Cornell University researchers, in partnership with NVIDIA, have developed an advanced method for creating digital images of cloth that more accurately capture textile textures. This new study was presented on December 16 at the Association for Computing Machinery's SIGGRAPH Asia 2025 meeting in Hong Kong.
Digitally rendering fabric has historically been challenging due to the complex ways light interacts with different woven or knitted yarns. The structure of fabric, composed of fibers twisted into plies and then into yarns, varies significantly by material (e.g., wool's oval fibers, cotton's kidney shape, silk's polygonal shape).
Professor Steve Marschner, who has led research in this area for over two decades, highlighted the difficulty in achieving realistic fabric renders. Doctoral student Yunchen Yu, the study's first author, noted the diverse nature of fabric structures makes a single universal model unlikely.
The new method models how light interacts with yarns, both as it passes through and reflects off the fabric. Yu's approach, developed with NVIDIA's Andrea Weidlich, utilizes a combination of ray optics for generating average color and highlights, and wave optics for simulating light passing through fabric and creating subtle glints. Initially, a pure wave optics simulation proved too computationally intensive; the combined method optimized this by using faster ray optics where suitable.
Marschner's lab has a long history in fabric rendering. Previous advancements include Piti Irawan's model for light reflection off fibers, and work with Shuang Zhao and Kavita Bala using microCT scanning to image woven fibers at a detailed scale. Later efforts, in collaboration with Doug James, focused on physical simulations of yarn arrangements in woven and knitted materials. A side project with Brooks Hagan enabled interior designers to visualize custom textiles.
Looking ahead, Yu aims to incorporate artificial intelligence to bypass the simulation step, making the model faster and more flexible. Marschner anticipates that integrating generative AI techniques will be crucial for more efficient fabric modeling, leading to higher quality renderings in the gaming and animation industries.