Recent advances in text-to-image generative models have raised concerns about their potential to produce harmful content when provided with malicious input text prompts. To address this issue, two main approaches have emerged: (1) fine-tuning the model to unlearn harmful concepts and (2) training-free guidance methods that leverage negative prompts. However, we observe that combining these two orthogonal approaches often leads to marginal or even degraded defense performance. This observation indicates a critical incompatibility between two paradigms, which hinders their combined effectiveness. In this work, we address this issue by proposing a conceptually simple yet experimentally robust method: replacing the negative prompts used in training-free methods with implicit negative embeddings obtained through concept inversion. Our method requires no modification to either approach and can be easily integrated into existing pipelines. We experimentally validate its effectiveness on nudity and violence benchmarks, demonstrating consistent improvements in defense success rate while preserving the core semantics of input prompts.
TINA: Text-Free Inversion Attack for Unlearned Text-to-Image Diffusion Models
Harbin Institute of Technology (Shenzhen)(哈尔滨工业大学(深圳));City University of Hong Kong(香港城市大学);Shenzhen Loop Area Institute(深圳河套学院);Peng Cheng Laboratory(鹏城实验室);Shandong University(山东大学)
Regressor-free Molecule Generation to Support Drug Response Prediction
School of Computer Science(计算机学院);Wuhan University(武汉大学);University of Technology Sydney(悉尼科技大学);School of Information and Communication Technology(信息与通信技术学院);Griffith University(格里菲斯大学);School of Computing(计算机学院);Macquarie University(麦考瑞大学)
Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning
Institute of Data Science, National University of Singapore(数据科学研究所,新加坡国立大学);S-Lab, Nanyang Technological University(南洋理工大学S实验室);College of Computing and Data Science, Nanyang Technological University(计算与数据科学学院,南洋理工大学)
Safeguarding Facial Identity against Diffusion-based Face Swapping via Cascading Pathway Disruption
MoE Key Laboratory of Information Technology, Sun Yat-sen University, Guangzhou, China(信息科技教育部重点实验室,中山大学,广州,中国);State Key Laboratory of Mathematical Engineering and Advanced Computing, Zhengzhou, China(数学工程与先进计算国家重点实验室,郑州,中国)
QuantSparse: Comprehensively Compressing Video Diffusion Transformer with Model Quantization and Attention Sparsification
State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences(人工智能安全国家重点实验室,计算技术研究所,中国科学院);University of Chinese Academy of Sciences(中国科学院大学);ETH Zürich(苏黎世联邦理工学院);City College of New York, City University of New York, USA(纽约城市学院,纽约城市大学,美国);Shanghai Jiao Tong University(上海交通大学)
STITCH-OPE: Trajectory Stitching with Guided Diffusion for Off-Policy Evaluation
Department of Computer Science, University of Toronto(多伦多大学计算机科学系);University of Toronto Robotics Institute(多伦多大学机器人研究所);Toyota Research Institute(丰田研究院);Vector Institute(向量研究所)