Skip to content

DiffHuman

Probabilistic 3D human reconstruction from one RGB image

Visit official site

What DiffHuman is

The method is a probabilistic approach to photorealistic 3D human reconstruction. It predicts a distribution from one RGB image and samples detailed results.

What you can do with it

  • Sample several 3D human models
  • Get richer detail in uncertain areas
  • Speed up inference with a rendering net

How it works

  • One RGB image starts the process
  • A distribution over shapes is predicted
  • Iterative denoising refines samples
  • A generative net speeds up rendering

Pros & cons

✓ What we like

  • Several plausible models can be sampled
  • Inference speed is raised by the network

! What to watch out for

  • A single RGB image is the input
  • No setup steps are listed

FAQ

What is the input?

The material says a single RGB image.

How are results produced?

The card describes iterative denoising sampling.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review