Skip to content

MimicBrush

Zero-shot image editing from a reference image

Visit official site

What MimicBrush is

The model is described as a zero-shot image editing model that takes a marked region in a source image and an in-the-wild reference image, then captures the semantic correspondence between them and finishes the edit in one pass. The model is built on diffusion priors and self-supervised learning of semantic relations across images.

What you can do with it

  • Mark the region to edit in a source image with a white mask
  • Provide an in-the-wild reference image showing the desired region style
  • Let the model capture the semantic correspondence between the two images
  • Finish the editing task in one pass without complex parameter tuning

Getting started

  • Visit the MimicBrush web page and open the product page
  • Select or upload the source image to edit
  • Draw or upload a mask marking the region to edit
  • Provide or choose a reference image showing the desired editing effect

Example uses

  • Give the trees in a landscape photo the autumn colours of another image
  • Swap a product photo background texture for a more textured material
  • Recast an old photo as a modern piece of art

Pros & cons

✓ What we like

  • Editing needs no complex parameter tuning
  • A single pass produces diverse editing results

! What to watch out for

  • Editing requires both a masked source region and a reference image
  • The edit depends on the semantic correspondence the model captures

FAQ

What inputs does it need?

The material states a masked editing region in the source image and an in-the-wild reference image.

How is the model developed?

The material states diffusion priors and self-supervised learning of semantic relations across images.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review