Skip to content

DiffPortrait3D

Synthesise 3D-consistent new views from one portrait

Visit official site

What DiffPortrait3D is

The model is described as a conditional approach that synthesises realistic 3D-consistent new views from one in-the-wild portrait while keeping identity and expression.

What you can do with it

  • Synthesise new views from a single picture
  • Keep identity and expression information
  • Handle extreme expressions and varied art styles

Who it is for

  • Portrait restoration and editing
  • Portrait view synthesis
  • Portrait animation work

Pros & cons

✓ What we like

  • Appearance context is injected into the frozen UNet
  • A conditioning module controls the rendered view

! What to watch out for

  • Input is one RGB portrait image
  • The backbone relies on a pre-trained 2D model

FAQ

What does it need as input?

A single RGB portrait image is described.

What is kept in the new view?

The material names identity and facial expression.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review