Image composition with pre-trained diffusion models

This repository contains code accompanying the post Image composition with pre-trained diffusion models.
The code is a minimal modification of the Stable Diffusion pipeline in the diffusers library by Hugging Face to implement ideas from MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation . You can look at the Colab Nootebook for usage.
If you use the default Stable Diffusion model specified in the code, please check the model card for the license and Use Restrictions.