Researchers from Imperial College London, the University of Exeter and Zhejiang University conduct empirical studies comparing large language models, text-to-image, and text-to-3D AI tools across combinational creativity tasks, revealing how each model excels at ideation, sketch visualization, and prototype development.

Key points

• LLMs achieve highest performance in linguistic-based combinational tasks like interpolation and replacement, driving conceptual ideation.
• Text-to-image models effectively externalize design ideas into rapid visual sketches, improving mid-stage visualization accuracy.
• Text-to-3D models excel at spatial operations and prototype generation, facilitating robust physical deformation and structural evaluation.

Why it matters: This framework enables designers to match specialized AI models to each phase of the creative process, enhancing innovation and efficiency in design workflows.