Research map / Video / Video generation
PhysRAG: Enhancing Physics-Awareness in Video Generation via Retrieval-Augmented Generation
Kexu Cheng, Zicheng Liu, Mingju Gao, Chunhe Song, Hao Tang
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 PhysRAG retrieval and query injection | Wan2.2-5B · PhyGenBench | average score: 0.54 -> 0.58 | Improved |
| add PhysRAG | Wan2.2-5B · VBench | Low-Avg: 62.10% -> 65.48% | Improved |
| add PhysRAG | Wan2.2-5B · VBench | High-Avg: 81.18% -> 82.88% | Improved |
| add PhysRAG | Wan2.2-5B · VBench | Spatial Rel.: 78.91% -> 77.33% | Hurt |
| use learnable-query injection instead of concatenation (ablation) | PhysRAG · PhyGenBench | average score: 0.540 -> 0.578 | Improved |
| use learnable queries instead of direct cross attention (ablation) | PhysRAG · PhyGenBench | average score: 0.553 -> 0.578 | Improved |
| add retrieval to curated SFT (ablation) | Wan2.2-5B · PhyGenBench | average score: 0.546 -> 0.578 | Improved |
| train backbone jointly with queries instead of freezing backbone (ablation) | PhysRAG · PhyGenBench | average score: 0.551 -> 0.578 | Improved |
| add retrieval with only frozen-backbone query training over SFT (ablation) | Wan2.2-5B · PhyGenBench | average score: 0.546 -> 0.551 | No clear effect |
| filter training data with two-stage pipeline for SFT (ablation) | Wan2.2-5B · PhyGenBench | average score: 0.539 -> 0.546 | Improved |
| filter training data when using PhysRAG (ablation) | PhysRAG · PhyGenBench | average score: 0.544 -> 0.578 | Improved |
| inject at layers 0,1,2 instead of layer 0 only (ablation) | PhysRAG · PhyGenBench | average score: 0.478 -> 0.578 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CogVideoX 5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 2 of 2 shared results
- Lavie — LaVie: High-Quality Video Generation with Cascaded Latent Diffusion Models (IJCV 2024) · 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
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.91)
- MotionRAG: Motion Retrieval-Augmented Image-to-Video Generation (NeurIPS 2025 · similarity 0.88)
- Hierarchical Fine-grained Preference Optimization for Physically Plausible Video Generation (NeurIPS 2025 · similarity 0.87)
- WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025 · similarity 0.87)
- PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation (CVPR 2025 · similarity 0.87)
- How Far Is Video Generation from World Model: A Physical Law Perspective (ICML 2025 · similarity 0.87)
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.
