mlsanityData to 7 Oct 2026

Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models

arXiv 2407.12383 ↗First posted 17 Jul 2024cs.CVcited by 123

On the map: Diffusion models → Robustness, privacy and adversarial aspects

Abstract (arXiv)

Text-to-image models encounter safety issues, including concerns related to copyright and Not-Safe-For-Work (NSFW) content. Despite several methods have been proposed for erasing inappropriate concepts from diffusion models, they often exhibit incomplete erasure, consume a lot of computing resources, and inadvertently damage generation ability. In this work, we introduce Reliable and Efficient Concept Erasure (RECE), a novel approach that modifies the model in 3 seconds without necessitating additional fine-tuning. Specifically, RECE efficiently leverages a closed-form solution to derive new target embeddings, which are capable of regenerating erased concepts within the unlearned model. To mitigate inappropriate content potentially represented by derived embeddings, RECE further aligns them with harmless concepts in cross-attention layers. The derivation and erasure of new representation embeddings are conducted iteratively to achieve a thorough erasure of inappropriate concepts. Besides, to preserve the model's generation ability, RECE introduces an additional regularization term during the derivation process, resulting in minimizing the impact on unrelated concepts during the erasure process. All the processes above are in closed-form, guaranteeing extremely efficient erasure in only 3 seconds. Benchmarking against previous approaches, our method achieves more efficient and thorough erasure with minor damage to original generation ability and demonstrates enhanced robustness against red-teaming tools. Code is available at urlhttps://github.com/CharlesGong12/RECE.

Key facts. Cited by 123 papers in the map (data through 7 Oct 2026). Placed in Robustness, privacy and adversarial aspects, one of 14 directions in Diffusion models. Builds on 26 of its 36 references that are in the map. First posted on arXiv 17 Jul 2024.

Builds on

26 of 36 references.

Closest ancestors · 7

Same direction, newest first

  1. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now18 Oct
  2. 2023Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?16 Oct
  3. 2023Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts12 Sep
  4. 2023Unified Concept Editing in Diffusion Models25 Aug
  5. 2023Erasing Concepts from Diffusion Models13 Mar
  6. 2022Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models9 Nov
  7. 2022Red-Teaming the Stable Diffusion Safety Filter3 Oct

Other references · 19

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  4. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  5. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  6. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  7. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  8. 2021Diffusion Models Beat GANs on Image Synthesiscited by 8,185
  9. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  10. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  11. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,595
  12. 2022DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generationcited by 3,286
7 more
  1. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,388
  2. 2022Imagic: Text-Based Real Image Editing with Diffusion Modelscited by 1,172
  3. 2022Diffusion Models: A Comprehensive Survey of Methods and Applicationscited by 1,107
  4. 2016Tutorial on Variational Autoencoderscited by 404
  5. 2023Selective Amnesia: A Continual Learning Approach to Forgetting in Deep Generative Modelscited by 175
  6. 2023Editing Implicit Assumptions in Text-to-Image Diffusion Modelscited by 128
  7. 2021Attacking Video Recognition Models with Bullet-Screen Commentscited by 16

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 12; 2025: 52; 2026: 59.

123 papers.

Sort:

By area:

  1. 2024SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generationcited by 125
  2. 2025SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoderscited by 75
  3. 2025Set You Straight: Auto-Steering Denoising Trajectories to Sidestep Unwanted Conceptscited by 61
  4. 2024Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Conceptscited by 39
  5. 2025SPEED: Scalable, Precise, and Efficient Concept Erasure for Diffusion Modelscited by 33
  6. 2024Jailbreak Attacks and Defenses against Multimodal Generative Models: A Surveycited by 31
  7. 2025Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptationcited by 27
  8. 2024TraSCE: Trajectory Steering for Concept Erasurecited by 25
  9. 2025CURE: Concept Unlearning via Orthogonal Representation Editing in Diffusion Modelscited by 24
  10. 2025Training-Free Safe Denoisers for Safe Use of Diffusion Modelscited by 19
  11. 2025ACE: Anti-Editing Concept Erasure in Text-to-Image Modelscited by 17
  12. 2025Concept Pinpoint Eraser for Text-to-image Diffusion Models via Residual Attention Gatecited by 16
38 more
  1. 2024Unstable Unlearning: The Hidden Risk of Concept Resurgence in Diffusion Modelscited by 14
  2. 2025TRCE: Towards Reliable Malicious Concept Erasure in Text-to-Image Diffusion Modelscited by 14
  3. 2025One Image is Worth a Thousand Words: A Usability Preservable Text-Image Collaborative Erasing Frameworkcited by 14
  4. 2025When Are Concepts Erased From Diffusion Models?cited by 14
  5. 2025T2VUnlearning: A Concept Erasing Method for Text-to-Video Diffusion Modelscited by 12
  6. 2024Unveiling Concept Attribution in Diffusion Modelscited by 11
  7. 2025Concept Corrector: Erase concepts on the fly for text-to-image diffusion modelscited by 11
  8. 2025Video Unlearning via Low-Rank Refusal Vectorcited by 11
  9. 2025Erasing Concepts, Steering Generations: A Comprehensive Survey of Concept Suppressioncited by 10
  10. 2024Text-to-Image Synthesis: A Decade Surveycited by 9
  11. 2025A Comprehensive Survey on Concept Erasure in Text-to-Image Diffusion Modelscited by 9
  12. 2025Erased but Not Forgotten: How Backdoors Compromise Concept Erasurecited by 9
  13. 2025Erased or Dormant? Rethinking Concept Erasure Through Reversibilitycited by 9
  14. 2025Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearningcited by 9
  15. 2026When Safety Collides: Resolving Multi-Category Harmful Conflicts in Text-to-Image Diffusion via Adaptive Safety Guidancecited by 9
  16. 2025Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuningcited by 8
  17. 2025SAEmnesia: Erasing Concepts in Diffusion Models with Supervised Sparse Autoencoderscited by 8
  18. 2025EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignmentcited by 8
  19. 2026TraceRouter: Robust Safety for Large Foundation Models via Path-Level Interventioncited by 8
  20. 2025CE-SDWV: Effective and Efficient Concept Erasure for Text-to-Image Diffusion Models via a Semantic-Driven Word Vocabularycited by 7
  21. 2025Concept Steerers: Leveraging K-Sparse Autoencoders for Test-Time Controllable Generationscited by 6
  22. 2025CRCE: Coreference-Retention Concept Erasure in Text-to-Image Diffusion Modelscited by 6
  23. 2025Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Modelscited by 6
  24. 2025Training-Free Safe Text Embedding Guidance for Text-to-Image Diffusion Modelscited by 6
  25. 2025AutoPrompt: Automated Red-Teaming of Text-to-Image Models via LLM-Driven Adversarial Promptscited by 6
  26. 2026SafeMo: Linguistically Grounded Unlearning for Trustworthy Text-to-Motion Generationcited by 6
  27. 2026GRP-Obliteration: Unaligning LLMs With a Single Unlabeled Promptcited by 6
  28. 2025DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution Modelingcited by 5
  29. 2025Locality-Aware Continual Unlearning for Diffusion Modelscited by 5
  30. 2024Moderating the Generalization of Score-based Generative Modelcited by 4
  31. 2025DyME: Dynamic Multi-Concept Erasure in Diffusion Models with Bi-Level Orthogonal LoRA Adaptationcited by 4
  32. 2025EMMA: Concept Erasure Benchmark with Comprehensive Semantic Metrics and Diverse Categoriescited by 4
  33. 2026Forget Many, Forget Right: Scalable and Precise Concept Unlearning in Diffusion Modelscited by 4
  34. 2026Safety-Guided Flow (SGF): A Unified Framework for Negative Guidance in Safe Generationcited by 4
  35. 2026Neighbor-Aware Localized Concept Erasure in Text-to-Image Diffusion Modelscited by 4
  36. 2026Co-occurring Associated REtained concepts in Diffusion Unlearningcited by 4
  37. 2024Negative Token Merging: Image-based Adversarial Feature Guidancecited by 3
  38. 2025Human Motion Unlearningcited by 3

50 most cited of 123 All 123 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models
  2. 2024Erasing Concepts from Text-to-Image Diffusion Models with Few-shot Unlearning
  3. 2024MACE: Mass Concept Erasure in Diffusion Models
  4. 2024Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters
  5. 2023Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Erasers
  6. 2025M-ErasureBench: A Comprehensive Multimodal Evaluation Benchmark for Concept Erasure in Diffusion Models
  7. 2023All but One: Surgical Concept Erasing with Model Preservation in Text-to-Image Diffusion Models
  8. 2025ACE: Attentional Concept Erasure in Diffusion Models