Safety Alignment on VSA Violence (visual synonym attack prompts)
0ASRAEGIS
Evaluation Results
| Method | Links | |
|---|---|---|
| AEGISCategory=AEGIS (Ours), Base Model=SD 1.42026.07 | 0 | |
| AEGISCategory=AEGIS (Ours), Base Model=SD 2.12026.07 | 4 | |
| SafeClipCategory=Input-Space2026.07 | 5 | |
| SLD-StrongCategory=Trigger-Specific2026.07 | 5 | |
| SAFREECategory=Input-Space2026.07 | 8 | |
| SafeGuiderCategory=Input-Space2026.07 | 11 | |
| CADCategory=Feature Pruning2026.07 | 12 | |
| UCECategory=Trigger-Specific2026.07 | 15 | |
| RecelerCategory=Trigger-Specific2026.07 | 17 | |
| SDIDCategory=Trigger-Specific2026.07 | 21 | |
| ESD-uCategory=Trigger-Specific2026.07 | 30 | |
| SLD-MediumCategory=Trigger-Specific2026.07 | 32 | |
| Diff-QuickFixCategory=Trigger-Specific2026.07 | 45 | |
| SD 2.1Category=No Alignment2026.07 | 51 | |
| SLD-WeakCategory=Trigger-Specific2026.07 | 54 | |
| SD 1.4Category=No Alignment2026.07 | 65 |