Text-to-Video Generation on VBench (Quality and Efficiency Metrics)
77.17VBench Semantic ScoreCalibAtt
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CalibAttResolution=480p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 77.17 | 82.29 | 81.26 | 70.9 | 11.2 | 1.29 | — | |
| RadialAttentionResolution=480p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 77.02 | 82.46 | 81.37 | 48.4 | 12.7 | 1.14 | — | |
| SpargeAttentionResolution=480p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 76.4 | 82.08 | 80.94 | 49.6 | 13.5 | 1.07 | — | |
| SVG2Resolution=480p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 76.19 | 82.64 | 81.35 | 62.7 | 24 | 0.6 | — | |
| Dense (FA3)Resolution=480p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 76.06 | 82.28 | 81.04 | 0 | 14.5 | 1 | — | |
| CalibAttResolution=720p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 74.04 | 78.8 | 77.84 | 73.9 | 30.6 | 1.57 | — | |
| SpargeAttentionResolution=720p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 73.47 | 78.75 | 77.69 | 49.9 | 39.9 | 1.21 | — | |
| RadialAttentionResolution=720p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 73.24 | 78.68 | 77.6 | 52.5 | 37 | 1.3 | — | |
| SVG2Resolution=720p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 73.19 | 78.83 | 77.7 | 61.9 | 47.3 | 1.02 | — | |
| Dense (FA3)Resolution=720p, Num Frames=81, Sampling Steps=4, Backbone=LightX2V2026.03 | 73.17 | 78.74 | 77.63 | 0 | 48.3 | 1 | — | |
| 50% stepsMACs (T)=21.933, Speedup=1.86×, Sampling steps=20, Base model=Wan2.1-1.3B2026.04 | — | — | — | — | — | — | 43.14 | |
| DiffSparseMACs (T)=18.124, Speedup=2.05×, Sampling steps=20, Base model=Wan2.1-1.3B2026.04 | — | — | — | — | — | — | 43.83 | |
| DuCaMACs (T)=20.332, Speedup=1.69×, Sampling steps=20, Base model=Wan2.1-1.3B, Sparsity ratio (R)=54%2026.04 | — | — | — | — | — | — | 43.56 | |
| DuCaMACs (T)=18.124, Speedup=1.68×, Sampling steps=20, Base model=Wan2.1-1.3B, Sparsity ratio (R)=59%2026.04 | — | — | — | — | — | — | 43.3 | |
| Lumos-1Params=1.5B, Time (8×A100 GPU-Days)=85.42025.07 | — | — | — | — | — | — | 78.17 | |
| Open-Sora Plan v1.2Params=2.7B + 13B, Time (8×A100 GPU-Days)=> 5922025.07 | — | — | — | — | — | — | 75.98 | |
| Wan 2.1 - 1.3BMACs (T)=43.866, Speedup=1.00×, Sampling steps=20, Base model=Wan2.1-1.3B2026.04 | — | — | — | — | — | — | 43.82 |