Identity-Preserving Talking Head Generation on TalkVid
0.599Speaker SimilarityID-LoRA (Ours)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ID-LoRA (Ours)Video Backbone=LTX-22026.03 | 0.599 | 0.772 | 10.62 | 3.09 | 0.385 | 5.4 | |
| ID-LoRATransfer Setting=Celeb -> TalkVid, Video Backbone=LTX-22026.03 | 0.595 | 0.767 | 10.32 | 3.12 | 0.412 | 6.5 | |
| CosyVoice 3.0 + WAN2.2Video Backbone=WAN2.22026.03 | 0.579 | 0.77 | 12.34 | 1.2 | 0.315 | 22.3 | |
| Kling 2.6 ProCommercial Model=true2026.03 | 0.506 | 0.754 | 11.59 | 2.4 | 0.326 | 4 | |
| ElevenLabs + WAN2.2Video Backbone=WAN2.22026.03 | 0.491 | 0.772 | 12.42 | 1.31 | 0.319 | 6.4 | |
| VoiceCraft + WAN2.2Video Backbone=WAN2.22026.03 | 0.457 | 0.773 | 12.15 | 1.22 | 0.407 | 17.1 |