Multiview of AI-generated Facial Expressions for Entertainment: A Cross-style Analysis

The objective of this study is to investigate how artificial intelligence (AI)-generated characters’ expressions regarding emotions reflect the aspects of emotion in different styles and how this is perceived by an audience. Using the Matrix of Character Styles and Types (MCST) framework, we developed a dataset with multidimensional characters. On the vertical plane, it is characters from humans to animals and fantasy figures on the horizontal, it is the level of art from hyperrealism to abstract. Based on the stable diffusion model, characters in diverse styles exhibited six universal basic emotional expressions: anger, disgust, fear, happiness, sadness, and surprise. The audience’s perceptions were then analyzed by a survey on several dimensions, including the expressiveness of the emotions generated, the readability in different AI-generated styles for the target audience groups, and, in general, the acceptance of machine-made imagery. This study contributes to the field of cross-style and cross-character-type AI emotion generation and provides specific guidance on how to generate AI-based 3D content (still image and animation). This study aims to provide designers with a more profound understanding of audience preferences and target groups