Method of Generate Character's Appearance by User's Love Tendency

Abstract

In recent years, CG characters have been used in digital signage and other applications. The creation of such characters requires specialized knowledge, equipment, and time. Therefore, when user-supported characters become popular in the future, it will be difficult for users to create their favorite characters. Therefore, we investigate users' romantic tendencies and preferred characters, and study a method to create a character's appearance suitable for each user's romantic tendencies. Specifically, we collect data on users' romantic tendencies toward characters using Rubin's love and like scales. Using this data as training data, we train a generative model of the character using cDCGAN. The vector of love tendencies is compared by using the number of items collected as it is, or by converting it into a one-hot vector. The results show that there is no significant difference between the two methods. We also analyzed the collected data and found that there was not much difference in the users' love tendencies.