For anyone using GenAI who’s still struggling with image generation or finding prompts overwhelming, I’ve made major modifications to the code that have produced excellent results.
Key improvements include:
- An intelligent caching system where images are stored based on character state, so identical scenes reuse existing images instead of regenerating them.
- The prompt generation is now enhanced with a robust clothing detection and classification system that correctly handles all clothing types.
- In terms of performance, this has cut generation time by around 80% for repeated scenes.
- The cache persists across game sessions
From a technical standpoint, it now includes comprehensive clothing region mapping (torso_full, torso_upper, etc.), a visibility aware weighting system, deduplication, and error handling with fallbacks.
View attachment 5131973
View attachment 5131981
View attachment 5131984
View attachment 5131986
View attachment 5131990
The system is far from perfect so use at your own risk. I just wanted to share this with the community in case anyone finds it useful.