Django Manga Translator

A powerful, Django-based web application for automated manga translation. This is a forked and enhanced version of georgescutelnicu/Manga-Translator.

Unlike the original repository, which primarily handles local files, this fork introduces the ability to translate directly from any manga website URL using its integrated scraping engine.

🌟 Key Features

• Automated Detection: Uses YOLOv8 to precisely locate text bubbles in manga pages.
• State-of-the-Art OCR: Leverages manga-ocr for high-accuracy Japanese text recognition.
• Stable Translation: Optimized support for Google Translate and Microsoft Bing Translator.
• Modern UI/UX: Custom-built, high-performance vanilla CSS/JS interface with animated transitions and a simulated progress tracking system.
• Smart Scraping: Integrated cloudscraper to bypass Cloudflare and anti-bot protections on popular manga sites.
• Image Enhancement: Integrated 2x Lanczos upscaling (Super-Resolution) to improve OCR accuracy for small text and furigana.
• Professional Logging: Structured logging with loguru, featuring file rotation and retention.
• Dynamic Cleanup: Automatically manages temporary files and translated images to save storage.

🚀 Improvements in this Fork

• URL-Based Translation: Translates directly from manga chapter URLs (Scraping engine).
• Premium UI Overhaul: Replaced standard Bootstrap with a bespoke, mobile-friendly modern design featuring deep indigo accents.
• Fault-Tolerant Pipeline: Implemented resilient error handling that prevents application crashes during individual translation failures.
• Robust Scraper: Added fallback to cloudscraper for sites like RawKuma.
• OCR Optimization: Implemented image upscaling logic that significantly improves recognition of small characters.
• Furigana Handling: Optimized preprocessing specifically to capture small reading aids (furigana).
• Production Logging: Integrated a centralized logging system with file outputs for easier debugging.
• Chained Translation Fallback: Automatically switches between translation services (Google/Bing) if one fails or is rate-limited.
• Improved OCR Stability: Specialized regex-based filtering to ignore punctuation-only bubbles, preventing API hangs on common "..." sequences.
• Robust Text Rendering: Implemented failsafe logic in the font-sizing engine to prevent infinite loops in extremely small speech bubbles.
• Maintenance Logic: Automated cleanup of media files to prevent storage bloat.

🛠️ Installation

Prerequisites

• Python 3.10+
• PyTorch (for YOLO and Manga-OCR)
• CUDA-compatible GPU (optional, but recommended for speed)

Setup

1. Clone the repository:

   git clone https://github.com/YOUR_USERNAME/Django-Manga-Translator.git
   cd Django-Manga-Translator

2. Create a virtual environment:

   python -m venv env
   source env/Scripts/activate  # On Windows: env\Scripts\activate

3. Install dependencies:

   pip install -r mangaTranslator/requirements.txt

4. Model Setup:

   • Place your YOLOv8 bubble detection model (model.pt) in the mangaTranslator/model/ directory.

5. Environment Setup:

   • Copy the template file: cp .env.template .env
   • Open .env and add your DJANGO_SECRET_KEY.
   • (Optional) Customize SERVER_PORT or DJANGO_DEBUG.

6. Run Migrations:

   cd mangaTranslator
   python manage.py migrate

7. Start the server:

   python manage.py runserver

   Note: The server will log its start-up port and status to the daily log file.

📖 Usage

1. Open your browser and navigate to http://127.0.0.1:8000/.
2. Enter the URL of the manga chapter you want to translate.
3. Select your preferred font and translator.
4. Click Translate and wait for the magic to happen!

📂 Project Structure

• mangaTranslator/: Main Django project directory.
  • translate/: Main application logic (views, forms, scrapers).
  • utils/: Helper modules for OCR, detection, and translation.
  • media/: Storage for processed images.
  • logs/: Application logs (created automatically).

⚠️ Current Limitations & Future Roadmap

While this fork introduces major reliability and UI improvements, there are several known limitations where the engine can still be improved:

• Bubble Detection Bias: The current YOLO model is primarily optimized for standard white bubbles.
• Common "Misses":
  • Black/Dark Bubbles: Text in inverted bubbles may be missed or poorly OCR'ed.
  • Stylized/Textured Bubbles: Bubbles with complex patterns or rough edges are harder to detect.
  • Borderless Dialogue: Floating text without a clearly defined border is often skipped.
  • Narration Boxes: Rectangular narration panels aren't always recognized as bubbles.
• Pipeline Constraints:
  • No Fallback for Non-Bubble Text: The system does not currently attempt to find "free-floating" text outside of detected bubbles.
  • OCR Dependency: OCR only runs on successfully detected areas; if the YOLO model misses a bubble, the text within it is missed entirely.

🗺️ Future Improvements

• Support for black/dark bubble inversion and Stylized/Textured Bubbles before OCR.
• Integration of a secondary text-only detector (like DBNet) for borderless text.
• Support for Manhwa (Korean) and Manhua (Chinese) layouts.
• Moving to a background task queue (Celery) for multi-page chapters.

🙏 Acknowledgments

• Original project by georgescutelnicu.
• Manga-OCR by kha-white.
• Ultralytics for the YOLOv8 implementation.