Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GENEVA: GENErating and Visualizing branching narratives using LLMs

Jorge Leandro, Sudha Rao, Michael Xu, Weijia Xu, Nebosja Jojic, Chris Brockett, Bill Dolan

TL;DR

GENEVA addresses the challenge of crafting branching narratives for dialogue-based RPGs by leveraging GPT-4 to generate a narrative graph aligned with high-level designer constraints. The method uses a two-step prompt process: first to produce branching beats, then to encode them into a graph for visualization (D3JS). A case study across four classic tales grounded in multiple settings demonstrates that the approach can produce plausible, testable narrative graphs, though grounding quality varies by setting. The work suggests a practical, AI-assisted workflow to accelerate early-stage narrative design and iteration.

Abstract

Dialogue-based Role Playing Games (RPGs) require powerful storytelling. The narratives of these may take years to write and typically involve a large creative team. In this work, we demonstrate the potential of large generative text models to assist this process. \textbf{GENEVA}, a prototype tool, generates a rich narrative graph with branching and reconverging storylines that match a high-level narrative description and constraints provided by the designer. A large language model (LLM), GPT-4, is used to generate the branching narrative and to render it in a graph format in a two-step process. We illustrate the use of GENEVA in generating new branching narratives for four well-known stories under different contextual constraints. This tool has the potential to assist in game development, simulations, and other applications with game-like properties.

GENEVA: GENErating and Visualizing branching narratives using LLMs

TL;DR

GENEVA addresses the challenge of crafting branching narratives for dialogue-based RPGs by leveraging GPT-4 to generate a narrative graph aligned with high-level designer constraints. The method uses a two-step prompt process: first to produce branching beats, then to encode them into a graph for visualization (D3JS). A case study across four classic tales grounded in multiple settings demonstrates that the approach can produce plausible, testable narrative graphs, though grounding quality varies by setting. The work suggests a practical, AI-assisted workflow to accelerate early-stage narrative design and iteration.

Abstract

Dialogue-based Role Playing Games (RPGs) require powerful storytelling. The narratives of these may take years to write and typically involve a large creative team. In this work, we demonstrate the potential of large generative text models to assist this process. \textbf{GENEVA}, a prototype tool, generates a rich narrative graph with branching and reconverging storylines that match a high-level narrative description and constraints provided by the designer. A large language model (LLM), GPT-4, is used to generate the branching narrative and to render it in a graph format in a two-step process. We illustrate the use of GENEVA in generating new branching narratives for four well-known stories under different contextual constraints. This tool has the potential to assist in game development, simulations, and other applications with game-like properties.
Paper Structure (11 sections, 2 figures)

This paper contains 11 sections, 2 figures.

Table of Contents

  1. Introduction
  2. METHODOLOGY
  3. Generating the Story Lines
  4. Generating the Visualization
  5. Case Study
  6. Stories, settings and constraints
  7. Analysis of the generate narrative graphs
  8. Conclusion
  9. Appendix
  10. Prompt for generating the story lines
  11. Prompt for generating graph

Figures (2)

  • Figure 1: In the online interface of GENEVA, you can choose a particular story, pick the number of starts, ends and storylines and click on the 'Show Graph' button to see the narrative graph for that story. The above figure shows the narrative graph for the Frankenstein story but grounded in the 21st century. Additional constraints on the graph includes one start, two endings and four storylines.
  • Figure 2: In the online interface of GENEVA, you can see the detailed description of the sequence of beats that make up each storyline by clicking on the 'Show Details' button. The above figure shows detailed view of the four different storylines present in the narrative graph in Figure 1.