Prompts Optimization Including Kansei Impression in Image Generation AI

Abstract

This study proposes a method for using linguistic information in genotype representation in interactive evolutionary computation (IEC) to generate phenotypes through image generation AI. Conventionally, the IEC method has used physical parameters of evaluation objects as the genotype; however, this only generates limited variation of objects in the phenotype. In this study, we conducted an evaluation experiment with real users and confirmed the effectiveness of the proposed system from the viewpoints of various candidate reflected user Kansei information generation.