Research map / Video / Video generation
PerpetualWonder: Long-horizon Action-conditioned 4D Scene Generation
Jiahao Zhan, Zizhang Li, Hong-Xing Yu, Jiajun Wu
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 |
|---|---|---|---|
| use VPP (physics particles anchoring Gaussians) instead of standard 3D Gaussians for the foreground (ablation) | PerpetualWonder foreground · multi-view optimization, qualitative | dynamics plausibility | Improved |
| add progressive multi-view optimization instead of direct optimization from all views at once (ablation) | PerpetualWonder backward optimization · multi-view consistency, qualitative | multi-view consistency | Improved |
| propagate generative refinements back to physics particles (closed loop) instead of resetting Gaussians each window | hybrid generative simulator, long-horizon · castle scene, four-round interaction | shape integrity | Improved |
| use PerpetualWonder instead of WonderPlay | hybrid generative simulator · 10 scenes, WorldScore | Camera Controllability: 75.95 -> 93.26 | Improved |
| use PerpetualWonder instead of WonderPlay | hybrid generative simulator · 10 scenes, WorldScore | 3D Consistency: 63.93 -> 80.41 | Improved |
| use PerpetualWonder instead of WonderPlay | hybrid generative simulator · 10 scenes, WorldScore | Imaging Quality: 36.80 -> 66.98 | Improved |
| use PerpetualWonder instead of Veo3.1 | video generation baseline · 10 scenes, WorldScore | Imaging Quality: 67.82 -> 66.98 | No clear effect |
| use PerpetualWonder instead of Veo3.1 | video generation baseline · 2AFC human study | favor rate (%): 62.0% physics plausibility, 70.8% motion fidelity | Improved |
Later papers that compared with this method
1 later accepted paper reports PerpetualWonder as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- LivingWorld: Interactive 4D World Generation with Environmental Dynamics (ECCV 2026 · better on 6 of 6 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tora — Tora: Trajectory-oriented Diffusion Transformer for Video Generation (CVPR 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
- 4DWeaver: Bridging Reconstruction and Generation Via Compact Autoregressive Priors (TPAMI 2026 · similarity 0.89)
- 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Choreographing a World of Dynamic Objects (CVPR 2026 · similarity 0.88)
- Free4D: Tuning-free 4D Scene Generation with Spatial-Temporal Consistency (ICCV 2025 · similarity 0.88)
- DreamScene4D: Dynamic Multi-Object Scene Generation from Monocular Videos (NeurIPS 2024 · similarity 0.88)
- Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting (ICLR 2024 · 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.
