Deepfake Detection on OmniFake Cross-Architecture
94Accuracy (FF++)DDA
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DDA2026.03 | 94 | 80.9 | 97.1 | 97.1 | |
| VIGILBackbone=Qwen3-VL-8B2026.03 | 92.8 | 82.1 | 97.9 | 98.1 | |
| Veritas2026.03 | 92.5 | 74.2 | 97.8 | 96.5 | |
| D^32026.03 | 91.3 | 64.5 | 96.7 | 96.2 | |
| VIGIL Cold-StartBackbone=Qwen3-VL-8B, mode=Cold-Start2026.03 | 90.3 | 80.3 | 94.1 | 94.9 | |
| UnivFD2026.03 | 88 | 59.7 | 88.8 | 95.1 | |
| NPR2026.03 | 84.9 | 56.3 | 84.5 | 91.1 | |
| AIDE2026.03 | 82.4 | 65.3 | 93.8 | 93.5 | |
| Co-SPY2026.03 | 82.4 | 65 | 96.5 | 96.4 | |
| Gemini-3-Pro2026.03 | 81.5 | 62.3 | 88.7 | 85.6 | |
| C2P-CLIP2026.03 | 80.8 | 51.1 | 84.1 | 73.8 | |
| FakeVLM2026.03 | 79.3 | 69.4 | 81.5 | 80 | |
| GPT-5.22026.03 | 77.4 | 59.8 | 85.2 | 84.7 | |
| SAFE2026.03 | 74.9 | 56.4 | 72 | 74.3 | |
| Kimi-K2.52026.03 | 66.3 | 67.9 | 80.7 | 85.1 | |
| GLM-4.6V2026.03 | 54.1 | 53.7 | 53 | 74.3 | |
| Qwen3-VL-8Bzero-shot=true2026.03 | 50.7 | 53.6 | 60.6 | 62.4 |