An analysis of text-to-image generative models as creativity support tools

Abstract

This study examines the role of text-to-image (T2I) generative models as creativity support tools in architectural design education. It investigates how 28 first-year architecture students engage with prompt engineering to generate architectural representations using a T2I model. The methodology involved a structured task where students crafted and iterated prompts to recreate architectural forms, followed by a Creativity Support Index (CSI) survey and qualitative analysis of generated prompts and images. Findings indicate that structured prompts with precise modifiers significantly enhance output quality, while the trial-and-error nature of prompt crafting presents challenges. Students struggled to articulate complex visual concepts due to limited domain-specific vocabulary and lacked control over composition. CSI results showed strong support for exploration and expressiveness, though immersion and collaboration were limited. The study underscores the potential of T2I models in architectural pedagogy while highlighting the need for structured training in prompt engineering and improved AI-human interaction strategies.