OOTDiffusion is the official implementation of the AAAI 2025 paper “Outfitting Fusion based Latent Diffusion for Controllable Virtual Try-on.” Give it a photo of a person and a photo of a garment, and it generates a realistic image of that person wearing the garment. The repository has earned about 6,600 stars on GitHub and was released by the paper’s authors in 2024.
Virtual try-on is harder than generic image editing: the garment must drape over a specific body, preserve its pattern, logos and fine details, and stay consistent under different poses. OOTDiffusion introduces OOTNet, an outfitting fusion network that learns how clothing deforms, and conditions a latent diffusion model on both the garment features and the target body so the generated image keeps the clothing’s identity while fitting naturally. The codebase builds on the diffusers ecosystem and includes training and inference pipelines, pre-trained checkpoints for standard benchmarks such as VITON-HD and Dress Code, and a Gradio-style demo path for trying pairs of images quickly. As an official research release it is aimed at reproducibility and further research rather than being a polished consumer product.
Free. OOTDiffusion is released as open research code by the paper’s authors; check the repository for the exact terms covering the model checkpoints before commercial use.
Computer-vision researchers and e-commerce prototypers exploring AI virtual try-on who want the official, reproducible implementation of a leading diffusion-based method.