Prompt-guided Diffusion Approach for Deep Generative Vehicle Designs under User-defined Specifications

Engineering vehicle design presents significant challenges due to its resource-intensive nature and the need to satisfy diverse design requirements. While recent data-driven generative models have shown promise, key limitations remain: (1) the difficulty of simultaneously addressing multiple design factors, such as user intent and engineering validity, and (2) the lack of accessibility and user-friendliness for intuitive design exploration. To overcome these challenges, this study introduces a prompt-guided diffusion framework for generating diverse and high-fidelity vehicle designs. Specifically, our proposed method integrates a diffusion framework with language-based prompt conditioning to ensure multiple user-defined design requirements, including vehicle type, styling, and aerodynamic performance. Both qualitative and quantitative evaluations demonstrate substantial improvements over baseline models, achieving up to $87.7% $ improvement in Fréchet Inception Distance (FID) and $72.2% $ improvement in learned perceptual image patch similarity (LPIPS). Furthermore, the proposed method exhibits robust generalization, producing valid design candidates even under unseen or data-scarce conditions. This work underscores the potential of deep generative approaches to accelerate engineering design exploration, offering greater diversity and extensibility for user-centered engineering workflows.