Research map / Image generation / Text-to-image diffusion
TINA: Text-Free Inversion Attack for Unlearned Text-to-Image Diffusion Models
Qianlong Xiang, Miao Zhang, Haoyu Zhang, Kun Wang, Junhui Hou, Liqiang Nie
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 text-free inversion attack instead of UDA | unlearned SD v1.4 (AdvUnlearn) · Nudity erasure | ASR: 23.24 -> 78.87 | Improved |
| use TINA instead of UDA | unlearned SD v1.4 (SalUn) · Nudity erasure | ASR: 13.38 -> 71.13 | Improved |
| use TINA instead of UDA | unlearned SD v1.4 (FMN) · Nudity erasure | ASR: 97.89 -> 97.89 | No clear effect |
| use TINA instead of UDA | unlearned SD v1.4 (ESD) · Style erasure (Van Gogh) | ASR: 32.0 -> 70.0 | Improved |
| use TINA instead of UDA | unlearned SD v1.4 (AC, SPM) · Style erasure (Van Gogh) | ASR: 77.0 -> 74.0 (AC); 88.0 -> 80.0 (SPM) | Hurt |
| use TINA instead of CCE | unlearned SD v1.4 (Scissorhands) · Object erasure (tench) | ASR: 0.0 -> 78.0 | Improved |
| use TINA instead of CCE | unlearned SD v1.4 (SalUn) · Object erasure (tench) | ASR: 58.0 -> 72.0 | Improved |
| optimize inversion at each step vs standard null-text DDIM inversion (ablation) | TINA · Nudity erasure | reconstruction of target concept | Improved |
Closest papers by meaning
- Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- STEREO: A Two-Stage Framework for Adversarially Robust Concept Erasing from Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.89)
- Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation (NeurIPS 2024 · similarity 0.88)
- Precise, Fast, and Low-cost Concept Erasure in Value Space: Orthogonal Complement Matters (CVPR 2025 · similarity 0.88)
- Localized Concept Erasure for Text-to-Image Diffusion Models Using Training-Free Gated Low-Rank Adaptation (CVPR 2025 · similarity 0.88)
- ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning (ICLR 2025 · 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.
