Virtual Try-On Tool Voost Can Reproduce Fabric Texture and Wrinkle Details
Researchers introduced the Voost framework, which enables bidirectional learning for virtual dressing and undressing through a single diffusion transformer without additional labels. It innovatively uses attention temperature scaling and self-correcting sampling techniques to enhance model stability. It performs excellently in multiple benchmark tests, significantly improving garment-body alignment accuracy and generation quality. This technology brings breakthroughs to the digital fashion field, and the project is open-source.