Oxygen-TryOn: Fashion-Native Foundation Model for Any-item Virtual Try-On
Oxygen-TryOn is framed as a try-on model built for fashion from the ground up, not a general image editor adapted to clothes.
It takes one target image and one or more item references, including product shots or worn photos, and generates the subject wearing them. The paper says it can handle varied categories, body views, reference counts, and multi-item compositions while preserving identity and item appearance. Its training stack combines a dedicated try-on data engine, continued pre-training, supervised fine-tuning, and reinforcement learning with hybrid rewards. The authors report state-of-the-art or leading results against public benchmarks, an in-house bench, proprietary systems, and open-source models. HF Daily Papers' note
It takes one target image and one or more item references, including product shots or worn photos, and generates the subject wearing them. The paper says it can handle varied categories, body views, reference counts, and multi-item compositions while preserving identity and item appearance. Its training stack combines a dedicated try-on data engine, continued pre-training, supervised fine-tuning, and reinforcement learning with hybrid rewards. The authors report state-of-the-art or leading results against public benchmarks, an in-house bench, proprietary systems, and open-source models. HF Daily Papers' note
score 5