<p>Convenient and realistic garment texture transfer is essential for online shopping and a variety of applications, including virtual characters. Traditional texture transfer methods between real fabrics often rely on complex computations and physical simulations, which can be computationally expensive. Image-to-image garment model transformation frequently leads to incomplete geometries and textures, such as dynamic wrinkles on garments. In this paper, we propose a novel multi-stage architecture for synthesizing images of individuals with different garment textures. First, a semantic segmentation model is employed to extract the semantic map of the person in the image. Using the segmented garment images and masks, we construct a parametric garment model through 3D reconstruction. The garment images provide necessary depth information, while the segmentation mask constrains the model’s contours. During the rendering phase, parallax mapping is applied to project the fabric texture map onto the garment model’s surface. Lighting effects on the garment are adjusted by modifying the lighting parameters. Finally, the rendered garment image is merged with the original figure image and processed using a least mean filter to eliminate edge defects. We also analyze the effects of grid resolution and illumination parameters on the overall results. Both qualitative and quantitative experimental results demonstrate that our method achieves high-quality texture transfer, providing valuable insights for garment designers during the fit evaluation process.</p>