Research map / Video / Video generation
MoReGen: Multi-Agent Motion-Reasoning Engine for Code-based Text-to-Video Synthesis
Xiangyu Bai, He Liang, Bishoy Galoaa, Utsav Nandi, Shayda Moezzi, Yuhang He, Sarah Ostadabbas
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 |
|---|---|---|---|
| supervised fine-tune the text-parser/coder LLM on 1200 text-specification pairs (Qwen2.5-Coder-14B vs GPT-5 without SFT) (ablation) | MoReGen · MoReSet evaluation set | DTW: 15.47 -> 8.93 | Improved |
| add one iteration of evaluator feedback to untuned Qwen2.5-Coder-14B (ablation) | Qwen2.5-Coder-14B · MoReSet evaluation set | code executability | Improved |
| add one iteration of evaluator feedback to GPT-5 coder (ablation) | GPT-5 coder · MoReSet evaluation set | DTW: 15.47 -> 14.13 | Improved |
| add one iteration of evaluator feedback to GPT-5 coder (ablation) | GPT-5 coder · MoReSet evaluation set | Procrustes: 0.58 -> 0.51 | Improved |
| add one iteration of evaluator feedback to GPT-5 coder (ablation) | GPT-5 coder · MoReSet evaluation set (VideoPhy2) | Semantic Adherence: 2.60 -> 2.52 | Hurt |
| add one iteration of evaluator feedback to GPT-5 coder (ablation) | GPT-5 coder · MoReSet evaluation set (Trajan) | Average Jaccard: 0.32 -> 0.38 | Improved |
| use physics-simulator code generation instead of diffusion T2V | MoReGen vs Sora2 · MoReSet test (75 videos) | DTW: 11.21 -> 8.93 | Improved |
| fine-tune T2V models for physics validity (NewtonGen from Wan, WISA from CogVideoX) | Wan2.2 and CogVideoX · MoReSet test (75 videos) | DTW: 17.18 -> 17.88 (Wan2.2 vs NewtonGen) | No clear effect |
| scale to large commercial T2V models versus medium open-source models | T2V models · MoReSet test (75 videos) | DTW: 12.94 (CogVideoX-5B) vs 11.21 (Sora2) | No clear effect |
| evaluate with Trajan AJ instead of trajectory metrics | metric validity · MoReSet test (Trajan) | Average Jaccard: 0.79 (LTXV) vs 0.10 (MoReGen) | Hurt |
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.
- NewtonGen — NewtonGen: Physics-consistent and Controllable Text-to-Video Generation via Neural Newtonian Dynamics (ICLR 2026) · better on 2 of 2 shared results
- WISA — WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025) · better on 1 of 1 shared result
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 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
- PhyT2V: LLM-Guided Iterative Self-Refinement for Physics-Grounded Text-to-Video Generation (CVPR 2025 · similarity 0.89)
- Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning (NeurIPS 2024 · similarity 0.88)
- WISA: World simulator assistant for physics-aware text-to-video generation (NeurIPS 2025 · similarity 0.87)
- VideoPhy: Evaluating Physical Commonsense for Video Generation (ICLR 2025 · similarity 0.87)
- Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation (ICML 2025 · similarity 0.87)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.86)
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.
