Abstract

The emergence of virtual try-on technology signifies a crucial development in the progression of the retail sector, especially in online fashion. GAN and early diffusion-based models that integrates both high-level semantic features and low-level garment-specific attributes through a dual-modular architecture - leveraging an IP-Adapter for cross-attention with textual and visual embeddings, and a dedicated GarmentNet for detail preservation via self-attention layers. By using this type of models, user customize the person garments images with text captions. VITON - HD and Dress code datasets are used for evaluation for these models in for real world conditions. Virtual try-on models can be categorized into three primary types: image-based, multi-pose, and video virtual try-on models. Every category provides distinct features and technical methods to replicate the try-on experience. Image-driven models, represented by methods such as VITON and CP-VTON, produce authentic try-on outcomes.