생성형 AI를 활용한 뷰티 콘텐츠 제작: 문화적 다양성을 중심으로

Abstract

Conventional generative AI models are predominantly trained on Western-centric facial datasets, and often do not accurately represent distinctive features such as the broader cheekbones commonly found among East Asians or the richly pigmented skin textures characteristic of African populations. This study attempted the development and validation of prompt templates for the Stable Diffusion model that systematically incorporate regional aesthetic standards to facilitate the creation of culturally diverse beauty content. Using the proposed approach, a wide range of image samples were collected from ten cultural regions—South Korea, Japan, India, Nigeria, Mexico, Brazil, Saudi Arabia, Thailand, France, and Kenya—and the AI-generated outputs were analyzed using these customized prompts. Experimental results indicated that images produced with the proposed templates showed higher CLIP scores and improved consistency in facial structure and skin tone representation. This research is significant as it demonstrates the practical potential of prompt engineering for creation of global beauty content and serves as a foundational study for the future development of personalized, real-time makeup simulation technologies.