If you buy manga on Kobo, Google Play Books, or BookWalker, there's a good chance what you own is an EPUB. And if you've tried to run one through a translation tool, you've probably hit a wall — either the tool doesn't accept the file at all, or it opens it and finds nothing to translate.

Both failures have the same cause, and it's worth understanding before you try to work around it.

What's Actually Inside a Manga EPUB

An EPUB is a ZIP archive with a specific structure. Rename volume.epub to volume.zip, unzip it, and you'll find something like this:

volume.epub/
├── META-INF/
│   └── container.xml
└── OEBPS/
    ├── content.opf        ← metadata, file list, reading order
    ├── toc.ncx            ← table of contents
    ├── images/
    │   ├── page001.jpg    ← the actual manga pages
    │   ├── page002.jpg
    │   └── ...
    └── xhtml/
        ├── page001.xhtml  ← a wrapper that displays page001.jpg
        └── ...

That last part is the thing that trips people up. In a manga EPUB, the pages are images, and the XHTML files are just thin wrappers that tell the reader app to display one image per screen. There's no selectable text anywhere — the dialogue lives inside the JPEG, exactly as it does in a scan.

This is called a fixed-layout (or pre-paginated) EPUB, and it's a different animal from the reflowable EPUBs used for novels, where the text is real text that resizes with your font settings.

So when a translation tool "opens the EPUB and finds nothing," it's because the tool was looking for text nodes in the XHTML. The text isn't there. It's in the pixels.

Why This Differs From PDF and CBZ

All three formats end up holding page images, but they differ in what surrounds them:

	PDF	CBZ	EPUB (fixed-layout)
Container	PDF object structure	Plain ZIP	ZIP with XML metadata
Pages	Embedded images, sometimes with a text layer	Image files, sorted by filename	Image files, order defined in `content.opf`
Reading direction	Not recorded	Not recorded	Recorded explicitly
Nested compression	No	No	Yes — images sit inside a ZIP
DRM common?	Sometimes	Almost never	Often

Two rows there matter more than the rest.

Reading direction is metadata, not a guess

Open content.opf in a manga EPUB and look at the spine element. You'll typically see:

xml

<spine page-progression-direction="rtl">

rtl means right-to-left — the file is telling the reader app that this is Japanese-style manga and pages advance leftward.

This is genuinely useful, and it's something neither PDF nor CBZ gives you. With a CBZ, a tool has to infer reading direction from the artwork itself, which is a harder problem than it sounds. With an EPUB, the publisher already told you.

The practical consequence: if you rebuild an EPUB after translating and drop that attribute, the volume will read backwards. Preserve the spine.

The nesting affects quality

A CBZ is a flat ZIP of images — unzip it and you have the originals, byte for byte.

An EPUB stores its images inside a ZIP too, but the images themselves were usually already compressed by the publisher, sometimes aggressively, to keep file sizes manageable for reader apps. Some manga EPUBs ship pages at resolutions noticeably lower than the equivalent CBZ scan.

That matters for translation, because OCR accuracy degrades with resolution long before the page looks bad to a human eye — and sound effects degrade first. If you have the same volume in both formats, use the higher-resolution one.

The DRM Reality

Here's the part most guides skip.

EPUBs bought from major stores are usually wrapped in DRM — Adobe's ACS, Kobo's own scheme, or a store-specific variant. A DRM-protected EPUB isn't really a readable ZIP: the content is encrypted, and only an authorized reader app holding the right key can display it.

No translation tool can process a DRM-protected file. Not because of a missing feature, but because the images aren't accessible to anything except the authorized reader. If you unzip one, you get encrypted blobs.

We don't provide instructions for removing DRM. Circumventing it is illegal in many jurisdictions regardless of whether you paid for the book, and we're not going to pretend otherwise.

What you can do:

Check whether your file is DRM-free first. Rename a copy to .zip and try to unzip it. If you get a normal folder structure with viewable images, there's no DRM. If you get errors or unreadable files, there is.
Buy from DRM-free sources when it matters to you. Some publishers and storefronts sell manga EPUBs without DRM, and doujin and indie releases frequently are. Store listings usually say.
Use a screen-based approach for what you legitimately own. Reading a DRM-protected volume in its authorized app and translating what's on screen is a different situation from breaking the file open. It's slower and page-by-page, but it doesn't involve circumventing anything.

Translating a DRM-Free Manga EPUB

Assuming your file opens normally, the workflow is straightforward:

1. Confirm it's fixed-layout. Unzip a copy and look in the images folder. If you see one image per page, you have an image-based manga EPUB and you'll need OCR-based translation. If instead you find real text in the XHTML, you have a reflowable ebook and an ordinary text translator will handle it better.

2. Note the reading direction. Check the spine attribute in content.opf before you do anything else. Write it down.

3. Extract the page images in spine order. Not filename order — spine order. Most manga EPUBs happen to name files sequentially, but not all do, and the content.opf manifest is the authoritative sequence.

4. Translate the images. This is the same job as translating any manga scan: detect speech bubbles, OCR vertical Japanese, translate with page context, inpaint the original text, typeset the new text. Our manga translator accepts image batches directly, so extracted pages can go straight in.

5. Decide what to rebuild. Most people don't need an EPUB back — a folder of translated images or a CBZ is easier to read and easier to fix if one page came out wrong. Rebuild an EPUB only if you specifically want it in a reader app that requires the format.

If you do rebuild: keep the original content.opf structure, swap only the image files, keep the filenames identical, and preserve page-progression-direction. Then validate with EPUBCheck before loading it anywhere.

Which Format Should You Prefer?

If you get a choice for the same volume:

CBZ is the easiest to work with — flat, uncompressed structure, no DRM, no metadata to preserve.

PDF is fine, though page extraction quality varies depending on how the file was produced.

EPUB is the most awkward for translation, but it's what most legitimate purchases come as, and its reading-direction metadata is genuinely worth having.

For personal reading, the format is mostly a logistics detail — what actually determines the result is the resolution of the pages inside and the quality of the OCR and typesetting applied to them.

Want to try it on a few pages first? AI Manga Translator gives you 5 free pages with no signup — enough to check whether your extracted EPUB pages come out at usable quality before you process a whole volume.

New to this? Start with our complete guide to translating manga.