nxnai/Voost

What it solves

Voost is designed to handle both virtual try-on (adding a garment to a person) and virtual try-off (removing a garment from a person) within a single, unified model. It aims to provide high-quality, realistic results that remain robust across different human poses, garment types, backgrounds, and lighting conditions.

How it works

Voost utilizes a scalable Diffusion Transformer (DiT) architecture to process bidirectional image manipulation. This allows the model to jointly manage the same set of tasks—adding or removing clothing—within a single transformer framework, ensuring consistency and quality across various image compositions.

Who it’s for

This project is intended for researchers and developers working in computer vision, digital fashion, and AI-driven image editing, specifically those focusing on virtual try-on and try-off technologies.

Highlights

• Unified framework for both virtual try-on and try-off.
• Based on a scalable Diffusion Transformer architecture.
• Robust to variations in human poses, garment categories, and lighting.
• Publicly available demo on Hugging Face Spaces.