Abstract: To accurately grasp the users’ visual perception patterns and modeling preferences for products, and intelligent generation of product modeling design schemes that meet users’ emotional preferences, a design method that combines visual perception data with product modeling generation using a stable diffusion model is proposed. Firstly, users are segmented into clusters and intelligently identified based on their respective clusters using K-modes and the KNN algorithm. Then, according to the visual perception mode of design features, eye tracking technology is used to obtain users’ visual perception data of elemental features and semantic features, and the elemental features and semantic features micro datasets for different user clusters are constructed respectively. Finally, based on the multi-network combination architecture of DreamBooth and LoRA, the stable diffusion model is fine-tuned and trained to establish a composite model of product modeling generation with elemental feature generation first and semantic feature generation later. The model training time is 2 hours, and the single run time is less than 1 min. Meanwhile, the comprehensive visual evaluation is 1.24%~2.09% higher than the other models, and the matching degree of elemental features is mostly evaluated with a relative error of 0.08, which shows the advantages of the proposed method in design efficiency and demand matching degree. It can realize one-stop generation from user feature information to user requirement text to product styling design scheme with only a few samples for different user clusters.