Hutama Husen, Dharma
Unknown Affiliation

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Procedural Story Generation for Visual Novels Using Large Language Models and Text-to-Image Techniques Hutama Husen, Dharma; Istiono, Wirawan
Journal of Games, Game Art, and Gamification Vol. 10 No. 3 (2025)
Publisher : Bina Nusantara University

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.21512/jggag.v10i3.12743

Abstract

Story-based video games have content limitations. A survey conducted among Genshin Impact players, 67% of respondents had reached the end of the avail- able story. The game’s ending may lead to repetitive gameplay. To increase the story content inside video games, this research uses Generative AI for pro- ducing procedural stories. This research also measures the similarity of generated stories using a sentence similarity model. Text contents are generated using an LLM named mistralai/Mixtral-8x7B-Instruct-v0.1. Picture contents are generated using a Text-To-Image model named stabilityai/stable-diffusion- xl-base-1.0. Story similarities are measured using a sentence similarity model named sentence-transformers/all-MiniLM-L6-v2 with a similarity threshold of 73%. The visual novel application is implemented in Unity Engine 2022.3.26f1 version. The connection between Unity and the models are established through REST API provided by Hugging Face website. The result of this research is an implementation of a Visual Novel game with Generative AI-made content, as well as measuring the uniqueness level of 36,667% for low usage and up to 60% for high usage. The reduction of uniqueness level is attributed to the appearance of common genres, as well as the increasing number of comparisons required. This research shows how to implement Generative AI in a visual novel game, and shows that a Generative AI’s ability to create unique stories will decrease with an increase in game repetitions.