DMM: Building a Versatile Image Generation Model via Distillation-Based Model Merging

Official repository of our paper Building a Versatile Image Generation Model via Distillation-Based Model Merging.

Introduction

We propose a score distillation based model merging paradigm DMM, compressing multiple models into a single versatile T2I model.

Checkpoints

HuggingFace🤗: https://huggingface.co/MCG-NJU/DMM.

Usage

Install required packages with:

pip install -r requirements.txt

and initialize an Accelerate environment with:

accelerate config

An example of a training launch is in train.sh:

sh train.sh

An example of inference script is in inference.py:

python inference.py

Visualization

Results

Results combined with charactor LoRA

Results of interpolation between two styles

TODO

• Pre-training code.
• Model weight release.
• Incremental training code.
• Inference code with Diffusers.
• Journeydb dataset code.
• Evaluation code.
• Online demo.
• ComfyUI plugins.

Reference

@article{song2025dmm,
  title={DMM: Building a Versatile Image Generation Model via Distillation-Based Model Merging},
  author={Song, Tianhui and Feng, Weixin and Wang, Shuai and Li, Xubin and Ge, Tiezheng and Zheng, Bo and Wang, Limin},
  journal={arXiv preprint arXiv:2504.12364},
  year={2025}
}