Dependency-Driven Prompt Pipeline for Coherent RPG Generation

A recent study published on arXiv (2604.25482) introduces a multi-stage prompt pipeline that is sensitive to dependencies for generating content in procedural RPGs with Large Language Models. This method breaks down the generation process into distinct phases: world building, creation of non-player characters, development of player characters, planning of campaign-level quests, and expansion of quests. Each phase relies on structured JSON outputs from earlier stages, ensuring adherence to schemas and a clear data flow, which minimizes narrative inconsistencies, reduces hallucinations, and facilitates the scalable creation of interconnected narrative components. The system's effectiveness is assessed qualitatively through a human-centered analysis across various instances.