Idea
A virtual try-on platform that automates multi-layer clothing draping on diverse characters for fashion and entertainment industries.
Research Paper
Core Innovation
This paper introduces LUIVITON, which uniquely separates clothing-to-body draping into two tasks using SMPL as a proxy, solved by geometric learning and diffusion models. It enables fully automated, multi-layer clothing draping on diverse humanoid characters without 2D sewing patterns. The system also allows post-draping customization of clothing size and material properties, supporting a wide range of character types beyond humans.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: growing demand for virtual try-on in fashion, gaming, and entertainment sectors.
Potential Customers & Pain Points
- Fashion retailers needing realistic virtual try-on
- Game developers requiring customizable character clothing
- Animation studios seeking efficient costume design
- Robotics companies wanting apparel simulation
- Virtual reality creators needing diverse avatar dressing
Business Model
SaaS platform offering API access and licensing for fashion, gaming, and animation companies with tiered pricing based on usage and customization features.
Competitive Landscape
- Zalando Virtual Try-On
- Metail
- Vue.ai
Implementation Challenges
- High computational requirements for real-time rendering
- Integration with existing 3D character pipelines
- User adoption in traditional fashion retail
Validation Strategy
- Develop prototype integrating with popular 3D character models
- Pilot with fashion retailers for virtual try-on feedback
- Collaborate with game studios to test multi-layer clothing draping
Research Paper Overview
LUIVITON: Learned Universal Interoperable VIrtual Try-ON
Summary
LUIVITON is an end-to-end system for fully automated virtual try-on that drapes complex, multi-layer clothing onto diverse humanoid characters using SMPL as a proxy. It separates clothing-to-body draping into clothing-to-SMPL and body-to-SMPL correspondence tasks, solved via geometric learning and diffusion models respectively. The system supports customization of clothing size and material properties post-draping and generalizes to humans, robots, cartoons, creatures, and aliens without requiring 2D sewing patterns.