Skip to content

audio2photoreal

Synthesize lifelike talking avatars from audio

Visit official site

What audio2photoreal is

The project is described as an open-source PyTorch implementation that synthesises photorealistic human figures in conversation from audio. The project lets researchers and developers train their own models for speech-driven avatar synthesis.

What you can do with it

  • Generate lifelike human avatars from audio
  • Use the provided pretrained models and dataset access
  • Work with the included face and body models
  • Render realistic avatars from speech input

Model notes

  • A face diffusion model is included
  • A body diffusion model is included
  • A body VQ VAE model is included
  • A body guidance transformer model is included

Example uses

  • Train a model on collected speech data for a custom avatar
  • Synthesise a virtual figure from historical speech material
  • Fit character voice work into 3D games and virtual spaces

Pros & cons

✓ What we like

  • Training and test code are both provided
  • The PyTorch implementation is open source

! What to watch out for

  • Avatar rendering is driven by audio input
  • Training your own model uses the provided codebase

FAQ

What does the project generate?

The material describes photorealistic human avatars in conversation synthesised from audio.

What models are included?

A face diffusion model, a body diffusion model, a body VQ VAE model and a body guidance transformer are listed.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review