Manga-Image-Translator is an open-source Python pipeline for manga and comics. You pick a detector, an OCR engine, an inpainter, and a translator, then run the same job from a web UI, the CLI, an API, or a WebSocket. The license is GPL-3.0.
Definition
Manga-Image-Translator is the most comprehensive open-source pipeline for translating manga/comics. It supports one-click translation for many kinds of images, with many different detectors, OCR engines, inpainters, and translators. It is the foundational project that many other tools (cotrans, oomol-flows) build on.
Key Metrics
- Stars: 9,700
- Forks: 969
- Language: Python
- Docker image: ~15GB
Architecture - Modular Pipeline
┌─────────────────────────────────────────────────────────┐
│ manga-image-translator │
├──────────┬──────────┬──────────┬────────────────────────┤
│ Detector │ OCR │Inpainter │ Translator │
├──────────┼──────────┼──────────┼────────────────────────┤
│ CTD │ 48px │LaMa_large│ Sugoi / DeepL / ChatGPT│
│ CRAFT │ 48px_ctc │ LaMa_MPE │ Sakura / DeepSeek │
│ Paddle │ 32px │ SD/SDXL │ Groq / NLLB / Qwen2 │
│ DBConvNext│MangaOCR │ original │ Papago / Youdao / Baidu│
└──────────┴──────────┴──────────┴────────────────────────┘
│ Web UI / CLI / API / WebSocket │
└──────────────────────────────────────────────────────────┘
Detectors
| Detector |
Description |
| CTD |
Comic Text Detector (default) |
| CRAFT |
Character-aware text detection |
| Paddle |
PaddleOCR-based detection |
| DBConvNext |
DBConvNext-based detection |
OCR Engines
| OCR |
Description |
| 48px |
Default OCR model |
| 48px_ctc |
CTC variant |
| 32px |
Smaller model |
| MangaOCR (mocr) |
Specialized manga OCR by kha-white |
Inpainters
| Inpainter |
Description |
| LaMa_large |
Large resolution inpainting |
| LaMa_MPE |
Mask-aware inpainting |
| Stable Diffusion |
AI-based inpainting |
| original |
Basic inpainting |
Translators
| Translator |
Type |
| Sugoi |
Offline (Japanese-focused) |
| DeepL |
API (high quality) |
| ChatGPT / OpenAI |
API (LLM-based) |
| Sakura |
Offline (Japanese) |
| DeepSeek |
API (LLM-based) |
| Groq |
API (fast inference) |
| custom_openai |
API (compatible) |
| NLLB / M2M100 / mbart50 |
Offline (Meta) |
| Qwen2 |
Offline/Online |
| Papago |
API (Korean-focused) |
| Youdao / Baidu / Caiyun |
API (Chinese-focused) |
Installation
pip (venv recommended)
python -m venv venv
source venv/bin/activate # Linux/macOS
venv\Scripts\activate # Windows
pip install -r requirements.txt
Docker
docker pull zyddnys/manga-image-translator
Usage
Web UI
python -m manga_translator --mode webui
# Open http://localhost:5003
CLI
# Single image
python -m manga_translator --input input.jpg --output output.jpg
# Batch
python -m manga_translator --input ./manga_pages/ --output ./translated/
# With options
python -m manga_translator -i input.jpg -o output.jpg \
--detector ctd \
--ocr 48px \
--translator deepL \
--inpainter lama_large \
--target-lang ENG
WebSocket mode
python -m manga_translator --mode ws
Additional Features
| Feature |
Description |
| Upscaler |
ESRGAN, waifu2x |
| Colorizer |
mc2 (manga colorization) |
| API mode |
RESTful API |
| Batch processing |
CLI batch mode |
| Multi-language |
JA, ZH, EN, KO + 20 minor langs |
Tech Stack
| Component |
Technology |
| Language |
Python |
| Framework |
FastAPI |
| ML |
PyTorch |
| OCR |
PaddleOCR, MangaOCR |
| Inpainting |
LaMa, Stable Diffusion |
| Upscaling |
ESRGAN, waifu2x |
Pros
| Pro |
Description |
| Most comprehensive |
Widest variety of detectors/OCR/translators/inpainters |
| Modular |
Choose each component of the pipeline |
| Active community |
9,700 stars, many contributors |
| Multiple interfaces |
Web UI, CLI, API, WebSocket |
| Docker support |
Easy to deploy |
| Foundation for others |
cotrans and oomol-flows are built on this project |
Cons
| Con |
Description |
| Heavy dependencies |
Docker image ~15GB |
| GPL-3.0 |
Copyleft license |
| GPU recommended |
Needs a GPU for good performance |
| Complex setup |
Many models need to be downloaded |
When to Use
- Best overall choice: when you need the most comprehensive manga translation tool
- Custom pipeline: when you want to pick each component (detector, OCR, translator)
- Docker deploy: when you need to deploy it as a service
- Build upon: when you want to build your own tool on an existing foundation
- Batch processing: when you need to process many images
References:
See also: Cotrans · Comic Translate · Translator Comparison