Research on multimodal generative design of product appearance based on emotional and functional constraints

Abstract

The latest advancements in generative design have unveiled its potential in converting textual inputs into conceptual renderings, yet challenges remain in aligning these designs with users’ emotional and functional requirements. Additionally, research on integrating text, renderings, and 3D models within a multimodal product design framework is still inadequate. Consequently, this study introduces a novel multimodal generative design approach based on emotional and functional constraints. Firstly, a dataset of product emotional and functional vocabularies is constructed and utilized to train a stable diffusion model enhanced by LoRA. Subsequently, a Multi-stage Fuzzy Comprehensive Evaluation and Analytic Hierarchy Process are employed to assess the design proposals generated by the pre-trained model based on user requirements. Ultimately, the preferred 2D renderings are converted into detailed 3D models using the One-2–3-45++ method, with a case study on seat design validating its effectiveness. The primary contribution of this study lies in proposing a generative design method that accurately maps users’ personalized emotional and functional requirements to product styling, color, structure, and material, while also establishing a comprehensive and intelligent multimodal generative design framework incorporating a comprehensive evaluation system. Compared to existing methods, the proposed approach demonstrates superior performance in satisfying users’ emotional and functional requirements, significantly enhancing the personalization level of generative product design.