AI Character “Kohane”: Development and User Evaluation of a Dialogue Agent System Expressed in 2D Manga-Anime Style

Abstract

In this paper, we propose a dialogue agent system with a personality presented in a manga-anime style, aiming to facilitate dialogue between humans and manga-anime characters. First, we define the manga-anime style in Japanese Anime, analyze the characteristics of the expression, and design a system that allows a dialogue agent to use Large Language Models (LLMs) to behave like a manga-anime character. The proposed system comprises three main processes: (1) Understanding the user's speech intent using a LLM and dynamically switching character prompts in real-time, thereby reflecting the character's personality in the agent's responses in detail. (2) Determining manga-anime directions such as comic symbols, backgrounds, effects, and composition according to the content of the speech. (3) Generating parameters based on the content and emotion of the speech, which are then linked to the animation of objects that characterize the agent. Using this system, we created an AI character named "Kohane," whose emotional expression changes depending on the conversation with a human. We also conducted comparative experiments with a 3D dialogue agent. As a result, we confirmed that "Kohane," created by the proposed system, gave a stronger impression of being a manga-anime character to participants.

AI Character “Kohane”: Development and User Evaluation of a Dialogue Agent System Expressed in 2D Manga-Anime Style