Fable Studio

Showrunner Agents is a research system from Fable Studio that generates long-form episodic content for existing intellectual properties (IPs) using large language models, custom diffusion models, and multi-agent simulations. Unlike typical AI tools that produce short clips or single images from a prompt, Showrunner Agents aims to create entire seasons of TV-like episodes with consistent characters, plot progression, and behavioral control. The system lets users rewrite storylines, change endings, or insert themselves as characters — think of it as a showrunner you can direct.

The core innovation is a multi-agent simulation that tracks each character's history, goals, emotions, and relationships. This context feeds into the generation process, so scenes and visuals stay coherent with the IP's world. The system also addresses what the team calls the "slot machine effect" — the frustration of random AI outputs — by hiding the trial-and-error generation steps from the user. You only see the final episode, which makes the output feel more deliberate and satisfying. The paper mentions using GPT-4 as the underlying LLM and custom diffusion models for visuals.

This tool is best suited for storytellers, TV writers, and fans who want to explore alternative narratives for existing shows or books. It could also interest game developers building narrative-driven experiences. The system is still in research phase — there's no public demo or pricing yet — but the concept points toward a future where audiences can actively shape the stories they love.