Skip to content

ip-composition-adapter

A Stable Diffusion 1.5 adapter that carries over image composition

Visit official site

What ip-composition-adapter is

The Hugging Face page describes an adapter built for Stable Diffusion 1.5 that injects general image composition into the model while mostly ignoring the reference image's style and content. Its example is a portrait of someone waving generating an image of a completely different person in the same waving pose. The page says the approach allows more flexible control than ControlNets, which match a control image more strictly.

What you can do with it

  • Reuse a portrait's pose with a different generated subject
  • Carry a landscape composition into a new scene
  • Transfer a design layout into a fresh design image

What to watch out for

  • Built specifically for Stable Diffusion 1.5, per the source

Pros & cons

✓ What we like

  • The distinction from ControlNet-style matching is explained
  • The intended behaviour is illustrated with a concrete example
  • Authorship and training credits are given on the page

! What to watch out for

  • Compatibility is limited to the stated model version
  • No licence, hardware or quota details are published here
  • Rights to the reference composition are not addressed

FAQ

What does it transfer?

General image composition, while largely ignoring style and content.

How does it differ from ControlNet?

The page says it is more flexible and does not match the control image as strictly.

Which model is it for?

The page states Stable Diffusion 1.5.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review