InSpatio-World 1.5

InSpatio Team: Donghui Shen, Guofeng Zhang, Haomin Liu, Haoyu Ji, Hujun Bao, Hongjia Zhai, Jialin Liu, Jing Guo, Nan Wang, Siji Pan, Weihong Pan, Weijian Xie, Xianbin Liu, Xiaojun Xiang, Xiaoyu Zhang, Xinyu Chen, Yifu Wang, Yipeng Chen, Zhenzhou Fan, Zhewen Le, Zhichao Ye, Ziqiang Zhao

Read on arXiv

Type: model
Venue: InSpatio
Year: 2026
Source: arxiv
Access: public
Language: en
Added: 2026-09-30
Verified: 2026-09-30

Summary

InSpatio releases InSpatio-World 1.5, a real-time 4D world simulator that turns a single image, multiple images, a panorama, or a video into an explorable, navigable scene with wide viewpoint changes, target camera trajectories, and bullet-time creation. The core is a Spatiotemporal Autoregressive (STAR) architecture: an Implicit Spatiotemporal Cache aggregates reference and historical observations into a latent world representation for long-horizon consistency, and an Explicit Spatial Constraint Module enforces geometric structure and translates user interactions into precise camera trajectories, plus Joint Distribution Matching Distillation (JDMD) to avoid fidelity degradation from over-reliance on synthetic data. For video inputs, Depth Anything 3 automatically estimates missing depth and per-frame camera intrinsics/extrinsics. The 1.3B model scores 68.72 on WorldScore-Dynamic, ranking #1 among evaluated real-time and interactive methods, at up to 24 FPS. Code is Apache 2.0.

Keywords

world models · 4D · video generation · view synthesis · models · interactive simulation

Topics

4D world models, video generation, novel view synthesis, spatial reasoning

Research notes

Discovery: ModelScope (@ModelScope2022, verified) on 2026-09-30: https://x.com/ModelScope2022/status/2105158320764104812
Paper (arXiv 2604.07209, submitted 2026-04-08, revised v2 2026-04-13): INSPATIO-WORLD: A Real-Time 4D World Simulator via Spatiotemporal Autoregressive Modeling, InSpatio Team (alphabetical).
Model: https://modelscope.ai/models/InSpatio/inspatio-world-v1.5
Paper page: https://modelscope.ai/papers/2604.07209
Connects to the collection's world-model, video-generation, and 3D-reconstruction entries.