HyperErase：用于文本到图像模型中多概念擦除的尺度校准超网络

中文摘要 AI 辅助

文本到图像（T2I）生成领域的最新进展大幅提升了视觉合成质量，但也因其可能生成有害或不良内容而引发了日益增长的安全担忧。现有的概念擦除方法主要遵循静态权重范式，生成单一冻结适配器，难以适应多样化的提示词变化，并且在扩展到多概念时面临参数干扰问题。我们提出HyperErase，一种基于超网络驱动的提示条件参数合成的概念擦除框架。我们的方法首先将概念擦除重新定义为提示条件参数摊销，并训练一个超网络将文本描述映射为提示特定的LoRA更新，从而无需逐提示梯度优化或手动LoRA合并。为了进一步提升合成适配器的稳定性和精度，我们开发了一种解耦修正策略，将LoRA令牌解耦为模式和尺度子空间，应用平方根变换以抑制乘法性过度缩放，并利用教师派生的规范先验进行推理时修正。在主要概念类别上的大量实验表明，HyperErase持续改善了擦除有效性、图像质量和语义对齐之间的权衡，达到了与黄金标准单概念基线相当的性能。此外，所得模型能够在单次前向传播中为每个输入提示变体提供专门的LoRA，而无需在推理期间进行梯度更新。这一原理性且灵活的框架为T2I模型中的概念擦除提供了新范式。

英文摘要

Recent advances in text-to-image (T2I) generation have substantially improved visual synthesis, but have also raised increasing safety concerns due to their potential to generate harmful or undesirable content. Existing concept erasure methods predominantly follow a static weight paradigm, producing a single frozen adapter that struggles to adapt to diverse prompt variations and suffers from parameter interference when scaling to multiple concepts. We propose \textbf{HyperErase}, a framework for concept erasure based on hypernetwork-driven prompt-conditioned parameter synthesis. Our approach first reframes concept erasure as prompt-conditioned parameter amortization and trains a hypernetwork to map textual descriptions to prompt-specific LoRA updates, eliminating the need for per-prompt gradient optimization or manual LoRA merging. To further improve the stability and precision of synthesized adapters, we develop a decoupled rectification strategy, which disentangles LoRA tokens into pattern and scale subspaces, applies a square-root transform to curb multiplicative over-scaling, and leverages teacher-derived canonical priors for inference-time correction. Extensive experiments across major concept categories demonstrate that HyperErase consistently improves the trade-off between erasure effectiveness, image quality, and semantic alignment, achieving performance comparable to gold-standard single-concept baselines. Furthermore, the resulting models can provide specialized LoRAs for each input prompt variation in a single forward pass without requiring gradient updates during inference. These principled and flexible framework offers a new paradigm for concept erasure in T2I models.

发表机构

• Harbin Institute of Technology, Shenzhen（哈尔滨工业大学（深圳））
• Jilin University（吉林大学）
• Tsinghua Shenzhen International Graduate School, Tsinghua University（清华大学深圳国际研究生院）
• South China University of Technology（华南理工大学）
• Peng Cheng Laboratory（鹏城实验室）

机构由 AI 辅助整理，请以论文原文为准。