manga-eroico¶
Fully automatic manga/comic translation tool powered by RapidOCR + Hunyuan-MT2. Rust + Tauri 2 + React, visual workflow, AGPL-3.0.
| 项目信息 | |
|---|---|
| 仓库 | https://github.com/Maicarons/manga-eroico |
| 主要语言 | Rust |
| Stars | 1 |
| 类型 | 原创项目 |
| 许可证 | AGPL-3.0 |
| 最近更新 | 2026-09-10 |
## manga-eroico
**Fully-automatic manga translation, local-first & visual.**
**漫画全自动翻译工具 · 本地优先 · 全流程可视化**
Rust + React + Tauri · Powered by [RapidOCR](https://github.com/RapidAI/RapidOCR) & [Hy-MT2](https://github.com/Tencent-Hunyuan/Hy-MT2)
[](https://github.com/Maicarons/manga-eroico/blob/HEAD/github/workflows/ci.yml)
[](https://github.com/Maicarons/manga-eroico/releases)
[](https://maicarons.github.io/manga-eroico/)
[](https://github.com/Maicarons/manga-eroico/blob/HEAD/LICENSE)
✨ Features¶
- 🔁 Visual workflow — detect → OCR → inpaint → translate → polish (optional) → render, every node inspectable & re-runnable
- 🗂️ Project-based — every translation job is a
.meproproject: pages, chapters, glossary, artifacts, versioned & rollback-able - 🧠 Local AI — RapidOCR (PP-OCRv5/v6) for detection & OCR, Hy-MT2 (1.8B/7B/30B-A3B) for 33-language translation, all on-device
- 📈 Hardware-adaptive — detects GPU/RAM and picks a lite / standard / pro model tier automatically
- 🌐 OpenAI-compatible polish — optional chapter-level context-aware text polishing via any OpenAI-compatible API (cloud or LM Studio / Ollama)
- 🌏 i18n — UI in 简体中文 / English / 日本語 / 한국어; OCR & translation languages independent of UI language
- 🌙 Dark mode first-class citizen
- 📦 Small installer — app bundles contain no models; models download from ModelScope in-app with resume + checksum
🚀 Quick Start¶
## prerequisites: Rust (1.85+), Node 20+, pnpm 9+
pnpm install
pnpm dev # web preview (mocked pipeline)
pnpm tauri dev # full desktop app
cargo test # Rust unit tests
pnpm test # frontend unit tests
📚 Documentation¶
- 开发计划 / Development Plan
- Docs site (VitePress):
pnpm docs:dev→ published to GitHub Pages via CI - Third-party model licenses
📜 License¶
This project is licensed under AGPL-3.0. Model weights keep their own licenses — see docs/reference/licenses.md.
🌍 Languages¶
- Translate: 38 target languages, any pair (Hy-MT2) — CJK, Latin, Cyrillic, Arabic, Devanagari, Thai, Tamil, Telugu, Tibetan, Mongolian, Uyghur, ...
- OCR read: 14 recognizer script families (PP-OCRv5/v4/v3)
- UI: 中文 / English / 日本語 / 한국어
🖥️ Headless CLI¶
The whole pipeline also runs without the GUI:
cargo run -p me-server -- create MyManga.mepro --from ja --to zh
cargo run -p me-server -- import-folder MyManga.mepro ./my-manga/
cargo run -p me-server -- run-page MyManga.mepro --all --llm-url http://127.0.0.1:8990/v1
cargo run -p me-server -- status MyManga.mepro
See the docs for full guides (zh/en/ja/ko).