VBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
VBench Out-of-domain
83.98Semantic Fidelity
3
VBench2
91.64Anatomy
3
VBench long-video (60s)
97.86Temporal Flickering
3
VBench
94.29Background Consistency
3
VBench Wan2.1 1.3B extrapolation, 141x448x832 1.0 (test)
64.27AQ
3
VBench Wan2.1 1.3B extrapolation, 61x720x1280 1.0 (test)
65.49Aesthetic Quality
3
VBench Long (5-minute)
0.98Subject Consistency
3
VBench++ Long 1-minute AR video synthesis
98.58Subject Consistency
3
VBench I2V (test)
0.983Subject Fidelity Score
3
VBench 2.0 (test)
0.984Motion Smoothness
2
VBench Official Evaluation
0.703Imaging Quality
2
VBench
98.2Motion Smoothness Score
2
VBench 30-second WorldScore-sampled trajectories (test)
98.43Flicker Score
2
VBench-I2V 63-sample (test)
6.428Total Score
2
VBench Multiple Objects
76.37Temporal GRiT
2
VBench Multiple Object
76.37Multi-Object (GriT) Score
2
VBench Ghibli Studio Style
98.28Subject Consistency
2
VBench-Long 1 hour
63.05Aesthetic Quality
2
VBench Wan2.2-TI2V-5B
0.584Total Score
2
VBench 81 frames, 256x256
35.9UR
2
VBench
0.427CSD Score
2
VBench Low-Resolution
84.85Quality Score
2
VBench Human Preference 2.0
43.3Color Preference
2
VBench Self-Forcing
0.846Quality Score
2
VBench 200-sequence benchmark (held-out subset)
—Primary metric
0