Research map / Video / Video generation
VDOT: Efficient Unified Video Creation via Optimal Transport Distillation
Yutong Wang, Haiyu Zhang, Tianfan Xue, Yu Qiao, Yaohui Wang, Chang Xu, Xinyuan Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| distill with 4-step VDOT instead of 100-step VACE-Wan-14B | VACE-Wan2.1-14B · UVCBench Depth | VBench normalized average: 77.34% -> 78.50% | Improved |
| distill with 4-step VDOT instead of 100-step VACE-Wan-14B | VACE-Wan2.1-14B · UVCBench Extension | VBench normalized average: 77.10% -> 80.53% | Improved |
| distill with 4-step VDOT instead of 100-step VACE-Wan-14B | VACE-Wan2.1-14B · UVCBench R2V | VBench normalized average: 82.54% -> 81.32% | Hurt |
| distill with 4-step VDOT instead of 100-step VACE-Wan-14B | VACE-Wan2.1-14B · UVCBench Flow and Grey | VBench normalized average: Flow 80.35% -> 80.18%; Grey 72.44% -> 72.13% | No clear effect |
| add OTD loss to DMD+GAN distillation (ablation) | VDOT · UVCBench Depth | VBench normalized average: 78.05% -> 78.50% | Improved |
| add OTD loss to DMD+GAN distillation (ablation) | VDOT · UVCBench Pose and Flow | VBench normalized average: Pose 80.79% -> 80.54%; Flow 80.15% -> 80.18% | No clear effect |
| add OTD loss to DMD+GAN distillation (ablation) | VDOT · UVCBench R2V | VBench normalized average: 78.40% -> 81.32% | Improved |
| add GAN loss to DMD+OTD distillation (ablation) | VDOT · UVCBench R2V | VBench normalized average: 77.66% -> 81.32% | Improved |
| add GAN loss to DMD+OTD distillation (ablation) | VDOT · UVCBench Outpaint | VBench normalized average: 76.44% -> 76.54% | Improved |
| add OTD and GAN losses to DMD-only Self-Forcing distillation (ablation) | VDOT · UVCBench Depth | VBench normalized average: 76.89% -> 78.50% | Improved |
| add AccWanInit (init Wan blocks with few-step Wan) (ablation) | VDOT · UVCBench Extension | VBench normalized average: 79.76% -> 80.53% | Improved |
| add AccWanInit (init Wan blocks with few-step Wan) (ablation) | VDOT · UVCBench Depth, Pose, Flow, Grey | VBench normalized average: Depth 77.15% -> 78.50%; Scribble 77.19% -> 76.77% | Mixed |
Closest papers by meaning
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.89)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.89)
- Q-VDiT: Towards Accurate Quantization and Distillation of Video-Generation Diffusion Transformers (ICML 2025 · similarity 0.89)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.88)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
