EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers

TL;DR

EraseAnything is a novel method designed to effectively remove unwanted concepts from the latest flow-based text-to-image models while preserving overall image quality, addressing a significant gap in current concept erasure techniques.

Contribution

It introduces a bi-level optimization framework with LoRA tuning and attention regularization tailored for flow-based T2I models, along with a self-contrastive learning strategy to prevent performance degradation.

Findings

Achieves state-of-the-art results in concept erasure tasks

Successfully adapts to new flow-based T2I architectures

Maintains generative quality after concept removal

Abstract

Removing unwanted concepts from large-scale text-to-image (T2I) diffusion models while maintaining their overall generative quality remains an open challenge. This difficulty is especially pronounced in emerging paradigms, such as Stable Diffusion (SD) v3 and Flux, which incorporate flow matching and transformer-based architectures. These advancements limit the transferability of existing concept-erasure techniques that were originally designed for the previous T2I paradigm (e.g., SD v1.4). In this work, we introduce EraseAnything, the first method specifically developed to address concept erasure within the latest flow-based T2I framework. We formulate concept erasure as a bi-level optimization problem, employing LoRA-based parameter tuning and an attention map regularizer to selectively suppress undesirable activations. Furthermore, we propose a self-contrastive learning strategy to ensure that removing unwanted concepts does not inadvertently harm performance on unrelated ones. Experimental results demonstrate that EraseAnything successfully fills the research gap left by earlier methods in this new T2I paradigm, achieving state-of-the-art performance across a wide range of concept erasure tasks.