Outfit Anyone

Outfit Anyone is an advanced AI tool designed for virtual try-on technology. This tool addresses the limitations of previous methods in generating high-fidelity and detail-consistent results. It leverages a two-stream conditional diffusion model, enabling lifelike handling of garment deformation and scalability factors like pose and body shape. The tool is applicable in diverse scenarios, from anime to real-world images, demonstrating its readiness for practical deployment​​.

The core of Outfit Anyone includes a conditional Diffusion Model that processes model images, garments, and text prompts, using garment images as a control factor. The network operates in two streams, independently processing model and garment data, converging within a fusion network for embedding garment details onto the model's feature representation. It features two key components: the Zero-shot Try-on Network for initial try-on imagery and the Post-hoc Refiner for enhancing clothing and skin texture in the output images.

Outfit Anyone showcases versatility in outfit changes, handling a wide range of clothing styles, and catering to different body types, including fit, curve, and petite. This adaptability extends to creating new animation characters. The tool also integrates with Animate Anyone, a state-of-the-art pose-to-video model, enabling outfit changes and motion video generation for any character. The project is primarily for academic research and effect demonstration, with most models and clothing images sourced from the internet and public datasets.