Cross-Audio Talking Head Generation on HDTF, CelebV-HQ, and CelebV-Text 100 cross-audio pairs
5.89FIDFlashLips
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FlashLipsArchitecture=Transformer2025.12 | 5.89 | 29.4 | 73.84 | 0.682 | 0.37 | 0.81 | |
| FlashLipsArchitecture=UNet2025.12 | 6.23 | 33.57 | 73.58 | 0.681 | 0.38 | 0.81 | |
| KeySync2025.12 | 6.81 | 37.55 | 74.13 | 0.676 | 0.36 | 0.79 | |
| LatentSync2025.12 | 7.69 | 46.08 | 72.52 | 0.68 | 0.33 | 0.84 | |
| IP-LAP2025.12 | 9.05 | 43.13 | 69.49 | 0.67 | 0.25 | 0.81 | |
| Diff2Lip2025.12 | 9.55 | 80.43 | 69.01 | 0.667 | 0.31 | 0.76 | |
| TalkLip2025.12 | 17.64 | 134.7 | 64.62 | 0.645 | 0.21 | 0.74 | |
| DiffDub2025.12 | 18.31 | 123.34 | 57.91 | 0.668 | 0.3 | 0.59 |