Research map / Image generation / Text-to-image diffusion
TRCE: Towards Reliable Malicious Concept Erasure in Text-to-Image Diffusion Models
Ruidong Chen, Honglin Guo, Lanjun Wang, Chenyu Zhang, Weizhi Nie, An-An Liu
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 |
|---|---|---|---|
| add second-stage denoising trajectory steering to textual erasure (ablation) | TRCE · sexual concept, I2P | ASR: 5.05% -> 1.29% | Improved |
| add second-stage steering to textual erasure (ablation) | TRCE · sexual concept, MMA-Diffusion | ASR: 7.80% -> 1.40% | Improved |
| add first-stage textual erasure to trajectory steering (ablation) | TRCE · sexual concept, I2P | ASR: 13.86% -> 1.29% | Improved |
| add second stage at high knowledge preservation rate eta=0.02 (ablation) | TRCE · adversarial prompts | Adv ASR: 13.52% -> 2.36% | Improved |
| optimize [EoT] embedding instead of [KEY] only (ablation) | TRCE · I2P sexual | ASR: 22.8% -> 5.05% | Improved |
| also optimize [KEY] with [EoT] (ablation) | TRCE · I2P sexual | ASR: 5.05% -> 6.34% | Hurt |
| add guidance enhancement in second stage (ablation) | TRCE · I2P sexual | ASR: 4.65% -> 1.29% | Improved |
| add L_preserve in second stage (ablation) | TRCE · I2P sexual | FID_gen: 14.64 -> 12.08 | Improved |
| add contrastive L_erase instead of safe alignment only (ablation) | TRCE · I2P sexual | ASR: 1.93% -> 1.29% | Improved |
| raise knowledge preservation rate eta from 0.002 to 0.02 (stage one only) (ablation) | TRCE · I2P sexual | ASR: 3.54% -> 7.20% | Hurt |
| use TRCE vs MACE for multi-concept erasure | TRCE · I2P 7 categories | overall inappropriate rate: 5.6% -> 2.0% | Improved |
| use TRCE(T+V) vs AdvUnlearn on UnlearnDiff attack | TRCE · sexual concept | ASR: 3.52% -> 0.70% | Improved |
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 3 of 3 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
- Six-CD: Benchmarking Concept Removals for Text-to-image Diffusion Models (CVPR 2025 · similarity 0.90)
- Circumventing Concept Erasure Methods For Text-To-Image Generative Models (ICLR 2024 · similarity 0.89)
- ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning (ICLR 2025 · similarity 0.89)
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Growth Inhibitors for Suppressing Inappropriate Image Concepts in Diffusion Models (ICLR 2025 · similarity 0.89)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · similarity 0.88)
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.
