Unified Concept Editing in Diffusion Models

TL;DR

Unified Concept Editing (UCE) offers a single, training-free approach to simultaneously address bias, copyright, and offensive content issues in text-to-image diffusion models, improving safety and scalability.

Contribution

The paper introduces UCE, a novel closed-form, training-free method for concurrent concept editing in diffusion models, unifying multiple safety-related modifications.

Findings

Effective simultaneous debiasing, style erasure, and content moderation.

Scalable to multiple concurrent edits in diffusion models.

Outperforms prior methods in efficacy and scalability.

Abstract

Text-to-image models suffer from various safety issues that may limit their suitability for deployment. Previous methods have separately addressed individual issues of bias, copyright, and offensive content in text-to-image models. However, in the real world, all of these issues appear simultaneously in the same model. We present a method that tackles all issues with a single approach. Our method, Unified Concept Editing (UCE), edits the model without training using a closed-form solution, and scales seamlessly to concurrent edits on text-conditional diffusion models. We demonstrate scalable simultaneous debiasing, style erasure, and content moderation by editing text-to-image projections, and we present extensive experiments demonstrating improved efficacy and scalability over prior work. Our code is available at https://unified.baulab.info