Exploring Anime Character Image Generation Based on User Preferences

Abstract: This paper presents a method for generating images of anime characters based on user preferences. A questionnaire is developed to measure user preferences, inspired by Rubin's "Love and Liking Scale". Specifically, user preferences are collected by presenting anime character images to participants through a crowd survey and collecting their responses. The model responsible for generating the anime character images is trained using deep learning techniques, using the survey data as a training set. However, attempts to generate images using the trained model did not produce the expected results. When analyzing the survey data, it was found that there was limited variability in the "Love and Liking" scales for each anime character. This suggests that the trained models may not adequately reflect user preferences. Future work will focus on improving the model to accurately capture user preferences and developing a more appropriate model. This study provides fundamental knowledge and essential insights for the development and advancement of anime character image generation methods tailored to individual user preferences.

Keywords: Anime character image generation, User preferences, Deep Learning

DOI: 10.54941/ahfe1004269

Cite this paper:

​