Generative AI for Interactive Storytelling

Faculty: Edirlei Soares de Lima, Margot Neggers

Generative AI models have shown remarkable capabilities in creating diverse content, including text, image, and voice. However, Large Language Models often struggle to maintain thematic consistency and structure when used for narrative generation, especially when creating longer and more complex stories. When generative AI aims at producing high-quality narratives, having mechanisms to guide the creative process and ensure thematic consistency is crucial. This project explores new methods to guide the generation of narratives using AI, by incorporating narrative patterns, genre structures, and semiotic relations, to improve the overall quality and coherence of the generated stories.

Edirlei and Margot, introduced a new approach to interactive narrative generation based on semiotic relations and large language models, enabling the creation of new narratives from existing ones through combination, imitation, expansion, and reversal. Moreover, they proposed the use of narrative patterns, derived from folklore and literary traditions, as well as genre structures, to guide the generation of thematically consistent and coherent stories using large language models.

Holmes & Hobbit

Holmes & Hobbit: The Intersection of Worlds generated by ChatGeppetto

Test it yourself!

Publications