metadata
license: mit
tags:
- diffusion
- concept-erasure
- stable-diffusion
- ga
- chainsaw
datasets:
- imagenet
language:
- en
pipeline_tag: text-to-image
ga_chainsaw
This is a concept-erased Stable Diffusion model using the Gradient Ascent (GA) method to remove the concept "Chainsaw".
Method
Gradient Ascent (GA) actively pushes the model away from generating the target concept.
Usage
from diffusers import StableDiffusionPipeline
import torch
pipe = StableDiffusionPipeline.from_pretrained("ErasureResearch/ga_chainsaw", torch_dtype=torch.float16).to("cuda")
prompt = "a photo of a chainsaw"
image = pipe(prompt).images[0]
image.save("erased_chainsaw.png")
Citation
If you use this model in your research, please cite:
@article{concept_erasure_2024,
title={Concept Erasure in Diffusion Models},
author={ErasureResearch Team},
journal={Proceedings of...},
year={2024}
}