CoCoStyle: Reflecting Subjectivity of Style to Support Decision Making Process for Fashion Designers

Abstract

By harnessing the objectivity of AI, designers have gained support in the decision-making process. However, in fashion, where style is heavily influenced by designers’ subjective experience, expertise, and heuristics, designers have expressed dissatisfaction with the inability of AI to reflect this subjectivity. Our research aims to explore methods for incorporating the nuances of fashion style into AI, and to develop a tool that more effectively supports the design decision-making process. Based on a formative study with six fashion professionals, we identified the process of defining style and how it is applied in the design process, and formulated three design goals. We developed CoCoStyle, consisting of CoStyle, which incorporates style subjectivity into the AI by adapting fashion attributes and images; CoDesign, which generates new, synthetic images based on the user-defined style; and CoImprove, which recommends various ways to improve designs within the user-defined style. We conducted a user study with six fashion professionals and four students majoring in fashion design, confirming the effectiveness of CoCoStyle in capturing style subjectivity, its potential to reduce time in the design process, and its role in mitigating concerns about design uniformity caused by AI. We highlighted the importance of attribute- and image-based adjustments in incorporating style subjectivity into AI and suggested ways for AI to gain a deeper understanding of style nuances.