mlsanityData to 7 Oct 2026

Direct Unlearning Optimization for Robust and Safe Text-to-Image Models

arXiv 2407.21035 ↗First posted 17 Jul 2024cs.CVcited by 50

On the map: LLM safety and security → Vision-language and text-to-image safety

Abstract (arXiv)

Recent advancements in text-to-image (T2I) models have unlocked a wide range of applications but also present significant risks, particularly in their potential to generate unsafe content. To mitigate this issue, researchers have developed unlearning techniques to remove the model's ability to generate potentially harmful content. However, these methods are easily bypassed by adversarial attacks, making them unreliable for ensuring the safety of generated images. In this paper, we propose Direct Unlearning Optimization (DUO), a novel framework for removing Not Safe For Work (NSFW) content from T2I models while preserving their performance on unrelated topics. DUO employs a preference optimization approach using curated paired image data, ensuring that the model learns to remove unsafe visual concepts while retaining unrelated features. Furthermore, we introduce an output-preserving regularization term to maintain the model's generative capabilities on safe content. Extensive experiments demonstrate that DUO can robustly defend against various state-of-the-art red teaming methods without significant performance degradation on unrelated topics, as measured by FID and CLIP scores. Our work contributes to the development of safer and more reliable T2I models, paving the way for their responsible deployment in both closed-source and open-source scenarios.

Key facts. Cited by 50 papers in the map (data through 7 Oct 2026). Placed in Vision-language and text-to-image safety, one of 12 directions in LLM safety and security. Builds on 54 of its 67 references that are in the map. First posted on arXiv 17 Jul 2024.

Builds on

54 of 67 references.

Closest ancestors · 8

Same direction, newest first

  1. 2024Probing Unlearned Diffusion Models: A Transferable Adversarial Attack Perspective30 Apr
  2. 2024SafeGen: Mitigating Sexually Explicit Content Generation in Text-to-Image Models10 Apr
  3. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now18 Oct
  4. 2023Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?16 Oct
  5. 2023Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts12 Sep
  6. 2023SneakyPrompt: Jailbreaking Text-to-image Generative Models20 May
  7. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models30 Mar
  8. 2022Red-Teaming the Stable Diffusion Safety Filter3 Oct

Other references · 46

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  4. 2023GPT-4 Technical Reportcited by 18,132
  5. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  6. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  7. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  8. 2020Denoising Diffusion Implicit Modelscited by 9,655
  9. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
  10. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  11. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  12. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
34 more
  1. 2022Classifier-Free Diffusion Guidancecited by 5,717
  2. 2023Adding Conditional Control to Text-to-Image Diffusion Modelscited by 5,466
  3. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
  4. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,094
  5. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,595
  6. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,388
  7. 2022Imagen Video: High Definition Video Generation with Diffusion Modelscited by 1,765
  8. 2021Variational Diffusion Modelscited by 1,145
  9. 2023Diffusion Model Alignment Using Direct Preference Optimizationcited by 767
  10. 2019Advantage-Weighted Regression: Simple and Scalable Off-Policy Reinforcement Learningcited by 750
  11. 2023The Dawn of LMMs: Preliminary Explorations with GPT-4V(ision)cited by 677
  12. 2023Erasing Concepts from Diffusion Modelscited by 530
  13. 2022Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Modelscited by 501
  14. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 484
  15. 2023Unified Concept Editing in Diffusion Modelscited by 377
  16. 2022Understanding Diffusion Models: A Unified Perspectivecited by 337
  17. 2023Ablating Concepts in Text-to-Image Diffusion Modelscited by 312
  18. 2024MACE: Mass Concept Erasure in Diffusion Modelscited by 298
  19. 2023Large Language Model Unlearningcited by 288
  20. 2023Using Human Feedback to Fine-tune Diffusion Models without Any Reward Modelcited by 246
  21. 2023Task Arithmetic in the Tangent Space: Improved Editing of Pre-Trained Modelscited by 218
  22. 2023Unlearn What You Want to Forget: Efficient Unlearning for LLMscited by 211
  23. 2023Selective Amnesia: A Continual Learning Approach to Forgetting in Deep Generative Modelscited by 175
  24. 2023One-Dimensional Adapter to Rule Them All: Concepts, Diffusion Models and Erasing Applicationscited by 164
  25. 2024GPT-4V(ision) is a Human-Aligned Evaluator for Text-to-3D Generationcited by 151
  26. 2022Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content?cited by 150
  27. 2023Editing Implicit Assumptions in Text-to-Image Diffusion Modelscited by 128
  28. 2023Aligning Language Models with Preferences through f-divergence Minimizationcited by 118
  29. 2024Aligning Diffusion Models by Optimizing Human Utilitycited by 99
  30. 2023Circumventing Concept Erasure Methods For Text-to-Image Generative Modelscited by 81
  31. 2022On Reinforcement Learning and Distribution Matching for Fine-Tuning Language Models with no Catastrophic Forgettingcited by 80
  32. 2023Into the LAIONs Den: Investigating Hate in Multimodal Datasetscited by 54
  33. 2023CommonCanvas: An Open Diffusion Model Trained with Creative-Commons Imagescited by 15
  34. 2024Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion Modelscited by 12

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 6; 2025: 28; 2026: 16.

50 papers.

Sort:

By area:

  1. 2024SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generationcited by 125
  2. 2025Set You Straight: Auto-Steering Denoising Trajectories to Sidestep Unwanted Conceptscited by 61
  3. 2024Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Conceptscited by 39
  4. 2024Jailbreak Attacks and Defenses against Multimodal Generative Models: A Surveycited by 31
  5. 2024Fuzz-Testing Meets LLM-Based Agents: An Automated and Efficient Framework for Jailbreaking Text-To-Image Generation Modelscited by 24
  6. 2025CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Modelscited by 24
  7. 2025PromptGuard: Soft Prompt-Guided Unsafe Content Moderation for Text-to-Image Modelscited by 16
  8. 2025Integrating Reinforcement Learning with Visual Generative Models: Foundations and Advancescited by 11
  9. 2025Erasing Concepts, Steering Generations: A Comprehensive Survey of Concept Suppressioncited by 10
  10. 2025Erased but Not Forgotten: How Backdoors Compromise Concept Erasurecited by 9
  11. 2025Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearningcited by 9
  12. 2025Red-Teaming Text-to-Image Systems by Rule-based Preference Modelingcited by 8
38 more
  1. 2025SuMa: A Subspace Mapping Approach for Robust and Effective Concept Erasure in Text-to-Image Diffusion Modelscited by 8
  2. 2025EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignmentcited by 8
  3. 2024Safeguarding Text-to-Image Generation via Inference-Time Prompt-Noise Optimizationcited by 7
  4. 2025CE-SDWV: Effective and Efficient Concept Erasure for Text-to-Image Diffusion Models via a Semantic-Driven Word Vocabularycited by 7
  5. 2025Concept Steerers: Leveraging K-Sparse Autoencoders for Test-Time Controllable Generationscited by 6
  6. 2025Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Modelscited by 6
  7. 2025Plug-and-Play Interpretable Responsible Text-to-Image Generation via Dual-Space Multi-facet Concept Controlcited by 5
  8. 2025Concept Unlearning by Modeling Key Steps of Diffusion Processcited by 5
  9. 2025The Right to be Forgotten in Pruning: Unveil Machine Unlearning on Sparse Modelscited by 5
  10. 2026Neighbor-Aware Localized Concept Erasure in Text-to-Image Diffusion Modelscited by 4
  11. 2025Responsible Diffusion: A Comprehensive Survey on Safety, Ethics, and Trust in Diffusion Modelscited by 3
  12. 2025CGCE: Classifier-Guided Concept Erasure in Generative Modelscited by 3
  13. 2025Now You See It, Now You Don't - Instant Concept Erasure for Safe Text-to-Image and Video Generationcited by 3
  14. 2026Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shapingcited by 3
  15. 2025VCE: Safe Autoregressive Image Generation via Visual Contrast Exploitationcited by 2
  16. 2025Not Every Time and Frequency Need to Be Forgotten in Diffusion Unlearningcited by 2
  17. 2026SafeRoPE: Risk-specific Head-wise Embedding Rotation for Safe Generation in Rectified Flow Transformerscited by 2
  18. 2026Unlearning in Diffusion Models: A Unified Framework with KL Divergence and Likelihood Constraintscited by 2
  19. 2025Do Concept Replacement Techniques Really Erase Unacceptable Concepts?cited by 1
  20. 2025SecureT2I: No More Unauthorized Manipulation on AI Generated Images from Promptscited by 1
  21. 2025ZIUM: Zero-Shot Intent-Aware Adversarial Attack on Unlearned Modelscited by 1
  22. 2025Prompt-Based Safety Guidance Is Ineffective for Unlearned Text-to-Image Diffusion Modelscited by 1
  23. 2025Coffee: Controllable Diffusion Fine-tuningcited by 1
  24. 2026Critic-Guided Reinforcement Unlearning in Text-to-Image Diffusioncited by 1
  25. 2026Unlearning the Unpromptable: Prompt-free Instance Unlearning in Diffusion Modelscited by 1
  26. 2024Opt-In Art: Learning Art Styles Only from Few Examplescited by 0
  27. 2025Dual Caption Preference Optimization for Diffusion Modelscited by 0
  28. 2025Lifting Data-Tracing Machine Unlearning to Knowledge-Tracing for Foundation Modelscited by 0
  29. 2026You Don't Need All That Attention: Surgical Memorization Mitigation in Text-to-Image Diffusion Modelscited by 0
  30. 2026Relationship-Aware Safety Unlearning for Multimodal LLMscited by 0
  31. 2026When Safe Concepts Become Unsafe: Multi-Concept Compositional Vulnerabilities in Text-to-Image Modelscited by 0
  32. 2026Mosaic: Compositional Multi-Concept Erasure via Vector Field Blendingcited by 0
  33. 2026CoreUnlearn: Rethinking Concept Unlearning through Disentangled Component-Level Erasure in Text-guided Diffusion Modelscited by 0
  34. 2026TILDE: TILt-based Distributional Erasure for Concept Unlearningcited by 0
  35. 2026AutoAnchor: Stable Diffusion Unlearning Using Cross-Attention as a Manifold Surrogatecited by 0
  36. 2026FestDPO: Few-step Generator Alignment with Direct Preference Optimizationcited by 0
  37. 2026Weeding Out Bad Seeds: Initial-Noise-Robust Unlearning for Text-to-Image Diffusion Modelscited by 0
  38. 2026Continual Concept Erasure in Diffusion Models by Suppressing Cross-Edit Interferencecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models
  2. 2026Certifying Concept Unlearning in Text-to-Image Diffusion Models
  3. 2025GenBreak: Red Teaming Text-to-Image Generators Using Large Language Models
  4. 2023Adversarial Nibbler: A Data-Centric Challenge for Improving the Safety of Text-to-Image Models
  5. 2025Prompting Forgetting: Unlearning in GANs via Textual Guidance
  6. 2025A Comprehensive Survey on Concept Erasure in Text-to-Image Diffusion Models
  7. 2024ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users
  8. 2024AlignGuard: Scalable Safety Alignment for Text-to-Image Generation