List your company
ResearchFormerly WizWand

Defense against adversarial prompts on MMA (test)

0.4Attack Success RateDES
-2.541217.311937.16557.0181Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DESBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.070.4
SAGEBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.070.5
AdvUnlearnBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.072.1
SalUnBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.073.2
GLoCEBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.073.8
GuardT2IBase Model=SDv1.5, Evaluation Classifier=filtering accuracy2026.077.54
ESDBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.078.5
Latent GuardBase Model=SDv1.5, Evaluation Classifier=filtering accuracy2026.0712.64
OpenAIBase Model=SDv1.5, Evaluation Classifier=filtering accuracy2026.0724.84
MicrosoftBase Model=SDv1.5, Evaluation Classifier=filtering accuracy2026.0726.25
UCEBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.0733.3
SAFREEBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.0741.2
SLD-strongBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.0759.2
SPMBase Model=SDv1.5, Evaluation Classifier=NudeNet2026.0765.05
SDv1.5Base Model=SDv1.5, Evaluation Classifier=NudeNet2026.0773.93