AI book translator

An AI book translator for whole books and series

Libris Translate translates a complete EPUB, or a series volume after volume, with the language model you choose. It reads the book before translating it, keeps the same names from the first chapter to the last, and gives you an EPUB ready for the stores and a map of the passages to proofread first.

Figures read from Libris Translate 0.28.0 and from one book measured on 1 October 2026. Quality, languages, speed and cost depend on the model you connect.

How it translates a book

  1. Add the book: an EPUB, chapter files, or one long file split at its chapter headings. Import a book
  2. It reads first: a Book Bible of characters, places and terms, and a glossary shared by the volumes of a series. Book Bible
  3. It translates and reviews on its own: retries, fallback providers and arbitration of the remaining doubts, without asking you to approve each step. Autopilot
  4. You download: an EPUB checked with EPUBCheck, a bilingual EPUB, a Word file or the chapters as text, with the autopilot’s report. Exports

What a book needs that a sentence translator forgets

  • The same names everywhere: locked glossary terms, character genders and a series memory; volume 3 never reads volume 5. Series consistency
  • Where to proofread first: a score from 0 to 100 per passage, weakest first, computed without an extra model call. Quality
  • Your corrections win: a side-by-side editor with history; the autopilot never rewrites a passage you corrected. Editor
  • Webnovels still being published: new chapters are translated on their own, with the earlier ones as context. Source feeds

Formats and languages

  • In: EPUB 2 and 3, up to 256 MB per file by default, with layout, styles and images kept; chapters as TXT, Markdown, HTML or DOCX.
  • Out: an EPUB checked with EPUBCheck, a bilingual EPUB (interleaved or side by side), a Word file with your corrections as tracked changes, or text, Markdown and ZIP.
  • Languages: the pair you choose, as far as your model handles it. Interface in English and French.
  • Not handled: PDF, fixed-layout EPUB (comics, picture books) and office documents with their layout.

Where the manuscript goes

Libris Translate runs in Docker on your own Linux server. Books, translations and history stay there; only the passages being translated go to the model provider you configured, and with a local model nothing leaves your network. No account on this site, no upload to us. Install steps

What it costs

Two separate costs: the Libris Translate licence, and the model, which you pay to its provider (or nothing more than electricity on your own hardware).

  • Personal: 50 € once, for non-commercial use, 200 000 words a month.
  • Studio: 29 € a month or 290 € a year, 3 000 000 words a month, with the API and book sharing.
  • Pro: quoted, 12 000 000 words a month.
  • Trial: 30 days free, 100 000 words a month, no payment card. The free trial · All the plans

What a real book cost: measured, not estimated

Measured on 1 October 2026, Libris Translate 0.25.0, quality “high”, default autopilot: Alice’s Adventures in Wonderland (public domain), English → French, 26,530 words, translated end to end. 2,695 model calls, 39,832,243 tokens sent and 2,026,109 received, analysis, checks, three autopilot rounds, escalations and failed calls included. Translation model: GPT-6 Luna; GPT-6 Sol for the passages the autopilot escalates (wordplay, arbitration). The book was translated with a ChatGPT subscription through the Codex bridge: the amounts below apply the APIs’ list prices to the same tokens.

Brought to 100,000 words, that is $107 of model cost at list prices. Every model, at the same token counts

What it does not do

Questions

Is there a free version?
A 30-day trial with 100 000 words a month, without a payment card. You still pay your model provider, unless the model runs on your own hardware.
Which languages can it translate?
Any pair your model handles. Libris Translate does not bundle a model: quality depends on the one you connect, and you can compare several on the same passages first. Compare providers
Does it replace a human translator?
No. It writes a consistent first draft and shows where to look first; a person reads it before anything is published.
Do I need a GPU?
No, when the model runs elsewhere: a Linux (amd64) server with Docker is enough. A GPU only matters if you run the model yourself.
Can it translate a whole series?
Yes: volumes share a series memory and a glossary, and new chapters of a webnovel are translated as they arrive. Series