This study examined how generative artificial intelligence represents cultural identity in fashion images and explored the coexistence of aesthetic differentiation and homogenization. Ninety-six female fashion images were generated using Midjourney across eight conditions, combining four identity categories—White, Black, East Asian, and South Asian—with contemporary and traditional styles. Four coders evaluated the images independently using a structured framework that included clothing type, color, decorative patterns, cultural symbols, facial appearance, hairstyle, body shape, makeup, pose, and composition. Final data were established through inter-coder agreement assessment and consensus coding, frequency analysis, and visual analysis of representative images. The results revealed two main patterns. First, cultural identities were mainly differentiated through clothing silhouettes, decorative patterns, head coverings, jewelry, colors, and culturally identifiable symbols. Traditional conditions included more cultural elements, whereas the White contemporary condition was largely represented through modern clothing, suggesting an association between White identity and default modernity. Second, despite cultural differences, all identity conditions shared a highly homogeneous ideal of beauty characterized by youthfulness, smooth and refined skin, ideal facial features, and predominantly slim bodies. These findings indicate that generative AI creates cultural diversity primarily through external fashion elements while maintaining a standardized aesthetic template for female appearance. Therefore, apparent cultural diversity in AI-generated fashion images may not reflect broader aesthetic diversity but may reproduce existing cultural stereotypes and dominant beauty standards.

Abstract
I. Introduction
II. Literature Review
    1. Aesthetic bias in fashion media
    2. Aesthetic homogeneity
Ⅲ. Research Method
    1. Generation conditions and prompt design
    2. Image generation and data collection
    3. Coding items and analysis procedures
Ⅳ. Results and Discussion
    1. Intercoder agreement and consensus building
    2. Differences in clothing and culturalrepresentation across identity and style conditions
    3. Shared appearance and body representationsacross identity conditions
    4. The coexistence of aesthetic bias andaesthetic homogenization
Ⅴ. Conclusion
References

• Yanong Feng(Master's Student, Dept. of Clothing & Textile, Hanyang University, Korea) | 봉아농 (한양대학교 의류학과 석사과정)
• Se Jin Kim(Associate Professor, Dept. of Clothing & Textile, Hanyang University, Korea) | 김세진 (한양대학교 의류학과 부교수) Corresponding author