EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers
This addresses a critical safety and control issue for users of advanced text-to-image models, though it is incremental as it adapts erasure techniques to a new model paradigm.
The paper tackles the problem of removing unwanted concepts from new flow-based text-to-image models like Stable Diffusion v3 and Flux, where existing erasure methods fail, by introducing EraseAnything which formulates erasure as a bi-level optimization with LoRA tuning and attention regularization. It achieves state-of-the-art performance across various concept erasure tasks.
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.