Skip to content

MagicFace

Compose personalised portraits without training

Visit official site

What MagicFace is

The technique is described as a training free way to build high fidelity portraits by injecting reference concept features into the generated region.

What you can do with it

  • Generate a personalised portrait image
  • Integrate several concept features into one image
  • Apply the method to texture transfer

技術の要点

  • A coarse to fine flow starts with semantic layout construction
  • RSA and RBA mechanisms guide the feature injection
  • The user uploads reference images and picks style parameters

Pros & cons

✓ What we like

  • Multiple concept features can be integrated and customised
  • No training step is required to generate a portrait

! What to watch out for

  • The flow relies on uploaded reference images and concept features
  • Generation parameters such as style and concept are chosen by the user

FAQ

Does it need training?

The feature list states personalised portraits are generated with no training.

What are the two stages of the flow?

The material names semantic layout construction and concept feature injection.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review