Skip to content

DragonDiffusion

Edit images by dragging objects with diffusion

Visit official site

What DragonDiffusion is

The method is described as a diffusion model based approach to fine grained image editing that changes intermediate representations through gradients.

What you can do with it

  • Move an object inside an image
  • Resize an object in place
  • Replace the appearance of an object

How the editing works

  • A feature correspondence loss creates the edit gradient
  • Semantic and geometric alignment are weighed at several scales
  • Cross branch self attention keeps the result consistent with the original

Pros & cons

✓ What we like

  • Several editing modes share one framework
  • Multi scale alignment guides the edit gradient

! What to watch out for

  • Edit signals pass through feature correspondence loss
  • Consistency depends on cross branch self attention

FAQ

Which edits are supported?

The material lists object moving, resizing, appearance replacement and content dragging.

How are edits applied?

A feature correspondence loss converts them into gradients over intermediate representations.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review