Architecture
A local-first AI roleplay engine. The entire world, characters, narrative, and imagery generate on the fly from a local stack, with no branching paths or pre-written content.
Core Mechanics
- Dynamic Visuals & State: Qwen 2.5 14B drives the narrative; ComfyUI (Stable Diffusion) visualizes it. The scene updates every few turns.
- Visual Mode: Users can pause the text-based roleplay and ask the AI to generate a specific image from a description of the current scene.
- Smart Memory: The system automatically summarizes interaction history into a memory buffer as the session grows, keeping the AI within its context window without losing track of past events or characters.
- Gallery & Persistence: Maintains a gallery of all generated visuals and preserves world state across sessions.