Research map / Image generation / Text-to-image diffusion
Z-Erase: Enabling Concept Erasure in Single Stream Diffusion Transformers
Nanxiang Jiang, Zhaoxin Fan, Baisen Wang, Daiheng Gao, Junhang Cheng, Jifeng Guo, Yalan Qin, Yeying Jin, Hongwei Zheng, Faguo Wu, wenjun wu
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 L_attn to L_erase (ablation) | Z-Erase · I2P nudity | detected nudity: 206 -> 148 | Improved |
| add L_attn to L_erase (ablation) | Z-Erase · I2P nudity | CLIP: 29.86 -> 29.12 | Hurt |
| add L_pr to L_erase (ablation) | Z-Erase · I2P nudity | FID: 29.65 -> 26.42 | Improved |
| add L_pr to L_erase (ablation) | Z-Erase · I2P nudity | CLIP: 29.86 -> 31.28 | Improved |
| add L_pr to L_erase (ablation) | Z-Erase · I2P nudity | detected nudity: 206 -> 235 | Hurt |
| add L_erase to L_attn + L_pr (ablation) | Z-Erase · I2P nudity | detected nudity: 314 -> 161 | Improved |
| add L_attn to L_erase + L_pr (full model) (ablation) | Z-Erase · I2P nudity | detected nudity: 235 -> 161 | Improved |
| add L_pr to L_erase + L_attn (full model) (ablation) | Z-Erase · I2P nudity | FID: 30.04 -> 26.46 | Improved |
| larger tolerance epsilon (ablation) | Z-Erase · I2P nudity | erasure vs preservation | Mixed |
| vary step size beta (ablation) | Z-Erase · I2P nudity | erasure/utility | No clear effect |
| apply stream-disentangled framework (text-only LoRA) (ablation) | naive fine-tuning · Z-Image Turbo | generation quality | 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.
- EraseAnything — EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers (ICML 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
- EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers (ICML 2025 · similarity 0.93)
- Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters (CVPR 2025 · similarity 0.91)
- One-dimensional Adapter to Rule Them All: Concepts Diffusion Models and Erasing Applications (CVPR 2024 · similarity 0.89)
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Memories of Forgotten Concepts (CVPR 2025 · similarity 0.89)
- Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptation (CVPR 2025 · similarity 0.89)
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.
