Live intelligence

Latest AI updates

1 updates
Research

Multi-View Transformer with Geometry-Guided Multi-Modal ControlNet for 3D Virtual Try-On

Researchers introduce GS-VTON-MVT, a three-stage framework for consistent 3D Gaussian Splatting virtual try-on. It combines a multi-view inpainting transformer with full cross-view self-attention, a geometry-guided multi-modal ControlNet conditioned on garment images plus depth and surface normals, and pose-adaptive persona-aware attention. On the 3D-VTONBench benchmark, the method improves FID, KID, LPIPS-based cross-view consistency, and user preference over general-purpose 3D editors and prior 3D try-on methods.

3D virtual try-on3D Gaussian Splattingdiffusion models