Object Motion Control on T2V-CompBench Object-only motion binding
4.75VLM Score (SA)Wan + SIFT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Wan + SIFTModel Setting=Wan + SIFT (Ours)2026.06 | 4.75 | 4.72 | 3.98 | 3.84 | |
| CogVideoX + SIFTModel Setting=CogVideoX + SIFT (Ours)2026.06 | 4.38 | 4.25 | 3.75 | 3.68 | |
| Wan + SFTModel Setting=Wan + SFT2026.06 | 4.32 | 4.3 | 2.91 | 2.43 | |
| CogVideoX + SFTModel Setting=CogVideoX + SFT2026.06 | 4.25 | 3.69 | 3.18 | 2.86 | |
| CogVideoX + VideoREPAModel Setting=CogVideoX + VideoREPA2026.06 | 4 | 3.5 | 2.2 | 2.6 | |
| WanModel Setting=Wan2026.06 | 3.8 | 3.6 | 2.91 | 2.75 | |
| CogVideoXModel Setting=CogVideoX2026.06 | 3.75 | 2.81 | 2.73 | 2.39 |