<p>Translating abstract design intent into text-based prompts presents a practical barrier in generative AI–mediated design, particularly for novice users who lack stable design vocabulary. This study proposes a structured interaction framework that replaces free-form prompting with attribute-based design variable selection, implemented as a web-based fashion image generation system. The system organizes six core fashion design attributes into a taxonomy-driven variable space and converts user selections into prompts through a rule-based translation pipeline. To examine the proposed framework, the study combined exploratory image-generation analysis with a user study. The evaluation investigated whether structured prompt construction could preserve attribute-level interpretability across diffusion-based image generation conditions and how users perceived the interaction structure during fashion ideation tasks. The results suggest that the proposed interaction framework supports relatively stable attribute reflection and enables users to explore design variations through more structured and interpretable interaction processes. The study contributes to human–AI interaction design by demonstrating how domain-specific structure can be embedded into generative AI interaction architectures to reduce reliance on prompt-writing skills and support more transparent relationships between user input and generated output.</p>