A comparative study of writing and speaking prompts with refinement support : user preferences in creating AI-generated 3D designs

Abstract

Artificial intelligence has become an increasingly flexible tool for creative content generation, enabling users to produce images, skyboxes, and simple 3D assets through natural language prompts. Despite lowering technical barriers, many users struggle to formulate prompts that accurately reflect their visual intentions. This thesis investigates how users employ written and spoken prompting when generating skybox visualisations and simple 3D models within a prototype AI application, and how refinement support assists this process. Twenty participants interacted with a multimodal prompting system using text or speech and were able to refine their prompts. User behaviour was observed and followed by a structured questionnaire. The findings indicate that written prompting is preferred for precision and control, while spoken prompting supports early exploration. Refinement improved prompt clarity by restructuring descriptions, indirectly enhancing generation outcomes.