Improvement Directions of Image Generation AI to Support Product Ideation

Background Recently, the use of large language models (LLMs)-based generative artificial intelligence (AI) that can create creative content by understanding, combining, and inferring textual information is gradually expanding in various industries. In the field of product design, various measures are actively studied to generate high-quality product ideas using image-generating AI. However, when product designers use image generation AI to generate initial product ideas, there is a lack of research that discovers the deficiencies of image generation AI based on user experience to explore the directions of improvement of image generation AI. Methods First, through literature research, we investigated what role image generation AI can play in the process of designers coming up with product ideas. Second, we observed the product design ideation process of 10 participants using image generation AI and conducted in-depth interviews. Third, by analyzing the task performance records and interview responses, the users’ inconvenient experiences are identified, and the problems of image generation AI are identified. Fourth, we proposed the improvement directions of image generation AI tools to utilize image generation AI more effectively and efficiently in the product design ideation process. Results The experimental observation results are as follows. First, visual product images generated by image generation AI helped spread the inspiration of product designers. Second, it was difficult for the users to select the prompt to enter at the beginning of their ideation. Third, image generation AI generated a lot of product images that did not reflect the user’s intention. Fourth, image generation AI generated structurally incomplete product images. Fifth, when users tried to modify the product image, the shape was not well transformed only by text prompts. Sixth, users were uncomfortable with the process of repeatedly entering text prompts such as form, style, material, and color, which are repeatedly used when coming up with an idea. Conclusions In this study, we propose improvement directions for image generation AI tools that can help to create product design ideas. First, in the process of entering appropriate prompts necessary for idea creation, a method of directly exploring prompts within the tool should be supported so that designers can receive help. Second, there is a need for a method of managing idea generation results, which supports product designers to select valid ideas from numerous images generated by AI according to the design goals initially set. Third, if necessary, the AI should be fine-tuned with the data chosen by the product designer. Fourth, there is a need for a method that can intuitively modify the product image from the three-dimensional shape of the actual product. Fifth, there is a need for a method of supporting a user to customize and use design elements that are repeatedly used.