画像生成AIを用いた意匠設計法の検討

抄録

Recently, deep learning-based image generation AI such as Stable Diffusion, Midjourney, and DALL-E have been released and have attracted much attention. These methods are based on a latent diffusion model, a type of diffusion model, and can generate images that match the textual representation, called a prompt, input by the user. We believe that the use of image-generating AI in aesthetic design is promising because it can handle textual representations as prompts and can fine-tune the generated image by changing the textual representation of the prompt or by adding textual representation to the prompt. Therefore, as the first step in our research, we verified whether an image generation AI can generate product images that provide the impressions represented by Kansei words from the prompts that include those Kansei words, which is the most fundamental function of aesthetic design methods.