Research map / Video / Video generation
World-model video generation
Uses video generators as interactive world simulators and benchmarks their physical consistency.
110 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Video generation, keyword cluster “video generation · video diffusion · text video”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
3 papers in 2023–24 and 105 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.3% to 9.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2021 · Automatic Generation of Flexible Plans via Diverse Temporal Planning (AAAI 2021 · 0 citations)
- 2022 · Playable Environments: Video Manipulation in Space and Time (CVPR 2022 · 13 citations)
- 2024 · RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · 4 citations)
- 2025 · DriveDreamer-2: LLM-Enhanced World Models for Diverse Driving Video Generation (AAAI 2025 · 59 citations)
- 2026 · Pre-Trained Video Generative Models as World Simulators (AAAI 2026 · 4 citations)
Earliest papers here
- Automatic Generation of Flexible Plans via Diverse Temporal Planning (AAAI 2021 · 0 citations)
- Playable Environments: Video Manipulation in Space and Time (CVPR 2022 · 13 citations)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · 4 citations)
What has been tried
33 papers here report 377 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 4%, hurt 23%, mixed 1%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace SD1.4 with SVD while retaining the cross-view module | nuScenes val, first-frame multi-view image · FID | 14.9 -> 17.2 | Hurt | DriveDreamer-2: LLM-Enhanced World Models for Diverse Driving Video Generation (AAAI 2025 · 59 citations) |
| Add novel-trajectory training with λnovel=0.5 instead of λnovel=0 | Waymo lane change · NTA-IoU | 0.349 -> 0.405 | Helped | DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · 35 citations) |
| Add a KAN-based material field | 7 scenes · aesthetic quality | 4.86 -> 4.89 | Helped | DreamPhysics: Learning Physics-Based 3D Dynamics with Video Diffusion Priors (AAAI 2025 · 10 citations) |
| Fuse data-driven and physically predicted 3D poses | FineGym · MPJPE | 0.229 -> 0.178 | Helped | FinePhys: Fine-grained Human Action Generation by Explicitly Incorporating Physical Laws for Effective Skeletal Guidance (CVPR 2025 · 10 citations) |
| Add distractor balls to point-force training scenes | Point force localization · Force localization | Helped | Force Prompting: Video Generation Models Can Learn And Generalize Physics-based Control Signals (NeurIPS 2025 · 9 citations) | |
| combine noise embedding fm(ϵ) with learnable parameters p in the mask token | OpenDV-2K validation · FVD | 126.71 -> 105.52 | Helped | MaskGWM: A Generalizable Driving World Model with Video Mask Reconstruction (CVPR 2025 · 5 citations) |
| Add numeric conditioning to vision | Collision (ID) · prediction error | No clear effect | How Far Is Video Generation from World Model: A Physical Law Perspective (ICML 2025 · 5 citations) | |
| Increase the history context from 1 to 64 frames | DOOM test trajectories · PSNR | 20.94 ± 0.044 -> 22.36 ± 0.033 | Helped | Diffusion Models Are Real-Time Game Engines (ICLR 2025 · 5 citations) |
| Add sparse historical keyframes | VBench · Temporal Flickering | 0.7260 -> 0.7580 | Helped | Video World Models with Long-term Spatial Memory (NeurIPS 2025 · 4 citations) |
| add text parsing to factorize task instructions | RT-1 human evaluation, seen tasks · Human | 85.5 -> 90.1 | Helped | RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · 4 citations) |
| Add the dynamic module | complex terrains · Penetration | 12.7% -> 0% | Helped | POMP: Physics-consistent Motion Generative Model through Phase Manifolds (CVPR 2025 · 3 citations) |
| Pre-train on human videos in addition to robot videos | Meta-World MT50-rand · success rate | 6.9% relative improvement | Helped | Learning an Actionable Discrete Diffusion Policy via Large-Scale Actionless Video Pre-Training (NeurIPS 2024 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 54 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- nuScenes · 6 of 54 papers
- LIBERO · 5 of 54 papers
- CALVIN · 4 of 54 papers
- Ego4D · 3 of 54 papers
- Waymo · 2 of 54 papers
- AMASS · 2 of 54 papers
- Something-Something-v2 · 2 of 54 papers
- Panda-70M · 2 of 54 papers
Baselines
- Drive-WM · 4 of 54 papers
- Panacea · 4 of 54 papers
- DynamiCrafter · 4 of 54 papers
- GPT-4o · 4 of 54 papers
- DrivingDiffusion · 3 of 54 papers
- OpenSora · 3 of 54 papers
- iVideoGPT · 3 of 54 papers
- DIAMOND · 3 of 54 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 108 papers; from OpenAlex and the papers' affiliations).
- Shanghai Jiao Tong University · 7 papers
- Tsinghua University · 7 papers
- University of Science and Technology of China · 5 papers
- Harbin Institute of Technology · 4 papers
- Hong Kong University of Science and Technology · 4 papers
- Shanghai AI Laboratory · 4 papers
- University of Oxford · 4 papers
- The Chinese University of Hong Kong · 4 papers
- Fudan University · 4 papers
- Alibaba Group · 4 papers
Use it
Download BibTeX of 21 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Efficient video diffusion models (283 papers)
- Compositional text-to-video diffusion (159 papers)
- Trajectory-guided motion control (184 papers)
- Autoregressive long video generation (156 papers)
Elsewhere in the topic
- Text-to-motion diffusion (140 papers)
- Diffusion-based human image animation (126 papers)
- Diffusion-based video editing (92 papers)
- Talking face generation (81 papers)
- Video-to-audio synthesis (76 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).