Video Editing on ShutterStock (30 unseen videos)
31.3Prompt Consistency (P.C.)TokenFlow
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TokenFlowTime (Inference efficiency)=150s2023.12 | 31.3 | 98.7 | 5.4 | 15.8 | |
| ControlNet per frameTime (Inference efficiency)=50s2023.12 | 30.4 | 94.2 | 25.9 | 45.1 | |
| Text2Video-zeroTime (Inference efficiency)=60s2023.12 | 30.4 | 98.1 | 16 | 33 | |
| MaskINTTime (Inference efficiency)=22s2023.12 | 30.4 | 97.1 | 8.6 | 22.3 | |
| ControlVideo-edgeTime (Inference efficiency)=120s2023.12 | 30.3 | 98.6 | 7.4 | 20.5 | |
| Tune-a-VideoTime (Inference efficiency)=20min2023.12 | 29.2 | 97.9 | 13.4 | 34.9 |