Prototype-Guided Concept Erasure in Diffusion Models

(CVPR2026) Yuze Cai, Jiahao Lu, Hongxiang Shi, Yichao Zhou, Hong Lu

Installation

Project Organization

Prototype-Guided Concept Erasure
	|---data
			(directory for prompts dataset and generated prompts)
	|---eval
			(directory for evaluate scripts)
	|---output
			(directory for experiment results and generated images)
	|---prototypes
			(directory for prototypes)
	|---src
			(directory for experiment scripts)

Environment Installation

conda create -n proto_ce python=3.12
concda activate proto_ce
pip install -r requirements.txt

Run

We have provided a script that covers the entire process from generating prompts, generating sample images, training, and erasing.

• When the concept to be erased is a concept in the I2P, the relevant concept name can be directly used as categories. When the concept to be erased is IP or style, the corresponding concept_facets in schema.json need to be modified.

bash run.sh

If you find our paper or this repo useful for your research, please consider citing:

@inproceedings{cai2026prototype,
        title={Prototype-Guided Concept Erasure in Diffusion Models},
        author={Cai, Yuze and Lu, Jiahao and Shi, Hongxiang and Zhou, Yichao and Lu, Hong},
        booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
        year={2026}
}