IDM-VTON is an open-source diffusion model for virtual try-on: given a photo of a person and an image of a garment, it generates a realistic image of the person wearing that garment, preserving logos, text and fine fabric detail that earlier try-on systems tended to smear. Presented at ECCV 2024, the project has more than 5,200 stars on GitHub, with code and model checkpoints released for research use.
Virtual try-on is a demanding test of image generation: the garment must drape plausibly on a new body while its identity — prints, lettering, texture — survives intact. IDM-VTON improves diffusion models for this with two complementary pathways: an IP-Adapter encodes the garment’s high-level semantics, while a parallel garment UNet (GarmentNet) extracts the low-level features that carry detail, and the main try-on UNet fuses both with the person’s pose and shape. The result set a strong open benchmark for “in the wild” try-on, handling varied poses, backgrounds and garment types better than prior GAN and diffusion baselines. The repository provides inference code, pretrained checkpoints and a demo setup, making it a common starting point for fashion-tech prototypes and for researchers comparing try-on methods. As with many research releases, the license terms restrict use to non-commercial research, so companies should review them before building on it.
Free. IDM-VTON is released for research use at no cost; check the repository license before any commercial use.
Fashion-tech developers prototyping virtual fitting, and computer-vision researchers working on diffusion-based image editing and try-on.