Researchers at Pusan National University, led by Professor Yoon Kyung Lee, investigated the application of generative AI – specifically ChatGPT-3.5, ChatGPT-4, and DALL-E 3 – to forecast menswear trends for the fall/winter 2024 season. Analysing historical data up to September 2021, the team generated 105 images with DALL-E 3, achieving successful prompt implementation 67.6% of the time. While some generated designs mirrored existing 2024 collections, the research highlighted the necessity of detailed prompts, as the AI struggled with nuanced concepts like gender fluidity when relying solely on trend keywords.
AI’s Emerging Role in Fashion Design
Researchers at Pusan National University explored the application of generative AI to visualise seasonal fashion trends, specifically analysing menswear trends up to September 2021 to predict those for the fall/winter 2024 season using ChatGPT-3.5 and ChatGPT-4. The research classified design elements into initial codes, subsequently refined and grouped into six final codes encompassing trends, silhouette elements, materials, key items, garment details, and embellishments. This systematic approach aimed to provide a structured framework for translating data into actionable insights for AI-driven design generation.
The researchers generated 35 prompts for DALL-E 3, each detailing a unique outfit, and consistently specified a template featuring a male model on a 2024 Fall/Winter runway, allowing for customisation of various parameters including aspect ratios and background. A total of 105 images were produced by running each prompt three times, and DALL-E 3 successfully implemented the prompts 67.6% of the time, with prompts including adjectives demonstrating a particularly high implementation rate.
While some generated images closely resembled actual 2024 Fall/Winter menswear collections, errors were noted, with a tendency towards ready-to-wear fashion and difficulties incorporating elements of gender fluidity. The study found that using trend keywords alone was insufficient for accurate image generation, indicating a requirement for more detailed and nuanced input to achieve desired results in fashion trend prediction.
Professor Yoon Kyung Lee noted that expertly worded prompts are necessary for accurate fashion design implementation using generative AI, highlighting the continued importance of fashion expertise in the process. The research suggests that generative AI models like DALL-E 3 have the potential to enhance both the efficiency and creativity of fashion designers, as well as enable broader understanding of fashion trends amongst non-experts.
Predictive Analysis and Methodology
The research utilised ChatGPT-3.5 and ChatGPT-4 to predict menswear trends for the fall/winter 2024 season, based on an analysis of historical data up to September 2021. Design elements were initially classified as codes, and subsequently regrouped into six categories: trends, silhouette elements, materials, key items, garment details, and embellishments. This coding system provided a framework for translating design data into prompts suitable for generative AI image creation.
Each of the 35 prompts generated for DALL-E 3 described a unique outfit, and a consistent template was specified, depicting a male model walking on a 2024 Fall/Winter runway. This template allowed for customisation of parameters including aspect ratios, event settings, camera angles, model appearance, runway design, background, and mood. Running each prompt three times resulted in a total of 105 generated images.
The study demonstrated that DALL-E 3 successfully implemented 67.6% of the prompts, with those containing adjectives exhibiting a particularly high implementation rate. While some generated images demonstrated a close resemblance to actual 2024 Fall/Winter menswear collections, the research also identified errors, including a tendency towards ready-to-wear fashion and difficulties in incorporating trend elements such as gender fluidity. The findings suggest that isolated trend keywords are insufficient for accurate image generation, implying a need for more nuanced input for effective fashion trend prediction.
Professor Yoon Kyung Lee noted the necessity of expertly worded prompts for accurate fashion design implementation with generative AI, thereby underlining the continuing importance of fashion expertise. The research, published in the Clothing and Textiles Research Journal (DOI: 10.1177/0887302X251348003), indicates that generative AI models like DALL-E 3 can potentially enhance the efficiency and creativity of fashion designers and facilitate a broader understanding of fashion trends.
Implementation and Future Potential
With further learning and improvements, generative AI models like DALL-E 3 promise to enhance fashion designers’ efficiency and creativity, while also enabling non-experts to understand fashion trends. This demonstrates that generative AI can be a powerful tool for both professionals and the general public, facilitating exploration, prediction, and styling of upcoming seasons with greater confidence.
The research, published in the Clothing and Textiles Research Journal (DOI: 10.1177/0887302X251348003), indicates that generative AI models like DALL-E 3 can potentially enhance the efficiency and creativity of fashion designers and facilitate a broader understanding of fashion trends.
Professor Yoon Kyung Lee is an Assistant Professor at Pusan National University, specialising in creativity and sustainability in fashion design, with a focus on AI, digital technology, and neuroscience. She holds an MSA and Ph.D. in dress aesthetics from Seoul National University, and previously ran her own brand, UginiO, showcasing at Seoul Fashion Week and Prt–Porter Paris. Her lab website is available at https://fashiondesign.pusan.ac.kr/fashiondesign/index.do, and her ORCID id is 0000-0002-5118-3789.
More information
External Link: Click Here For More
