Deepfake Detection on OmniFake Cross-Task
92Accuracy (BrushNet)VIGIL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VIGILBackbone=Qwen3-VL-8B2026.03 | 92 | 96.4 | 91.1 | 88.9 | 81.7 | |
| Co-SPY2026.03 | 88.4 | 75.5 | 71.4 | 70.1 | 68.3 | |
| DDA2026.03 | 86.7 | 79.3 | 83.5 | 81.9 | 79.2 | |
| VIGIL Cold-StartBackbone=Qwen3-VL-8B, mode=Cold-Start2026.03 | 85.4 | 86.5 | 83.3 | 84.2 | 85.6 | |
| Veritas2026.03 | 84 | 80.5 | 83.2 | 78.3 | 73.8 | |
| D^32026.03 | 73.8 | 62.1 | 60.6 | 62.2 | 57.5 | |
| AIDE2026.03 | 73.5 | 76 | 66.9 | 62.8 | 64.5 | |
| FakeVLM2026.03 | 72.8 | 74.8 | 76.5 | 74.2 | 72.3 | |
| UnivFD2026.03 | 72 | 62.1 | 64.1 | 58.3 | 66.8 | |
| Gemini-3-Pro2026.03 | 68.2 | 72.9 | 56.8 | 56.4 | 57.8 | |
| NPR2026.03 | 67.6 | 59 | 68.1 | 64.6 | 74.5 | |
| Kimi-K2.52026.03 | 65.2 | 72.1 | 53.4 | 57.5 | 61.8 | |
| C2P-CLIP2026.03 | 63.4 | 70.4 | 68.8 | 65.2 | 68.8 | |
| GPT-5.22026.03 | 63.4 | 66.9 | 56.9 | 56.9 | 54.3 | |
| GLM-4.6V2026.03 | 54.1 | 65 | 49.7 | 50.4 | 52 | |
| SAFE2026.03 | 50.4 | 59.8 | 74.5 | 62.1 | 75.5 | |
| Qwen3-VL-8Bzero-shot=true2026.03 | 48.9 | 61.2 | 48.4 | 51.2 | 53.5 |