KMReader 7.2 adds manga translation to the reader: text bubbles are detected, recognized, translated, and typeset back into the page — on iPhone, iPad, and Mac. Here’s how it works and why it’s built the way it is.
On-device OCR, replaceable models
Text detection and recognition run entirely on your device. The CoreML model packs are downloaded on demand from a catalog, verified against their published hashes, and stored outside the app bundle — you can delete them at any time to reclaim storage, and detection and recognition models can be mixed to trade speed for accuracy.
Because OCR is local, page images never leave your device. Not to us — we don’t run servers for KMReader — and not to anyone else.
Your translator, your choice
Recognized text still needs a translation engine, and that choice is yours:
- Apple’s built-in translation works with zero setup, so recognized text stays on your device too.
- Any OpenAI-compatible provider — your endpoint, your model, your API key. Only the recognized text (and optionally the comic’s title) is sent to the provider you configured. Page images are never part of the request.
Set a default source language and a target language once in Settings → Manga Translation, and every book after that is one tap away.
Lettered back into the page
Translations are rendered back into the bubbles rather than dumped into a side panel: vertical text stays vertical, and the result works across paged, right-to-left, split, and Webtoon presentations. Translated pages are kept for the whole reading session, upcoming pages are translated ahead as you read, and a long-press on the floating translation action jumps straight into settings when a book needs a different setup.
Try it when it ships
KMReader 7.2 is currently in preparation. Once it lands on the App Store: download a model pack in Settings → Manga Translation, open any manga, and tap the floating translation action. The feature guide walks through the details, and the KMReader page has the full feature overview.