List your company
ResearchFormerly WizWand

Text-to-Video Generation on VBench (Aesthetic, Imaging, Consistency, and Motion Metrics)

0.6784Aesthetic QualityOur SFT Baseline + DPO
0.570240.598320.62640.65448Jun 29, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Our SFT Baseline + DPOBase Model=Proprietary 4.5B SFT model, Optimization=DPO, Resolution=720p2026.060.67840.73970.26770.9840.9637
Our SFT Baseline + Shell-LCCBase Model=Proprietary 4.5B SFT model, Optimization=Shell-LCC, Resolution=720p2026.060.67350.76310.2670.990.9682
Our SFT BaselineBase Model=Proprietary 4.5B SFT model, Resolution=720p2026.060.67240.75090.26370.98910.968
UltraWan-T2V-1.3B + Shell-LCCBase Model=UltraWan-T2V-1.3B, Optimization=Shell-LCC, Resolution=720p2026.060.62990.73960.22360.99180.9658
Wan-T2V-1.3B + Shell-LCCBase Model=Wan-T2V-1.3B, Optimization=Shell-LCC, Resolution=720p2026.060.62530.70370.22920.98710.9715
Wan-T2V-1.3BBase Model=Wan-T2V-1.3B, Resolution=720p2026.060.62440.66290.22710.98480.9654
UltraWan-T2V-1.3BBase Model=UltraWan-T2V-1.3B, Resolution=720p2026.060.57440.67560.22880.96570.9225