Showrunner AI - generative AI TV fan and parody episodes

What can do:

The AI startup focuses on using large language models (LLMs), diffusion models, and multi-agent simulations to generate high-quality episodic content for Intellectual Property (IP). They aim to enhance creative storytelling through contextualization, story progression, and behavioral control.


  1. Powerful LLMs: The startup utilizes powerful LLMs like GPT-4 trained on a large corpus of TV show data. This enables them to rewrite entire seasons of TV shows with the right guidance.
  2. Multi-Agent Simulation: By using a multi-agent simulation, the startup incorporates data points such as character history, goals, emotions, simulation events, and localities to generate coherent and aligned scenes and image assets.
  3. Diffusion Models: The startup developed custom diffusion models trained on South Park show data. These models gradually transform random noise into coherent images, allowing for the generation of characters and backgrounds that align with the show's style.
  4. Episode Generation: The startup prompts the story system with high-level ideas and major events to generate full episodes. The showrunner system handles casting characters, scene progression, and dialogue generation based on plot patterns.
  5. Reducing Latency: To reduce latency in generation, the startup hides most of the generation time during user interactions and uses faster models for specific prompts. They also implement a buffering system for voice clips to avoid pauses in scene playback.

Use Cases

  1. Rewriting TV Show Endings: Users can ask the AI to rewrite the endings of TV shows, addressing issues like dissatisfaction with Game of Thrones' last season. The AI can generate alternative endings, allowing users to explore different story directions.
  2. Creating Personalized TV Episodes: Users can put themselves into their favorite TV shows as main characters or create their own personalized episodes. The AI can generate scenes and dialogue based on user prompts, providing an interactive and engaging storytelling experience.
  3. Enhancing Creative Writing: The AI startup's multi-agent simulation and prompt-chaining techniques can assist writers in brainstorming and refining ideas for their stories. By simulating creative thinking, the AI system can generate prompt chains and provide suggestions, improving the writing process.
  4. Generating Coherent South Park Episodes: The startup's use of large language models, diffusion models, and multi-agent simulations allows for the generation of South Park episodes that align with the show's humor, characters, and style. Users can experience new episodes that match the existing IP.
  5. Exploring New Show Concepts: Users can use the AI system to generate episodes for existing TV shows or create entirely new show concepts. By utilizing prompt-chaining and customization features, users can iterate on their ideas and explore new storytelling possibilities.

Overall, the AI startup leverages advanced techniques in language models, simulations, and generative models to provide users with the ability to rewrite TV show endings, personalize episodes, enhance creative writing, generate coherent South Park episodes, and explore new show concepts.

Prompt type:

Generate image, Write video script, Generate video


The authors present an approach using large language models, diffusion models, and multi-agent simulations to generate high-quality episodic content for intellectual property.