This work investigated the potential of utilizing large language models (LLMs) to enhance narrative quality and player engagement in visual novel-style games. A system for generating dialogues based on LLMs was used in a prototype game. Player studies were conducted to evaluate narrative immersion, engagement, and overall game satisfaction in both the LLM-generated dialogue version and a version with predefined dialogues. The results showed that participants experienced greater satisfaction and interest in interacting with the LLM-operated NPCs. This confirmed the hypothesis that leveraging LLMs can effectively improve narrative quality in games by providing more individualized and responsive experiences for players. The work also discusses potential further research directions and the development of LLM-based systems in the context of video games.
Thesis available here: https://github.com/KyattPL/masters-backup/blob/master/LaTeX%20praca/thesis.pdf