Infinite-World

Scaling Interactive World Models to 1000-Frame Horizons via Pose-Free Hierarchical Memory

Ruiqi Wu^1,2,3*, Xuanhua He^4,2*, Meng Cheng², Tianyu Yang², Yong Zhang^2‡, Chunle Guo^1,3†, Chongyi Li^1,3, Ming-Ming Cheng^1,3

¹Nankai University ²Meituan ³NKIARI ⁴HKUST

*Equal Contribution †Corresponding Author ‡Project Leader

Arxiv Code

Abstract

We propose Infinite-World, a robust interactive world model capable of maintaining coherent visual memory over 1000+ frames in complex real-world environments. While existing world models can be efficiently optimized on synthetic data with perfect ground-truth, they lack an effective training paradigm for real-world videos due to noisy pose estimations and the scarcity of viewpoint revisits.

To bridge this gap, we first introduce a Hierarchical Pose-free Memory Compressor (HPMC) that recursively distills historical latents into a fixed-budget representation. By jointly optimizing the compressor with the generative backbone, HPMC enables the model to autonomously anchor generations in the distant past with bounded computational cost, eliminating the need for explicit geometric priors. Second, we propose an Uncertainty-aware Action Labeling module that discretizes continuous motion into a tri-state logic. This strategy maximizes the utilization of raw video data while shielding the deterministic action space from being corrupted by noisy trajectories, ensuring robust action-response learning.

Furthermore, guided by insights from a pilot toy study, we employ a Revisit-Dense Finetuning Strategy using a compact, 30-minute dataset to efficiently activate the model’s long-range loop-closure capabilities. Extensive experiments, including objective metrics and user studies, demonstrate that Infinite-World achieves superior performance in visual quality, action controllability, and spatial consistency.

Methodology

Overview of Infinite-World architecture.

(a) Hierarchical Pose-free Memory Compressor: The HPMC recursively compresses raw historical latents into a fixed memory budget via hierarchical compression with local and global stages. The compressor is jointly optimized with the DiT backbone to autonomously anchor generations in the distant past with constant computational cost.

(b) Uncertainty-Aware Action Labeling: Continuous poses are decoupled into translation and rotation primitives. A tri-state logic filters out "Uncertain" motion to ensure robust action-response learning.

(c) Data Strategy: Pre-training on open-domain video is followed by finetuning on a revisit-dense dataset to activate 1000-frame memory consistency.

Video Gallery

A street in a fantasy city where buildings are carved into gargantuan ancient trees, glowing sap running through bark, misty floor.

A serene indoor Zen pool room, the water surface perfectly still reflecting a circular skylight, smooth concrete walls, a single bonsai on a floating wooden platform.

Looking across a vast salt flat after rain, a thin layer of water creating a perfect sky mirror, pastel sunset colors.

A cozy attic workshop with exposed wooden beams, a large window showing a rainy sky, a clean drafting table with neatly organized architectural sketches.

Looking out at a calm glacial lake, clear turquoise water, smooth gray stones on the shore, distant snow-capped mountains reflecting on the surface.

A wide view of a turquoise coral reef seen through shallow clear water from a boat, bright tropical sun, colorful underwater life.

Comparison

Quantitative comparison on VBench (motion smoothness, dynamic degree, aesthetic quality, image quality) and user study (memory, fidelity, action, ELO rating). Best in bold, second best underlined.

Model	VBench					User Study
	Mot. Smo.↑	Dyn. Deg.↑	Aes. Qual.↑	Img. Qual.↑	Avg. Score↑	Memory↓	Fidelity↓	Action↓	ELO Rating↑
Hunyuan-GameCraft	0.9855	0.9896	0.5380	0.6010	0.7785	2.67	2.49	2.56	1311
Matrix-Game 2.0	0.9788	1.0000	0.5267	0.7215	0.8068	2.98	2.91	1.78	1432
Yume 1.5	0.9861	0.9896	0.5840	0.6969	0.8141	2.43	1.91	2.47	1495
HY-World-1.5	0.9905	1.0000	0.5280	0.6611	0.7949	2.59	2.78	1.50	1542
Infinite-World	0.9876	1.0000	0.5440	0.7159	0.8119	1.92	1.67	1.54	1719

BibTeX

@article{wu2026infiniteworld,
  title={Infinite-World: Scaling Interactive World Models to 1000-Frame Horizons via Pose-Free Hierarchical Memory},
  author={Wu, Ruiqi and He, Xuanhua and Cheng, Meng and Yang, Tianyu and Zhang, Yong and Kang, Zhuoliang and Cai, Xunliang and Wei, Xiaoming and Guo, Chunle and Li, Chongyi and Cheng, Ming-Ming},
  journal={arXiv preprint arXiv:2602.02393},
  year={2026}
}