Best EPUB Translation Tools: 15 Options Tested and Compared

Translating an EPUB is not the same problem as translating a document. A Word file is mostly a linear stream of text. An EPUB is a zipped website: a folder of XHTML chapter files, a stylesheet, an OPF manifest listing every asset, a navigation document, embedded fonts, cover metadata, and internal anchors that link footnotes to their markers. Run that through the wrong tool and you get back 400 pages of correct Spanish with a broken table of contents, orphaned footnotes, and a cover image the reader app refuses to display.

So the question is never just “which translator is most accurate.” It is which tool understands the container.

This roundup covers 15 EPUB translation tools, from a free Calibre plugin that costs nothing but your own API key, through per-book web services, up to enterprise translation management systems that publishers use to keep a nine-language series in sync. Every entry says what it actually does to your file, what it costs, and who should skip it.

The short answer

What actually breaks when you translate an EPUB

Before comparing tools, it helps to know the specific failure modes. Every tool below succeeds or fails on these six things.

Inline markup gets swallowed

A sentence like <p>She read <em>Middlemarch</em> twice.</p> contains three text nodes and one tag. Naive tools flatten the paragraph to plain text, translate it, and paste it back without the emphasis. Good tools segment around the tags and reattach them to the right words in the target language, which is harder than it sounds when word order flips.

The navigation document goes stale

EPUB 3 files carry a nav.xhtml; EPUB 2 files carry a toc.ncx. Both store chapter titles as their own strings, separate from the headings inside the chapters. Translate the chapters but not the nav and your reader shows a Spanish book with an English table of contents.

Metadata stays in the source language

The OPF file holds title, author, description, and a language code. That language code matters: it drives hyphenation, text-to-speech voice selection, and how stores categorise the book. A translated EPUB still declaring <dc:language>en</dc:language> will be read aloud by an English voice.

Footnote anchors lose their partner

Footnotes are two elements linked by an ID. Anything that rewrites IDs, reorders content, or strips attributes turns a working footnote into dead text. Academic and non-fiction titles suffer worst here.

Text expands and the layout does not

English to Spanish or French typically runs 15 to 25 percent longer. German can run longer still, with compound nouns that overflow fixed-width elements. Reflowable EPUBs mostly absorb this. Fixed-layout EPUBs, the kind used for cookbooks, children’s books, and technical manuals with diagrams, do not.

The file stops validating

Amazon KDP, Apple Books, and Kobo all run structural checks on upload. An EPUB that opens fine in Calibre can still be rejected for an unescaped ampersand introduced by a translation engine. Run EPUBCheck before you submit anything.

How these tools were judged

Five criteria, applied consistently:

1. Container awareness. Does it parse the EPUB structure, or does it convert to something flatter and convert back?
2. Output you can keep. A downloadable, valid EPUB, or a rendering you can only read inside the tool?
3. Engine choice. Locked to one machine translation provider, or can you point it at DeepL, GPT, Claude, or Gemini?
4. Cost shape. Per book, per character, per month, or free with your own API key.
5. Human override. Can a translator or editor correct segments, or is the output take-it-or-leave-it?

Pricing was checked in August 2026 and moves often. Treat the figures as a shape, not a quote.

1. Calibre plus the Ebook Translator plugin

The strongest free option, and the one most people land on eventually.

Calibre already understands EPUB better than almost any consumer software, because ebook conversion is what it was built for. The Ebook Translator plugin by bookfere adds a translation layer on top: it walks the XHTML files, extracts translatable strings, sends them to an engine of your choosing, and writes a new EPUB with the structure intact.

Engine support is the selling point. Google Translate, DeepL, ChatGPT, Claude, Gemini, and custom endpoints all work, so you can run a cheap pass with Google to check the pipeline, then rerun the same book through a better model once you trust the setup. It handles 48 input formats and 20 output formats, and it can produce bilingual output with source and translation stacked paragraph by paragraph.

The cost is your time. You install Calibre, install the plugin, get an API key, configure the engine, and tune the request interval so you do not hit rate limits halfway through chapter nine. First run on a full-length book usually involves at least one restart.

Use it if you are translating more than two or three books, care about which model does the work, and would rather own the pipeline than rent it.

2. Immersive Translate

Built for reading, not for publishing, and it is very good at reading.

Immersive Translate started as a browser extension that renders bilingual web pages with the translation interleaved under each original paragraph. The same treatment now applies to EPUB files, which makes it the default recommendation for language learners working through a novel in the original.

It supports over 20 translation engines, including DeepL, OpenAI, Claude, and Gemini on the paid tier. Formatting fidelity is better than most competitors because the tool was designed around preserving reading layout rather than producing a clean deliverable.

That last distinction is the catch. Bilingual reading inside the app is not the same as a clean monolingual EPUB you can sideload to a Kindle, hand to an editor, or list on a store. If your goal is a publishable file, this is the wrong shape of tool.

Pro sits around $9.99 per month with cheaper annual and lifetime rates. The free tier is usable for reading and gets tight on batch work.

3. BookTranslator.ai

Upload EPUB, choose language, download EPUB. That is the entire product, and there is real value in that.

Pricing is fixed per book rather than per character, which removes the anxiety of watching a meter run: roughly $5.99 per 100,000 words on the standard tier and $9.99 per 100,000 words on the Pro tier, which routes to more capable models for better contextual accuracy. Support runs past 99 languages.

For a single book, the maths is hard to argue with. A 90,000-word novel translated for under ten dollars, with no API keys, no rate limit tuning, and no plugin installs, is a genuinely good trade for anyone who is not doing this weekly.

What you give up is control. You cannot pin a glossary of character names across a series, you cannot edit segments and reuse them, and you cannot swap the model when a chapter comes out flat.

4. OpenL Translate

OpenL is a general-purpose AI translator with a dedicated EPUB path. It parses the EPUB structure, translates the content files, and aims to leave CSS, images, and navigation untouched.

Plans start near $7.90 per month, with a limited free tier. Because you are paying for a subscription rather than per book, it works out better than per-book services once you cross three or four titles a month, and worse if you translate one book a year.

The subscription also covers documents, web pages, and images, so it is a reasonable pick if EPUB is one of several formats you deal with rather than the only one.

5. EPUB Translator (epubtranslation.com)

A purpose-built single-job tool. It reads the EPUB structure directly and translates content while leaving CSS, navigation, and chapter hierarchy alone.

There is not much more to say about it, which is the point. Narrow tools that do one thing tend to break in fewer places than broad platforms with an EPUB feature bolted on. If your book is structurally simple, reflowable prose with a normal chapter structure, a tool like this will usually get you 95 percent of the way in one pass.

Complex titles with sidebars, tables, indexes, or fixed layouts are where narrow tools run out of road.

6. DocTranslator

The free-tier workhorse. DocTranslator accepts EPUB and covers 120 or more languages without an account for smaller files.

Quality sits where you would expect for a free service: adequate for understanding, not for selling. Use it to sanity-check whether a translation project is viable before you spend anything, or to make a foreign-language reference book readable for your own use.

Do not put its output on a storefront.

7. DeepL (Pro and API)

DeepL is the engine, not the pipeline, and that distinction trips people up constantly.

Its document translation handles Word, PowerPoint, PDF, and a few others. EPUB is not on that list. What DeepL gives you is the highest-regarded neural translation quality for European language pairs, exposed through an API that every serious EPUB tool on this page can call.

The API free tier allows 500,000 characters per month at no cost, which is enough to translate roughly one full-length novel. Beyond that, DeepL API Pro runs a small monthly fee plus a per-million-character rate. Document translation bills a minimum of 50,000 characters per file, which matters if you are pushing many small files rather than one large one.

The web and desktop subscriptions are a separate product from the API with separate limits. People buy the wrong one regularly.

Practical setup: DeepL API key plugged into Calibre’s Ebook Translator plugin. That combination is the quality-per-dollar leader for European languages.

8. Google Cloud Translation Advanced

Same situation as DeepL, with wider language coverage and weaker literary feel.

Google Cloud Translation Advanced supports document translation for DOCX, PPTX, PDF, and XLIFF. EPUB is absent, so you build an extract, translate, repackage pipeline yourself or let a tool do it for you. Where Google wins is breadth: language pairs that DeepL simply does not offer, particularly across South Asian, Southeast Asian, and African languages.

Glossary support is genuinely useful for series work. You upload a term list, and character names, place names, and invented vocabulary stay consistent across every chapter and every book.

Plain Google Translate, the free web tool, remains fine for spot-checking a paragraph. It is not a book pipeline.

9. Smartcat

A translation management system with an AI pre-translation step and a built-in marketplace of human linguists, which is the combination that makes it interesting for books.

The workflow: machine-translate the whole book, then hire an editor through the platform to post-edit the segments that matter, with translation memory capturing every correction. Book two costs less than book one because the memory already knows how you translated your recurring phrases.

The editor supports over 70 file formats. EPUB is not one of them natively, so you extract to XLIFF, DOCX, or HTML first, translate, then repackage.

That extra step is real friction for a single title and rounding error for a catalogue.

10. Crowdin

Crowdin comes from the software localisation world and brings that discipline to publishing: version control, branching, translation memory, glossaries, and reviewer approval gates.

Like Smartcat, it does not ingest EPUB directly. The usual route is extracting XLIFF or HTML from the EPUB, translating inside Crowdin, then repackaging.

Where it earns the setup cost is ongoing revision. A textbook that ships a new edition every year, a manual that updates quarterly, a series where book four retroactively renames a character: these are localisation problems, and Crowdin was built for exactly this. Change one source string and every language flags for review automatically.

Translating one standalone novel this way is using a forklift to move a chair.

11. memoQ

The professional translator’s choice, and the tool your freelance translator will likely already own.

memoQ combines a feature-dense CAT editor with project management, which is why language service providers standardised on it. It handles 30-plus file formats across 100-plus languages, with the segment-level control, concordance search, and quality assurance checks that professional literary translation actually requires.

For EPUB you pair it with an extraction step, most commonly Okapi. The payoff is that a human translator can work at professional speed with machine translation as a suggestion layer rather than a replacement.

If you are paying a human to translate your book, ask which CAT tool they use before you pick a platform. Fighting your translator’s toolchain wastes more money than the licence costs.

12. Okapi Framework (Rainbow)

Free, open source, and the missing piece for every TMS on this list.

Okapi is a set of components for building localisation pipelines. Rainbow is its desktop front end. Between them they provide filters that read structured formats and emit XLIFF, translate it through whatever engine or CAT tool you like, then merge the translated XLIFF back into the original structure.

The merge step is what makes it valuable. Reassembling a valid EPUB from translated fragments by hand means editing the OPF, fixing the nav, and re-zipping in the correct order with the mimetype file stored uncompressed and first. Okapi does that reliably.

Steep learning curve. Zero licence cost. If you plan to translate books repeatedly and want a pipeline nobody can price you out of, learn this one.

13. OmegaT

A free-libre CAT tool under GPLv2, and the open-source answer to memoQ.

OmegaT gives you translation memory, glossaries, fuzzy matching, and machine translation plugins without a subscription. With the Okapi plugin installed it reads more than 50 formats. It is unglamorous, stable, and used daily by professional translators who prefer not to rent their working environment.

Pair it with Okapi Rainbow for extraction and you have a complete, free, offline book translation pipeline. Nothing leaves your machine except the API calls you choose to make, which matters for unpublished manuscripts under embargo.

14. oomol-lab/epub-translator

An open-source Python tool that translates EPUBs with large language models and keeps the original text alongside the translation, producing a bilingual book.

It needs Python 3.11 or newer and a terminal. In exchange you get full control over the prompt, the model, and the chunking strategy, plus output that is genuinely useful for language learners and for cross-referencing a translation against its source during review.

The wider GitHub ecosystem here is worth a look. quantrancse/epub-translator handles Google Translate with custom dictionary support, and several projects wrap LLM APIs with glossary and batch handling. One project reports translating a 300-page book in about two hours for roughly six dollars in API spend.

Maintenance is the risk. Community scripts break when APIs change and nobody is contractually obliged to fix them.

15. Claude, ChatGPT, or Gemini driving a script

The newest pattern, and the one improving fastest.

Instead of a translation API that sees one segment at a time, you feed a large language model whole chapters with instructions: keep this character’s voice formal, preserve these invented terms exactly, match the register of the source, return valid XHTML. Long context windows mean the model can hold a book’s terminology in mind in a way that segment-based machine translation cannot.

Tools like vbalko-claimate/epub-translator formalise this, driving the translation through an AI coding assistant with glossary and batch support.

Two real problems. Models sometimes rewrite rather than translate, quietly improving a sentence the author wrote deliberately. And they occasionally return malformed XHTML, which is fatal in an EPUB. Validate every chapter.

For literary fiction, marketing copy, and anything where voice matters more than literal accuracy, this approach now beats traditional machine translation. For legal, medical, or technical text where a mistranslation carries consequences, it does not.

Side-by-side comparison

Tool	Native EPUB	Engine choice	Output	Cost shape	Best for
Calibre + Ebook Translator	Yes	Google, DeepL, GPT, Claude, Gemini, custom	Clean or bilingual EPUB	Free plus API usage	Repeat translators who want control
Immersive Translate	Yes	20+ engines	Bilingual reading	~$9.99/mo	Language learners
BookTranslator.ai	Yes	Fixed by tier	Clean EPUB	~$5.99-9.99 per 100k words	One-off books, zero setup
OpenL Translate	Yes	Managed	Clean EPUB	From ~$7.90/mo	Mixed-format translation needs
EPUB Translator	Yes	Managed	Clean EPUB	Per file	Simple reflowable books
DocTranslator	Yes	Managed	Clean EPUB	Free tier	Feasibility checks
DeepL	No	DeepL only	API strings	500k chars free, then metered	European language quality
Google Cloud Translation	No	Google only	API strings	Metered	Rare language pairs, glossaries
Smartcat	No, extract first	Multiple plus humans	XLIFF to repackage	Platform plus linguist fees	MT draft plus paid post-editing
Crowdin	No, extract first	Multiple	XLIFF to repackage	Tiered platform	Series and revised editions
memoQ	No, pair with Okapi	Multiple	Professional CAT output	Licence	Working with a human translator
Okapi Rainbow	Extraction and merge	Engine agnostic	Valid rebuilt EPUB	Free	Building your own pipeline
OmegaT	With Okapi plugin	MT plugins	Translated project files	Free, GPLv2	Offline and confidential work
oomol-lab/epub-translator	Yes	LLM of choice	Bilingual EPUB	Free plus API usage	Developers and learners
LLM-driven scripts	Depends on script	Any frontier model	Whatever you build	API usage	Voice-sensitive fiction

The two tools you need regardless of which translator you pick

Sigil

Sigil is a free, open-source EPUB editor for Windows, macOS, and Linux, handling both EPUB 2 and EPUB 3. It has been maintained since 2009.

You will need it. Something will come back wrong: a heading that lost its style, a nav entry pointing at the wrong file, a stray tag. Sigil lets you open the EPUB, see the actual XHTML, fix the specific thing, and save without rebuilding the whole book.

EPUBCheck

The official validator from the W3C. Run it after every translation pass and before every store upload. It catches the structural problems that reader apps silently tolerate and retail platforms reject.

Sigil can run EPUBCheck for you, which makes the two a natural pair.

A workflow that survives contact with a real book

Here is the sequence that avoids the expensive mistakes.

Translate one chapter first. Not the whole book. Pick chapter three, run it through your chosen tool, and read the output carefully. You are checking whether the tool understands your book’s specific quirks before you spend money on 400 pages of the same mistake.

Build a glossary before the full run. Character names, place names, invented terminology, product names, recurring phrases with a fixed rendering. Feed it to the tool if it accepts one. If it does not, that tells you something about whether the tool fits a book-length job.

Keep the source file untouched. Work on copies. You will run the pipeline more than once.

Validate immediately after translation, before you look at quality. A structurally broken file is not worth reading.

Get a native speaker on the first and last chapters. Openings and endings do the most work, and they are where machine translation flatness is most visible to a reader deciding whether to buy. Two chapters of professional review costs a fraction of a full post-edit and catches the systemic problems.

Test on the actual device. Kindle, Kobo, Apple Books, and a phone. Rendering differs, and a Kindle in particular treats CSS very differently from a desktop reader.

What machine translation still gets wrong in books

Engine quality has improved enough that a machine draft is now genuinely readable. Readable is not the same as publishable, and the gap shows up in predictable places.

Register drifts. A character who speaks in clipped, working-class sentences comes back speaking neutral standard prose. Machine translation defaults to a middle register regardless of the voice in the source.

Pronoun and formality choices go unmarked. Languages with formal and informal second person force a decision the English source never made. Get it wrong and two characters who should be intimate address each other like strangers for 300 pages.

Wordplay dies quietly. Puns, rhymes, and acrostics come back as literal nonsense with no flag telling you something was lost.

Cultural references land flat. A reference that carries weight for an American reader may mean nothing to a Brazilian one, and a good human translator substitutes rather than translates.

Consistency erodes across long documents. Segment-based engines have no memory of chapter one when translating chapter twenty. This is the single strongest argument for translation memory or a glossary, and it is why LLMs with long context windows have an edge for fiction.

For non-fiction, technical documentation, and reference material, machine output with a light edit is often good enough to sell. For literary fiction, poetry, and anything where a reader is paying for the prose, it is not.

The rights question nobody asks until it matters

Translating a book you did not write and distributing that translation is a separate act of copyright infringement from copying the original. The translation right is held by the copyright owner, and it is routinely sold separately by territory and language.

Translating a purchased ebook for your own reading sits in a very different place from listing that translation for sale. If you are working with your own book, none of this applies. If you are not, get the rights in writing before you spend a cent on tooling.

Public domain titles are the safe playground. Project Gutenberg has plenty, and they make excellent test files for evaluating any tool on this list.

Selling the translated edition

Once you have a clean file, distribution is its own decision. Amazon KDP, Apple Books, Kobo, and Google Play Books all accept EPUB uploads and handle territory rights per language edition.

Selling direct keeps the margin. A WordPress site running WooCommerce or Easy Digital Downloads handles EPUB delivery, licence keys, and per-region pricing without giving up 30 to 65 percent to a platform. If you are setting that up, our guide to building an ecommerce store in WordPress covers the storefront side, and the same site can host the translated marketing pages.

Which raises the obvious follow-up: your book is now in six languages and your website is in one. The WordPress translation plugin comparison covers the site-side tools, and the approach in scaling ecommerce with AI product description translation applies directly to book listings and blurbs.

One thing people miss: the same glossary you built for the book should govern the store copy. Nothing undermines a translated edition faster than a product page calling the protagonist by a different name than the book does.

Frequently asked questions

What is the best free EPUB translation tool?

Calibre with the Ebook Translator plugin. The software is free and open source, it produces a valid EPUB, and paired with DeepL’s free API tier of 500,000 characters per month you can translate a full-length novel at no cost. For a quick no-install option, DocTranslator handles EPUB free at lower quality.

Can Google Translate handle an EPUB file?

Not directly. Google Translate’s document upload does not accept EPUB, and Google Cloud Translation Advanced supports DOCX, PPTX, PDF, and XLIFF but not EPUB. You need a tool that extracts the content, calls Google, and repackages the file.

How much does it cost to translate a book with AI?

A 90,000-word book runs between zero and about fifteen dollars in machine translation costs, depending on whether you use a free tier, an open-source script with an LLM API, or a per-book service. Human post-editing to publishable quality typically runs $1,500 to $4,500 for the same book.

Will translation break my EPUB formatting?

It depends entirely on whether the tool parses the EPUB container or flattens it. Tools that work at the XHTML level, including Calibre’s plugin and the purpose-built EPUB services, preserve CSS, navigation, and images. Anything that converts your EPUB to plain text or DOCX and back will cost you formatting. Run EPUBCheck afterwards either way.

Is AI translation good enough to sell?

For technical documentation, manuals, and most non-fiction, machine output with a light native-speaker edit sells fine. For literary fiction and poetry, readers notice. The pragmatic middle path is a machine draft plus paid post-editing of the opening chapters, sample pages, and blurb, which is where buying decisions are made.

Which languages should a self-publisher translate into first?

Spanish, German, and French have the largest ebook-buying markets outside English and the best machine translation quality, which makes them the lowest-risk starting point. Portuguese and Italian follow. Machine translation quality drops noticeably for languages with less training data, so budget more for human review there.

Can I translate a DRM-protected EPUB?

No tool on this list will open a DRM-protected file, and stripping DRM is illegal in many jurisdictions regardless of what you intend to do next. Buy or produce a DRM-free copy.

Choosing, in one paragraph

Translating one book with no technical appetite: BookTranslator.ai or OpenL. Translating books regularly and willing to spend an afternoon on setup: Calibre plus the Ebook Translator plugin with a DeepL or Claude key, which will outperform every paid consumer service on quality per dollar. Reading rather than publishing: Immersive Translate. Running a catalogue with revisions and multiple languages: Crowdin or memoQ with an Okapi extraction step. Working with a professional translator: ask them first, then buy what they already use.

Whichever you pick, the two non-negotiables are the same. Translate one chapter before you commit to the whole book, and validate the file before anyone else sees it.