Skip to content

PhysAvatar

Estimate body shape and clothing physics from multi-view video

Visit official site

What PhysAvatar is

The framework is described as combining inverse rendering and inverse physics to estimate body shape, appearance and clothing parameters from multi-view video.

What you can do with it

  • Feed multi-view video into the estimation pipeline
  • Track dynamic meshes with a 4D Gaussian grid
  • Estimate clothing physics through gradient-based optimisation

Application scenarios

  • Render high-fidelity human 3D models for film work
  • Build realistic garment animation for games and virtual reality
  • Show how clothing fits on an e-commerce platform

Pros & cons

✓ What we like

  • Works with motions and lighting beyond the training set
  • Estimates intrinsic material properties with a physics-based renderer

! What to watch out for

  • Input is multi-view video data
  • Clothing parameters come from optimisation against a simulator

FAQ

What does it estimate?

Body shape, outer appearance and physical parameters of clothing from multi-view video.

What can it render?

New views of avatars in loose clothing under unseen motions and lighting.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review