Video Jailbreaking on COCO 2017 (test)
93Attack Success RateRunaway Evil
Evaluation Results
| Method | Links | |
|---|---|---|
| Runaway EvilVideo Generation Model=Wan, Evaluation Model (MLLM)=GEMMA2025.12 | 93 | |
| Runaway EvilVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=LLAVA2025.12 | 92 | |
| Runaway EvilVideo Generation Model=Opensora, Evaluation Model (MLLM)=LLAVA2025.12 | 92 | |
| Runaway EvilVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=GEMMA2025.12 | 89.5 | |
| Runaway EvilVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=LLAVA2025.12 | 89 | |
| Runaway EvilVideo Generation Model=Opensora, Evaluation Model (MLLM)=GEMMA2025.12 | 88 | |
| Runaway EvilVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=QWEN2025.12 | 86.5 | |
| Runaway EvilVideo Generation Model=Wan, Evaluation Model (MLLM)=QWEN2025.12 | 86 | |
| Runaway EvilVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=GEMMA2025.12 | 85.5 | |
| Runaway EvilVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=QWEN2025.12 | 85.5 | |
| Runaway EvilVideo Generation Model=Opensora, Evaluation Model (MLLM)=QWEN2025.12 | 83.5 | |
| Runaway EvilVideo Generation Model=Wan, Evaluation Model (MLLM)=LLAVA2025.12 | 81 | |
| PGJVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=LLAVA2025.12 | 47 | |
| PGJVideo Generation Model=Opensora, Evaluation Model (MLLM)=LLAVA2025.12 | 46.5 | |
| DACAVideo Generation Model=Opensora, Evaluation Model (MLLM)=LLAVA2025.12 | 46 | |
| PGJVideo Generation Model=Opensora, Evaluation Model (MLLM)=QWEN2025.12 | 45 | |
| DACAVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=LLAVA2025.12 | 44 | |
| PGJVideo Generation Model=Wan, Evaluation Model (MLLM)=LLAVA2025.12 | 42.5 | |
| PGJVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=LLAVA2025.12 | 41.5 | |
| PGJVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=QWEN2025.12 | 41.5 | |
| DACAVideo Generation Model=Wan, Evaluation Model (MLLM)=LLAVA2025.12 | 41 | |
| PGJVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=QWEN2025.12 | 37 | |
| DACAVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=LLAVA2025.12 | 36.5 | |
| SneakyVideo Generation Model=Opensora, Evaluation Model (MLLM)=LLAVA2025.12 | 35 | |
| PGJVideo Generation Model=Opensora, Evaluation Model (MLLM)=GEMMA2025.12 | 32.5 | |
| SneakyVideo Generation Model=Wan, Evaluation Model (MLLM)=LLAVA2025.12 | 31.5 | |
| PGJVideo Generation Model=Wan, Evaluation Model (MLLM)=QWEN2025.12 | 31.5 | |
| PGJVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=GEMMA2025.12 | 31 | |
| SneakyVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=LLAVA2025.12 | 30.5 | |
| SneakyVideo Generation Model=Opensora, Evaluation Model (MLLM)=QWEN2025.12 | 28.5 | |
| DACAVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=QWEN2025.12 | 27 | |
| SneakyVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=QWEN2025.12 | 26 | |
| PGJVideo Generation Model=Wan, Evaluation Model (MLLM)=GEMMA2025.12 | 24.5 | |
| DACAVideo Generation Model=Opensora, Evaluation Model (MLLM)=QWEN2025.12 | 24 | |
| SneakyVideo Generation Model=Wan, Evaluation Model (MLLM)=QWEN2025.12 | 23 | |
| SneakyVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=LLAVA2025.12 | 23 | |
| SneakyVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=QWEN2025.12 | 21.5 | |
| PGJVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=GEMMA2025.12 | 21.5 | |
| DACAVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=GEMMA2025.12 | 21 | |
| DACAVideo Generation Model=Opensora, Evaluation Model (MLLM)=GEMMA2025.12 | 21 | |
| DACAVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=GEMMA2025.12 | 21 | |
| SneakyVideo Generation Model=Opensora, Evaluation Model (MLLM)=GEMMA2025.12 | 16.5 | |
| DACAVideo Generation Model=Wan, Evaluation Model (MLLM)=GEMMA2025.12 | 16.5 | |
| DACAVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=QWEN2025.12 | 16.5 | |
| DACAVideo Generation Model=Wan, Evaluation Model (MLLM)=QWEN2025.12 | 11 | |
| SneakyVideo Generation Model=Wan, Evaluation Model (MLLM)=GEMMA2025.12 | 10 | |
| SneakyVideo Generation Model=Cogvideo, Evaluation Model (MLLM)=GEMMA2025.12 | 10 | |
| SneakyVideo Generation Model=DynamiCrafter, Evaluation Model (MLLM)=GEMMA2025.12 | 6.5 |