TryOnDiffusion

About

TryOnDiffusion is a state-of-the-art, diffusion-based Virtual Try-On (VTON) framework that generates realistic try-on images in a zero-shot manner by combining an arbitrary person image with an in-shop garment. Its two-stage pipeline—Warping followed by Try-On Diffusion—leverages a Parallel-UNet design to robustly handle pose and garment misalignments while preserving fine clothing details. Trained and evaluated on DeepFashion and VITON, TryOnDiffusion demonstrates strong performance across challenging poses and clothing categories, with open-source code and pretrained models available and results validated in CVPR 2023.