Comparing AI-Generated and Human-Crafted T-Shirt Layouts through an XAI Lens

In recent years, the apparel industry has also begun to draw attention to the use of generative AI for design assistance. This study utilized 30 existing T-shirt illustrations provided by graniph and compared: (1) the layouts proposed by the generative AI system “Gemini,” (2) the actual product layouts, and (3) the choices made by five professional designers. First, when evaluated based on eight strictly defined categories (sub-categories), Gemini's layouts matched the actual product in only 12 of the 30 cases. However, by grouping similar placement categories into five broader categories, the number of matches increased to 20. Furthermore, in terms of these broad categories, the matches between Gemini and the five designers ranged from 18 to 24, indicating that Gemini's proposals exhibit a certain level of commonality with human decisions. In addition, when Gemini explained in text why it chose a particular layout, over half of the illustrators found these explanations “reasonable,” suggesting that explainable AI can effectively support collaborative design. This study clarifies both the practical and academic significance of using generative AI in the fashion design field and highlights the importance of interaction design and explanation methods to facilitate smoother collaboration between humans and AI.