Research map / Video / Video generation
Infinite-World: Scaling Interactive World Models to 1000-Frame Horizons via Pose-Free Hierarchical Memory
Ruiqi Wu, Xuanhua He, Meng Cheng, Tianyu Yang, Yong Zhang, Zhuoliang Kang, Xunliang Cai, Xiaoming Wei, Chun-Le Guo, Chongyi Li, Ming-Ming Cheng
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 revisit-dense dataset finetuning (ablation) | Infinite-World baseline · Five-chunk horizon | Memory consistency rank: 2.40 -> 1.83 | Improved |
| add revisit-dense dataset finetuning (ablation) | Infinite-World baseline · Five-chunk horizon | Action responsiveness rank: 2.95 -> 1.61 | Improved |
| add uncertainty-aware action labeling (ablation) | Infinite-World baseline · Five-chunk horizon | Action responsiveness rank: 2.95 -> 2.17 | Improved |
| add uncertainty-aware action labeling (ablation) | Infinite-World baseline · Five-chunk horizon | Visual fidelity rank: 2.10 -> 1.78 | Improved |
| add UAL on top of RDD finetuning (ablation) | Infinite-World · Five-chunk horizon | Action responsiveness rank: 1.61 -> 1.38 | Improved |
| add UAL and RDD together on top of baseline (ablation) | Infinite-World · Five-chunk horizon | Memory consistency rank: 2.40 -> 1.69 | Improved |
| use hierarchical compression instead of no compression or direct compression (ablation) | HPMC · GPU memory on 80GB H800 | GPU memory vs context length: plateau at approximately 45GB; no compression OOM at over 180 frames | Improved |
| scale pilot training data from 100 to 1,000 sequences (ablation) | toy DiT generator with memory · pilot study on synthetic 3D scenes | loop-closure quality | No clear effect |
| infer on 6 chunks when trained with max 4-chunk context (ablation) | toy DiT generator with memory · pilot study on synthetic 3D scenes | memory stability | Hurt |
| Infinite-World versus Yume 1.5 on VBench average | Infinite-World · Long-horizon interactive benchmark | VBench average score: 0.8141 vs 0.8119 | No clear effect |
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.
- Yume 1.5 — Yume1.5: A Text-Controlled Interactive World Generation Model (CVPR 2026) · better on 3 of 4 shared results
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
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.89)
- Learning World Models for Interactive Video Generation (NeurIPS 2025 · similarity 0.89)
- Learning Visual Feature-Based World Models via Residual Latent Action (NeurIPS 2026 · similarity 0.89)
- WorldMem: Long-term Consistent World Simulation with Memory (NeurIPS 2025 · similarity 0.88)
- Trimming the Long-Tail of Visual World Modeling Evaluation (NeurIPS 2026 · similarity 0.88)
- WorldWeaver: Generating Long-Horizon Video Worlds via Rich Perception (NeurIPS 2025 · similarity 0.88)
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.
