Video Generation on 182 motion prompts (test)
26.98FIDWan-FT + LIC (Ours)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Wan-FT + LIC (Ours)Backbone=Wan2.2 I2V, Adaptation=LoRA fine-tuned, Loss Function=flow-matching loss + Identity Consistency Loss (LIC), Lambda=0.52026.06 | 26.98 | 10.14 | 8.35 | 6.68 | 7.51 | |
| Wan-FTBackbone=Wan2.2 I2V, Adaptation=LoRA fine-tuned, Loss Function=flow-matching loss2026.06 | 27.95 | 10.63 | 8.3 | 6.67 | 7.48 | |
| Wan-BaseBackbone=Wan2.2 I2V, Adaptation=off-the-shelf, no adaptation2026.06 | 66.65 | 21.5 | 8.02 | 6.48 | 7.37 |