From わからない (I don't understand) to わかった (I understand)!

🇫🇷 Version Française | 🇯🇵 日本語版 (coming soon)

Real-time universal translator for video games using OCR/Vision and local AI (Ollama).

Instantly translate untranslated game texts by capturing any screen area.

📋 Release validation: See the phase-by-phase validation plan for startup time, OCR latency, CI, and platform checks.

• Three translation modes: Tesseract (fast), EasyOCR (accurate), or Vision (context-aware)
• Automatic language detection: Detects source language automatically (Japanese, Korean, Chinese, etc.)
• Customizable hotkeys: Prevent conflicts with games using modifier combinations
• Global hotkey to capture and translate on the fly
• Mode switching hotkey to cycle between translation modes in real-time
• Visual area selection for precise text capture
• 100% local translation via Ollama (privacy-friendly)
• Elegant transparent overlay with auto-close (configurable timeout)
• Configuration GUI for easy setup

• Pipeline: Screenshot → Tesseract OCR → LLM translation
• Speed: ⚡ Very fast (2-5 seconds)
• Accuracy: ✅ Good for clear text
• GPU Usage: Minimal (great for gaming)
• Best for: Active gameplay, simple text, European languages

• Pipeline: Screenshot → EasyOCR → LLM translation
• Speed: ⚡⚡ Fast (5-10 seconds)
• Accuracy: ✅✅ Excellent for Asian languages
• GPU Usage: Medium (with CUDA) or CPU
• Best for: Japanese, Korean, Chinese, 80+ languages
• Note: Requires Python 3.11 or 3.12

• Pipeline: Screenshot → Vision Model → Direct translation
• Speed: 🐌 Slow (10-30 seconds)
• Accuracy: ✅✅✅ Excellent, understands visual context
• GPU Usage: High (may cause lag during gameplay)
• Best for: Screenshots, stylized text, paused games

In-game: Press Ctrl+Shift+M to cycle through modes!

Tesseract → Ctrl+Shift+M → EasyOCR → Ctrl+Shift+M → Vision → Ctrl+Shift+M → Tesseract...

1. Download the project (ZIP or git clone)
2. Double-click INSTALL.bat
3. Wait for installation to complete
4. Done! 🎉

1. Python 3.11 or 3.12 (Download here)
   • ⚠️ IMPORTANT: Check "Add Python to PATH" during installation!
2. Ollama (Download here)
3. Tesseract OCR (Download here)
   • Add to PATH: C:\Program Files\Tesseract-OCR
4. Git (optional, for updates) (Download here)

# 1. Create virtual environment
python -m venv venv311

# 2. Activate environment (Windows)
venv311\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

1. Install Ollama: https://ollama.ai
2. Download models:

   # For OCR modes (fast, recommended)
   ollama pull gemma2:2b
   
   # For Vision mode (accurate, optional)
   ollama pull gemma3:4b

3. Start Ollama server:

   ollama serve

Note about EasyOCR: EasyOCR is optional but recommended for:

• 🇯🇵 Japanese games (kanji, hiragana, katakana)
• 🇨🇳 Chinese games (simplified and traditional)
• 🇰🇷 Korean games (hangul)
• 🎨 Exotic/stylized fonts

To enable EasyOCR: Uncomment lines in requirements.txt and re-run INSTALL.bat

Double-click CONFIGURE.bat to open the configuration interface and customize:

• 🎯 Keyboard shortcuts (customizable hotkeys)
• 🌍 Source and target languages
• ⚙️ Default translation mode
• ⏱️ Overlay display duration
• 🤖 Ollama models

You can also directly edit config.json:

{
  "translation_mode": "tesseract",
  "vision_model": "gemma3:4b",
  "ollama_model": "gemma2:2b",
  "ollama_url": "http://localhost:11434",
  "source_lang": "en",
  "target_lang": "fr",
  "ocr_languages": ["ja", "en"],
  "auto_detect_language": true,
  "hotkey": "ctrl+shift+t",
  "toggle_mode_hotkey": "ctrl+shift+m",
  "overlay_timeout": 60
}

• overlay_timeout: Overlay display duration in seconds

  • Default: 60 (1 minute)
  • Fast readers: 30 seconds
  • Slow readers: 90 or 120 seconds

• Supported language codes: en, fr, es, de, it, pt, ja, ko, zh

Double-click RUN.bat to start WakaranAI.

# Start Ollama (in separate terminal)
ollama serve

# Launch WakaranAI
venv311\Scripts\python main.py

• Ctrl+Shift+T: Translate screen area
• Ctrl+Shift+M: Switch mode (Tesseract ↔ EasyOCR ↔ Vision)
• Escape: Cancel selection
• Ctrl+C: Quit application (in console)

💡 Tip: Use combinations with modifiers (Ctrl, Shift, Alt) to avoid conflicts with games!

1. Launch your game
2. Press Ctrl+Shift+T to capture
3. Draw a rectangle around the text to translate
4. Wait for translation (2-5s in Tesseract, 5-10s in EasyOCR, 10-30s in Vision)
5. Read the translation in the overlay (stays 60 seconds by default)
6. If too slow, press Ctrl+Shift+M to switch to fast Tesseract mode!

Double-click UPDATE.bat to automatically update code and dependencies.

1. Download latest version from GitHub
2. Replace files (except config.json)
3. Run UPDATE.bat to update dependencies

• Check if your shortcut is used by the game
• Try a combination with Ctrl+Shift+...
• Launch CONFIGURE.bat to change shortcuts

• Switch game to windowed mode (windowed or borderless)
• Use Escape to cancel selection
• This is a limitation with fullscreen DirectX/OpenGL games

• Check Ollama is running: ollama serve
• Test URL: open http://localhost:11434 in browser
• Check port 11434 isn't blocked by firewall

• Check Tesseract is installed: tesseract --version
• On Windows, add to PATH or re-run INSTALL.bat

• Selected area may be too small or blurry
• Try EasyOCR mode: Ctrl+Shift+M to switch modes
• Text contrast may be too low

• Re-run INSTALL.bat
• Or manually: venv311\Scripts\pip install -r requirements.txt

• Use Tesseract mode for gaming
• Check CPU/GPU (Vision mode is very intensive)
• Reduce selected area size

WakaranAI/
├── INSTALL.bat            # 🔧 Automatic installation
├── RUN.bat                # ▶️ Quick launcher
├── CONFIGURE.bat          # ⚙️ Configuration interface
├── UPDATE.bat             # 🔄 Automatic update
├── main.py                # Main entry point
├── config_gui.py          # Configuration interface
├── config.json            # Configuration (created on first run)
└── venv311/               # Virtual environment (created by INSTALL.bat)

• Linux/Mac support
• Automatic text zone detection
• Translation history
• Multi-monitor support
• Full GUI (without console)
• Customizable overlay themes
• More LLM model support

Contributions are welcome! Feel free to:

• Open an issue to report a bug
• Suggest a feature
• Submit a pull request

MIT License - This project is under MIT license. Feel free to adapt it to your needs!

• Tesseract OCR
• EasyOCR
• Ollama
• Gemma (Google)

Developed with ❤️ for the gaming community

If you like this project, don't forget to give it a ⭐ on GitHub!