Combining AIGC and Virtual Simulation Technology to Enhance the Personalization of Apparel Designs

Abstract

With the increasing demand of consumers for personalized experience and fast response, the design and production model of traditional apparel companies is facing serious challenges. In order to adapt to market changes and maintain competitive advantages, the industry urgently needs to adopt emerging technologies. In this paper, based on CycleGAN, we make the generated images more natural by improving the network model and adding background optimization loss to achieve clothing image style migration and fusion. Based on the LoRA-DAE framework, the method of LoRA integration into the stable diffusion model is proposed, which enhances the model adaptation ability. A fashion clothing model dataset is established to provide data support for the model. Through comparative experiments to evaluate the viewing experience of the method proposed in this paper, the model of this paper scores 27.1585, 14.4955, 17.6485, and 13.8486 on overall, shape, shadow, and detail texture, respectively.Meanwhile, it is also superior to other algorithms on subjective scores, and the average score of users' subjective evaluations is 27.9588.The model of this paper is mounted on the apparel personalization applet, and the experimenter's evaluation of this paper's apparel personalization apparel is conducted. When the model of this paper is installed on the apparel personalized customization app, the “gaze duration” of the experimenters on the apparel customization app is 1834.6985ms, which is 41.7156% higher than that of the control group, and the mean value of the subjective experience rating is 4.2663, which is higher than that of the control group, and based on the model of this paper, the model provides a higher psychological experience.