Research on Generating Multi-angle Static Try-on Images for E-commerce Based on Conditional Diffusion Model

With the development of e-commerce industry, users' demand for virtual try-on experience of clothing products is increasing. Aiming at the problem of high cost and poor real-time performance of traditional 3D modeling, this paper proposes a multi-angle static try-on image generation method for e-commerce based on conditional diffusion model. The system takes clothing products provided by merchants freely selected by users as input, and combines with preset 2D human posture templates to generate static try-on renderings from front and left and right 45 angles through conditional diffusion models (such as Stable Diffusion). Technically, clothing segmentation and key point alignment technology are adopted to accurately match the clothing selected by users to the preset posture template; At the same time, through the conditional guidance diffusion model, a natural fusion multi-view try-on image is generated. Experiments show that this method has achieved high clothing alignment accuracy on the e-commerce clothing data set, and the structural similarity (SSIM) of the generated images is above 0.89. In the subjective evaluation of users, 85% think that the effect is "close to real people trying on". Compared with the existing methods, this system can ensure the generation quality and shorten the generation time of a single image to less than 2 seconds. The commercial application shows that the click-through rate of e-commerce pages equipped with this system is increased by 37%, and the conversion rate of user orders is increased by 22%. This study provides an efficient and low-cost solution for e-commerce virtual fitting, and promotes the application of AI technology in clothing industry.