FlowErase-OPD: Multi-Concept Erasure via Anchored On-Policy Distillation in Flow Matching Models

2026-08-07

AI Summary

This work addresses the inefficiency and degraded generative performance commonly encountered by existing flow-matching models when simultaneously erasing multiple harmful concepts. To overcome these limitations, the authors propose FlowErase-OPD, a novel framework that compresses multiple single-concept erasure models into a unified LoRA module via in-policy distillation. The approach introduces two key innovations: Anchored Multi-Teacher Distillation (AMTD) and Adaptive Retention Control (ARC), which jointly optimize erasure efficacy and content fidelity. By integrating dynamic sampling with adaptive loss weighting, FlowErase-OPD achieves state-of-the-art performance across diverse erasure tasks—including nudity, object removal, and artistic style elimination—demonstrating significantly improved erasure completeness, image quality, semantic alignment, and adversarial robustness.

Abstract

Recent advances in flow matching models have substantially improved the quality of text-to-image generation, but have also raised increasing safety concerns due to their potential to generate harmful or undesirable content. Existing concept erasure methods for flow matching models predominantly focus on removing individual concepts, while effectively erasing multiple concepts simultaneously remains challenging. We propose FlowErase-OPD, a framework for multi-concept erasure based on on-policy distillation (OPD). Our approach first distills multiple single-concept erased models into a unified LoRA module and introduces Anchored Multi-Teacher Distillation (AMTD), which incorporates a retention teacher to mitigate the trade-off between concept erasure and preservation of generative capabilities. To further improve the coordination of multiple erasure objectives, we develop Adaptive Retention Control (ARC), which dynamically adjusts the sampling frequency and loss weight of each erasure teacher, together with the relative contribution of erasure and retention teachers throughout training. Extensive experiments on nudity, object, and artistic-style erasure demonstrate that FlowErase-OPD consistently improves the trade-off between erasure effectiveness, image quality, and semantic alignment, achieving state-of-the-art performance across diverse multi-concept erasure settings. Furthermore, the resulting models exhibit strong robustness against adversarial attacks. These results highlight the potential of on-policy distillation as a principled framework for safe and controllable generation in flow matching models.

Problem

• multi-concept erasure
• flow matching models
• text-to-image generation
• safety concerns
• undesirable content

Innovation

• on-policy distillation
• multi-concept erasure
• flow matching
• Anchored Multi-Teacher Distillation
• Adaptive Retention Control