Install Extension
Documentation/Troubleshooting/Common Issues

Common Issues

Solutions to the most frequently encountered problems.

Translation Not Starting

  • No manga images detected — Images must be at least 150x150px with a height/width ratio above 0.4. Very small thumbnails or banner images are ignored.
  • Backend not configured — Make sure you have entered an API key (OpenAI) or started the server (Ollama/FastAPI) and clicked "Check Connection".
  • API key invalid — Verify your OpenAI API key is correct and has available credits. Check in the Backend tab.

Translation Quality Issues

  • Inaccurate translations — Try switching to a more capable model (e.g., GPT-4o instead of GPT-5.4 Nano).
  • Inconsistent names — Add character names to the glossary and enable multi-page context.
  • Text not fitting bubbles — Adjust the font scale down (e.g., to 80%) in the Display tab.

Bubbles Not Detected

  • No bubbles found — Some manga styles (manhwa with colored speech boxes, hand-lettered text) may not be detected well. The white-region fallback handles most cases.
  • Too many or too few bubbles — Adjust merge distance and strict bubble separation in Display tab Advanced settings.

Images Not Loading

If images fail to load for translation, the extension uses a 5-step fallback chain:

  1. Canvas drawImage from the loaded img element
  2. Cross-origin Image load with crossOrigin="anonymous"
  3. Background script fetchImageAsBase64 (bypasses CORS)
  4. Direct fetch from content script
  5. Screenshot visible tab and crop to image area

See CORS & Images for details.

Extension Running Slowly

  • Large images — Very high-resolution pages take longer to process. The PaddleOCR model resizes to 1280px max internally.
  • Too many bubbles — Pages with 20+ speech bubbles take longer. Each bubble is a separate crop sent to the AI.
  • Ollama on CPU — CPU-only inference is significantly slower than GPU. Consider using a smaller model or OpenAI Vision instead.
← Advanced CORS & Images →