OmniTryOn: Video Try-On Anything at Once!
Researchers have introduced OmniTryOn, a new framework capable of simultaneously transferring multiple types of wearable items onto a person in a video. This approach moves beyond previous limitations of single-garment transfer and reliance on explicit garment masks, which often degrade visual quality and physical dynamics. To support this new task, they also released TryAny-Bench, a benchmark dataset and evaluation protocol, along with their OmniTryOn model which utilizes a novel First Frame Wearable Cache and Spatiotemporally Consistent RoPE for improved motion and dynamic preservation. AI
IMPACT Enables more realistic and complex virtual try-on experiences in video, potentially impacting e-commerce and fashion.