pan around a page you cannot see all of. An EPUB reflows to your screen, at your font, in your size.

So converting is worth doing, and it is more than renaming the file. It means recovering a structure the PDF does not contain.

Chapters out of headings

A PDF has no chapters and no headings. It has glyphs at coordinates, and what you read as a heading is a line set larger than the ones around it. The same reading that produces our Markdown and HTML output works that out: blocks of lines, a larger block is a heading, a block starting with a bullet or a number is a list item.

The book is then split at the top-level headings, and those become the table of contents. A document with no headings becomes a single chapter, which is long but honest, and a reader handles a long chapter perfectly well.

There is one guard worth explaining. Heading levels come from font sizes, so a report with a large title stamped on every page produces a top-level heading per page. Split on those and a three-hundred-page report becomes three hundred chapters of one paragraph each, which is unusable in a way a long chapter is not. Past a limit the whole document becomes one chapter instead.

The one rule e-readers actually enforce

An EPUB is a zip with a particular shape, and readers are relaxed about most of it. They are not relaxed about the first entry: a file called mimetype, stored rather than compressed, holding exactly application/epub+zip.

Get that wrong and you have a file that opens perfectly in any zip utility and is rejected by every e-reader. That is the kind of bug that ships, because a zip utility is what a developer checks it in. So the tests here pack a real book and open it the way a reader does: first entry, stored, exact bytes, then the package file, the navigation document and every chapter it claims to have.

Each book gets its own identifier

An EPUB needs an identifier that no other book shares, and there is nothing in a PDF that reliably is one. So each conversion gets a fresh UUID, which says “this file” rather than “this work”. That is what a reader’s library actually keys on, and it means converting the same PDF twice gives you two entries rather than one silently replacing the other.

What it does not carry over

Pictures, tables as tables, columns as columns, footnote links. Text is what a PDF makes recoverable with confidence; the rest is layout that would have to be inferred, and inferring it produces a book that looks nearly right and is wrong where you cannot see it. For pictures, Extract images pulls them out at full size.

If the file is a scan, there is no text layer to recover and the tool says so rather than handing you an empty book. Run it through OCR first.

Nothing leaves the device

The document is read and the book is packed by a worker inside your browser. No upload, no account, no server copy, and no size limit imposed by somebody else’s free tier.