A Study on NPC Narrative Generation Using LLM Systems
Hongjun Choi, Jeongwon Han, DongLyeor Lee, Byung Pyo Kyung · Korean Institute of Smart Media · 2025
This study proposes a system for efficiently generating in-game NPC narratives using large language model (LLM) technology. To create natural and consistent NPC narratives, we developed a prompt design methodology consisting of three components: basic prompts, information prompts, and variable prompts. This approach significantly improves development efficiency and NPC diversity compared to traditional manual NPC design methods. In an evaluation conducted with 48 game industry professionals, naturalness scored an average of 4.35 and completeness scored an average of 4.06, both indicating high satisfaction. Additionally, 62.5% of evaluators expressed an intention to adopt the system in actual game development. A strong positive correlation (r=0.813, p=0.05) was observed between the two evaluation metrics. The research results suggest that LLM-based NPC generation technology can enhance immersion in games and increase player engagement. Future research directions include studying the integration of NPC dialogue generation systems with appearance generation.