Research map / Video / Video generation
Predictive Structure Improves Video Diffusion Dynamics
Mi Luo, Yujia Chen, Alex Dimakis, Kristen Grauman, Wen-Sheng Chu, Du Tran
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 |
|---|---|---|---|
| add PDA and GRPO post-training (LDO) | CogVideoX-5B · VideoPhy | Physical Commonsense: 32.0 -> 45.6 | Improved |
| add LDO post-training | CogVideoX-5B · VideoPhy | Semantic Adherence: 69.2 -> 71.8 | Improved |
| add LDO post-training | CogVideoX-5B · VideoPhy (Fluid-Fluid) | Physical Commonsense: 60.0 -> 80.0 | Improved |
| add LDO post-training | Open-Sora · PisaBench (real) | Trajectory L2: 0.079 -> 0.070 | Improved |
| add LDO post-training | Open-Sora · PisaBench (simulated) | Chamfer Distance: 0.076 -> 0.064 | Improved |
| add LDO post-training | Open-Sora · PisaBench (simulated) | IoU: 0.155 -> 0.154 | No clear effect |
| add PDA (Gram-matrix alignment) alone (ablation) | CogVideoX-5B · VideoPhy | Physical Commonsense: 32.0 -> 39.2 | Improved |
| add GRPO with V-JEPA reward alone (ablation) | CogVideoX-5B · VideoPhy | Physical Commonsense: 32.0 -> 40.4 | Improved |
| add GRPO to PDA (ablation) | LDO · VideoPhy | Physical Commonsense: 39.2 -> 45.6 | Improved |
| add PDA to GRPO (ablation) | LDO · VideoPhy | Physical Commonsense: 40.4 -> 45.6 | Improved |
| add GRPO to PDA (ablation) | LDO · VideoPhy | Semantic Adherence: 72.4 -> 71.8 | Hurt |
| add PDA to GRPO (ablation) | LDO · VideoPhy | Semantic Adherence: 70.6 -> 71.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- VideoREPA-5B — VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 2025) · better on 5 of 5 shared results
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 5 of 5 shared results
- WISA — WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.90)
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · similarity 0.90)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.89)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.89)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.89)
- AdaState: Self-Evolving Anchors for Streaming Video Generation (NeurIPS 2026 · similarity 0.89)
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.
