MITHRIL: A Multi-modal and Immersive Generative AI Approach for Roleplaying Tabletop Games

We present MITHRIL, a multi-modal generative system that automatically produces rich narrative text and accompanying images for tabletop role-playing games. Our system uses a multi-agent architecture with specialized agents for character backstory generation, plot development, image creation, and personalization to ensure coherence across story elements. Notably, MITHRIL can personalize content by embedding a player’s likeness into character avatars and tailoring narrative details, demonstrating a new level of controllable generation in interactive storytelling. In evaluation with experienced Dungeon Masters, the system significantly reduced content creation time and produced stories that users found coherent and engaging, showcasing the potential of generative AI to augment human storytellers.Note: This is an unpublished paper.