Research map / Image generation / Text-to-image diffusion
Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models
Hoigi Seo, Byung Hyun Lee, Jaehyun Cho, Sungjin Lim, Se Young Chun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Student-t mixture instead of Gaussian mixture (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 8.96 -> 0.24 | Improved |
| use affine optimal transport instead of surrogate mapping (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%): 86.48 -> 89.37 | Improved |
| add Gaussian noise anchors to boundary-sampled anchors (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 1.12 -> 0.24 | Improved |
| add boundary-sampled anchors to noise anchors (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%): 88.08 -> 89.37 | Improved |
| use structured noise instead of full-rank noise in NIR (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%) after restore: 79.28 -> 89.37 | Improved |
| use MoE instead of linear module (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 7.89 -> 0.24 | Improved |
| use MoE instead of FFN-GLU (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 0.48 -> 0.24 | Improved |
| add AdaLN-Zero to MoEraser (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%): 89.32 -> 89.37 | No clear effect |
| increase mapping concepts from 3 to 10 (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 0.24 -> 3.92 | Hurt |
| raise confidence level from 0.9 to 0.999 (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%): 89.37 -> 82.52 | Hurt |
| lower confidence level from 0.9 to 0.7 (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_t (%): 0.24 -> 2.32 | Mixed |
| raise noise scale from 2.0e-3 to 2.0e-2 (ablation) | ETC · 50-celebrity erasure (SDv1.4) | Acc_r (%): 89.37 -> 85.21 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAFREE — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.93)
- Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters (CVPR 2025 · similarity 0.93)
- All but One: Surgical Concept Erasing with Model Preservation in Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.92)
- Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptation (CVPR 2025 · similarity 0.92)
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.92)
- Concept Pinpoint Eraser for Text-to-image Diffusion Models via Residual Attention Gate (ICLR 2025 · similarity 0.92)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
