ACE: Anti-Editing Concept Erasure in Text-to-Image Models

TL;DR

The paper introduces ACE, a novel method that effectively erases target concepts in text-to-image models during both generation and editing, improving control over harmful or unwanted content.

Contribution

ACE is the first approach to simultaneously erase concepts during generation and editing in text-to-image diffusion models, enhancing content control and safety.

Findings

ACE outperforms existing methods in concept erasure during editing.

ACE effectively filters out target concepts in various editing scenarios.

ACE demonstrates superior erasure of explicit concepts and artistic styles.

Abstract

Recent advance in text-to-image diffusion models have significantly facilitated the generation of high-quality images, but also raising concerns about the illegal creation of harmful content, such as copyrighted images. Existing concept erasure methods achieve superior results in preventing the production of erased concept from prompts, but typically perform poorly in preventing undesired editing. To address this issue, we propose an Anti-Editing Concept Erasure (ACE) method, which not only erases the target concept during generation but also filters out it during editing. Specifically, we propose to inject the erasure guidance into both conditional and the unconditional noise prediction, enabling the model to effectively prevent the creation of erasure concepts during both editing and generation. Furthermore, a stochastic correction guidance is introduced during training to address the erosion of unrelated concepts. We conducted erasure editing experiments with representative editing methods (i.e., LEDITS++ and MasaCtrl) to erase IP characters, and the results indicate that our ACE effectively filters out target concepts in both types of edits. Additional experiments on erasing explicit concepts and artistic styles further demonstrate that our ACE performs favorably against state-of-the-art methods. Our code will be publicly available at https://github.com/120L020904/ACE.