Video Generation on VBench (Imaging, Aesthetic, Motion, Consistency, and Dynamic Metrics)
71.36Imaging QualityOurs (Joint)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Ours (Joint)Latency (s)=11.78, Speedup vs. 50-step=27.66×, ∆ vs. Distill=1.09×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 71.36 | 66.45 | 98.17 | 80.56 | 95.63 | 95.8 | 43.68 | 25.86 | |
| LightX2VSpeedup vs. 50-step=47.50×*, ∆ vs. Distill=1.90 ×*, Model Scale=Wan-14B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 71.15 | 68.6 | 98.45 | 58.8 | 96.7 | 96.26 | 47.31 | 26.17 | |
| Ours (Distill)Latency (s)=12.93, Speedup vs. 50-step=25.20×, ∆ vs. Distill=1.00×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 70.67 | 67.81 | 98.61 | 68.06 | 96.26 | 96.35 | 44.11 | 25.88 | |
| TurboDiffusionSpeedup vs. 50-step=93.00×*, ∆ vs. Distill=3.72×*, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 69.47 | 64.62 | 97.8 | 86.11 | 94.21 | 94.34 | 43.31 | 25.46 | |
| Wan-14B (50 step)Latency (s)=1222.16, Speedup vs. 50-step=1.00×, Model Scale=Wan-14B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 69.43 | 66.07 | 98.3 | 65.46 | 98.09 | 97.52 | 45.75 | 25.91 | |
| Ours (Joint)Latency (s)=40.72, Speedup vs. 50-step=30.01×, ∆ vs. Distill=1.20×, Model Scale=Wan-14B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 69.31 | 69.36 | 98.9 | 62.17 | 96.98 | 97.15 | 45.57 | 26.12 | |
| Static → DistillLatency (s)=12.92, Speedup vs. 50-step=25.22×, ∆ vs. Distill=1.01×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 69.05 | 66.49 | 98.72 | 56.94 | 95.92 | 97.19 | 45.06 | 25.57 | |
| Distill → DynamicLatency (s)=12.48, Speedup vs. 50-step=26.11×, ∆ vs. Distill=1.04×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 68.23 | 55.66 | 98.32 | 11.11 | 92.59 | 93.43 | 32.85 | 22.4 | |
| Dynamic → DistillLatency (s)=12.14, Speedup vs. 50-step=26.84×, ∆ vs. Distill=1.07×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 67.57 | 64.37 | 98.74 | 65.28 | 95.68 | 95.31 | 40.84 | 25.24 | |
| Wan (50 step)Latency (s)=325.84, Speedup vs. 50-step=1.00×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 67.01 | 65.46 | 98.52 | 65.19 | 97.93 | 97.56 | 45.06 | 25.57 | |
| Distill → StaticLatency (s)=12.90, Speedup vs. 50-step=25.26×, ∆ vs. Distill=1.01×, Model Scale=Wan-1.3B, Resolution=480 × 832, FPS=16 fps, Duration=5 s2026.07 | 58.21 | 63.48 | 98.39 | 26.39 | 95.11 | 94.58 | 39.1 | 24.88 |