Combinediff: a GenAI creative support tool for image combination exploration
Zhouyang Wang, Zhengyu Tan, Yujia Ma
Abstract
Recent advancements in generative AI models showcase significant potential in the creative industry. This study explores the application of state-of-the-art (SOTA) visual foundation models to enhance combinational creativity in product design. ‘CombineDiff,’ a creativity support tool, was developed and evaluated through a mixed-methods study involving 18 participants. The study focused on assessing user interaction with the tool using the Creativity Support Index (CSI) and surprise ratings. The research also examined and suggested parameter settings that enhance creative outcomes. Our findings indicate that combinations yielding unexpected images receive higher surprise ratings, demonstrating the tool’s effectiveness in fostering P-creativity. Additionally, qualitative case studies highlighted the tool’s adaptability across different image types, showcasing its utility in both functional and aesthetic design combinations. This study contributes to the understanding of how visual foundation models can be effectively utilised in creative design processes.