Video Generation on VBench Self-Forcing Wan 2.1-1.3B (test)
0.846Quality ScoreDense Attention
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Dense AttentionGeneration Steps=4-step, Model Type=Distilled2026.02 | 0.846 | 0.8 | 0.837 | |
| Dense AttentionGeneration Steps=50-step, Model Type=Base2026.02 | 0.846 | 0.81 | 0.839 | |
| MonarchRTGeneration Steps=4-step, Model Type=Distilled, Attention Sparsity=95%2026.02 | 0.842 | 0.788 | 0.832 | |
| MonarchRTGeneration Steps=50-step, Model Type=Base, Attention Sparsity=95%2026.02 | 0.841 | 0.812 | 0.835 | |
| VSAGeneration Steps=4-step, Model Type=Distilled, Attention Sparsity=85%2026.02 | 0.828 | 0.793 | 0.821 | |
| VSAGeneration Steps=50-step, Model Type=Base, Attention Sparsity=85%2026.02 | 0.827 | 0.785 | 0.819 |