Research map / Video / Video generation
Dynamic World Generation Made Efficient
Fengrui Tian, Jinqi Luo, Uday Kiran Reddy Tadipatri, Hancheng Min, Rene Vidal
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 Spatial Token Memory (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | Temporal consistency: 2.112 -> 2.442 | Improved |
| add Spatial Token Memory (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | CLIP score: 23.12 -> 24.55 | Improved |
| add 3D token geometry (reproject KV tokens) vs naive KV caching (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | CLIP score: 19.98 -> 24.55 | Improved |
| add token depth alignment to static-world depth (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | Temporal consistency: 1.667 -> 2.442 | Improved |
| add token depth alignment (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | Factual consistency: 2.145 -> 2.233 | Mixed |
| add background completion (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | T2I alignment: 2.340 -> 2.997 | Improved |
| add foreground Gaussians (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | Factual consistency: 1.627 -> 2.233 | Improved |
| add Spatial Token Memory vs removal on T2I alignment (ablation) | DynWorld · Dynamic world views (VideoScore + CLIP) | T2I alignment: 2.658 -> 2.997 | Improved |
| parallel generation with memory instead of sequential DynamicVoyager | DynWorld · Generation time, 8 views (16 frames) | runtime: 2 hours 56 minutes -> 18-22 minutes | Improved |
| inject memory only into last attention layers #40-53 (ablation) | DynWorld · Generation time vs FID | FID: 173.2 -> 182.2 | Mixed |
| DynWorld vs Voyager overall WorldScore | DynWorld · WorldScore | Overall score: 54.53 -> 66.11 | Improved |
Closest papers by meaning
- V3D: Video Diffusion Models are Effective 3D Generators (TPAMI 2025 · similarity 0.90)
- WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories (CVPR 2026 · similarity 0.90)
- SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints (ICLR 2025 · similarity 0.89)
- World from Motion: Generative Dynamic Gaussian Reconstruction from Monocular Video (NeurIPS 2026 · similarity 0.89)
- SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025 · similarity 0.89)
- GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025 · 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.
