Improving Fine-Grained Control via Aggregation of Multiple Diffusion Models

[arXiv]

Aggregation of Multi Diffusion Models (AMDM) algorithm can aggregate features from different conditional control diffusion models with the same theoretical foundations into a specific model, enabling fine-grained conditional control.

In this GitHub repository, we provide three examples to verify the effectiveness of the algorithm.

How to use

1. Installing conda environment.

   conda create --name AMDM python=3.8
   conda activate AMDM
   pip install -r requirements.txt
   

2. Prepare corresponding checkpoint from hugginface and setting path in configs.

3. Runing xxxx_ui.py. For example: python interactmigc_ui.py.

Thanks to InteractDiffusion, MIGC and IP-Adapter.