ControlVideo: Training-free Controllable Text-to-Video Generation
About
Text-driven diffusion models have unlocked unprecedented abilities in image generation, whereas their video counterpart still lags behind due to the excessive training cost of temporal modeling. Besides the training burden, the generated videos also suffer from appearance inconsistency and structural flickers, especially in long video synthesis. To address these challenges, we design a \emph{training-free} framework called \textbf{ControlVideo} to enable natural and efficient text-to-video generation. ControlVideo, adapted from ControlNet, leverages coarsely structural consistency from input motion sequences, and introduces three modules to improve video generation. Firstly, to ensure appearance coherence between frames, ControlVideo adds fully cross-frame interaction in self-attention modules. Secondly, to mitigate the flicker effect, it introduces an interleaved-frame smoother that employs frame interpolation on alternated frames. Finally, to produce long videos efficiently, it utilizes a hierarchical sampler that separately synthesizes each short clip with holistic coherency. Empowered with these modules, ControlVideo outperforms the state-of-the-arts on extensive motion-prompt pairs quantitatively and qualitatively. Notably, thanks to the efficient designs, it generates both short and long videos within several minutes using one NVIDIA 2080Ti. Code is available at https://github.com/YBYBZhang/ControlVideo.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Video Editing | 20 in-the-wild cases | CLIP score26.87 | 8 | |
| Video Motion Editing | User Study 20 video cases | M-A Score94.1 | 7 | |
| Video Editing | DAVIS (40 selected object-centric videos) | Prompt Consistency (P.C.)31.4 | 6 | |
| Video Editing | ShutterStock (30 unseen videos) | Prompt Consistency (P.C.)30.3 | 6 | |
| Depth-conditioned Video Generation | UVCBench | Aesthetic Quality64.5 | 5 | |
| Accident Video Generation | MM-AU 1.0 (test) | CLIP S22.51 | 5 | |
| Pose-conditioned Video Generation | UVCBench | Aesthetic Quality63.32 | 5 | |
| Video Editing | User Study | Preference Rate (Ours)84.1 | 5 | |
| Scribble-conditioned Video Generation | UVCBench | Aesthetic Quality54.79 | 4 | |
| Video-to-Video Translation | 23 videos (test) | Frame Accuracy93.2 | 4 |