Skip to content

Visual Anagrams

Build multi-view illusions with a zero-shot method

Visit official site

What Visual Anagrams is

The method is described as a simple zero-shot way to generate multi-view visual illusions, with theory and practice shown for a wide set of transformations.

What you can do with it

  • Generate illusions seen from more than one view
  • Apply rotation, flipping and colour inversion
  • Use tilt, jigsaw reordering or random permutation

How it works

  • A pre-trained diffusion model estimates noise across views
  • The noise estimates are aligned and averaged
  • The averaged estimate drives each diffusion step

Pros & cons

✓ What we like

  • The approach needs no extra training as a zero-shot method
  • A wide range of transformations is supported

! What to watch out for

  • Outputs depend on the pre-trained diffusion model
  • Illusions follow the chosen view transformations

FAQ

Which transforms are supported?

Rotation, flipping, colour inversion, tilt, jigsaw reordering and random permutation are listed.

How are images produced?

The material states noise estimates from different views are aligned, averaged and used for diffusion steps.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review