SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models

Junchao Huang, Guian Fang, Shengju Qian, Xianghao Kong, Zhuoran Zhao, Wei Huang, Yihua Du, Zixin Zhang, Justin Cui, Yuchao Gu, Yukang Chen, Xinting Hu, Tianyu He, Shaoshuai Shi, Zhuotao Tian, Xin Wang, Mike Zheng Shou, Li Jiang

Project Page | Code | 🤗 Dataset | 🤗 Models | arXiv

• SolarWM provides an open foundation spanning 1.43 million clips from 10 source datasets and a unified three-stage training recipe across four 5B–33B video backbones, enabling real-time interactive rollouts from minutes to hours after training only on 5-second clips.

StereoPilot: Learning Unified and Efficient Stereo Conversion via Generative Priors

Guibao Shen^*, Yihua Du^*, Wenhang Ge^*, Jing He, Chirui Chang, Donghao Zhou, Zhen Yang, Luozhou Wang, Xin Tao, Ying-Cong Chen^†

Project Page | Code | 🤗 Daily Paper | 📰 机器之心 | arXiv

• Converting a 5-second monocular video into a stereoscopic video takes only 11 seconds. Built both parallel and converged 3D dataset.

VideoMemory: Toward Consistent Video Generation via Memory Integration

Jinsong Zhou^*, Yihua Du^*, Xinli Xu^*, Luozhou Wang, Zijie Zhuang, Yehang Zhang, Shuaibo Li, Xiaojun Hu, Bolan Su, Ying-Cong Chen^†

Project Page | Code | arXiv

• Propose an entity-centric, dynamic memory bank framework that enables coherent long-form, multi-shot narrative video generation by retrieving and updating character/prop/background states to preserve identity across temporal gaps.

RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models

Yaoyang Liu^*, Dongyu Yan^*, Yihua Du, Shengju Qian, Zeke Xie, Xin Wang, Ying-Cong Chen, Long Chen^†

Project Page | 🤗 Daily Paper | arXiv

• RAR uses Stride-RoPE and synchronized multi-stride video and control sampling to generate at controllable temporal rates, switch rates mid-rollout without resetting history, and extend physical horizons.

A Mechanistic View on Video Generation as World Models: State and Dynamics

Luozhou Wang^*, Zhifei Chen^*, Yihua Du, Dongyu Yan, Wenhang Ge, Guibao Shen, Xinli Xu, Leyi Wu, Man Chen, Tianshuo Xu, Peiran Ren, Xin Tao, Pengfei Wan, Ying-Cong Chen^†

Code | 🤗 Daily Paper | arXiv | 📰 机器之心

• Bridge this divide: (1) Taxonomy: Unifying Implicit Context vs. Explicit Latent Compression (2) Evolution: Shifting benchmarks from “Visual Fidelity” to “Physical Persistence” and Causality.

ReWorld: An Interactive World Model with Long-Horizon Memory

Zhifei Chen^*, Luozhou Wang^*, Guibao Shen, Dongyu Yan, Shuai Yang, Tianshuo Xu, Yihua Du, Wei Wang, Tianyi Gui, Lianghua Huang, Ying-Cong Chen^†

Project Page | Code | 🤗 Daily Paper | arXiv

• ReWorld decouples short-horizon action control from long-horizon memory and combines a fixed-budget KV cache with a pose-indexed landmark bank, enabling real-time 704×1280 interactive video streaming with consistent recall of revisited scenes.