# Ziqi Cai (蔡子祺)

> Ph.D. student in the Camera Intelligence Lab at Peking University (School of Computer Science), advised by Prof. Boxin Shi, and currently a visiting scholar at Institute of Science Tokyo. Research: physically grounded visual generation and interactive world modeling.

Ziqi Cai studies how generative models can internalize the laws of light, geometry, and persistent 3D state, enabling them to estimate scene illumination, synthesize controllable radiance, and simulate coherent multi-agent worlds. Before Peking University, Ziqi Cai earned a bachelor's degree from Beijing Jiaotong University and was a research intern at the Institute of Computing Technology, Chinese Academy of Sciences (ICT, CAS). At Peking University, Ziqi Cai was awarded the President's Scholarship.

- Homepage: https://caiziqi.com/
- Email: zqtsai@gmail.com
- https://scholar.google.com/citations?user=o_E15d8AAAAJ
- https://github.com/ghostcai
- https://www.cnblogs.com/ghostcai

## Publications

- [Video Generation Models Are Inherent Lighting Estimators](https://caiziqi.com/research/vlite/): Ziqi Cai, Shuchen Weng, Kaiqi Liu, Zifeng Wang, Zhiquan Zhang, Minggui Teng, Han Jiang, Boxin Shi. European Conference on Computer Vision (ECCV), 2026. Paper: https://arxiv.org/pdf/2607.04674; Project: https://caiziqi.com/research/vlite/.
- [Lighting-grounded Video Generation with Renderer-based Agent Reasoning](https://caiziqi.com/research/liver/): Ziqi Cai, Taoyu Yang, Zheng Chang, Si Li, Han Jiang, Shuchen Weng, Boxin Shi. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026. Paper: https://arxiv.org/pdf/2604.07966; Project: https://caiziqi.com/research/liver/.
- [SPW-Nav: A Streaming Panoramic World Model for Language-Guided Navigation](https://alaya-lab.github.io/SPW-Nav/): Yunheng Liu*, Ziqi Cai*, Siqi Yang, Yimu Wang, Minggui Teng, Jiaming Tan, Shuchen Weng, Erwin Wu, Kaipeng Zhang, Boxin Shi. Technical Report, arXiv:2610.08941. *Equal contribution. Paper: https://arxiv.org/pdf/2610.08941; Project: https://alaya-lab.github.io/SPW-Nav/.
- [MASS: Multiplayer World Models with Authoritative Shared State](https://alaya-lab.github.io/MASS/): Ziqi Cai, Siqi Yang, Yimu Wang, Zixian Gao, Yunheng Liu, Shuchen Weng, Erwin Wu, Kaipeng Zhang, Boxin Shi. Technical Report, arXiv:2608.06257. Paper: https://arxiv.org/pdf/2608.06257; Project: https://alaya-lab.github.io/MASS/.
- [MAVIN: Multi-Shot Audio-Visual Generation with Customized Narrative Control](https://arxiv.org/pdf/2606.29473): Kaiqi Liu, Yunyao Mao, Ziqi Cai, Zheng Geng, Jing Wang, Qiulin Wang, Xintao Wang, Pengfei Wan, Kun Gai, Shuchen Weng, Boxin Shi. European Conference on Computer Vision (ECCV), 2026. Spotlight. Paper: https://arxiv.org/pdf/2606.29473.
- [ReContraster: Making Your Posters Stand Out with Regional Contrast](https://arxiv.org/pdf/2604.10442): Peixuan Zhang, Zijian Jia, Ziqi Cai, Shuchen Weng, Si Li, Boxin Shi. Annual Meeting of the Association for Computational Linguistics (ACL), 2026. Paper: https://arxiv.org/pdf/2604.10442.
- [PAWA: Panoramic Action-conditioned World Modeling for Multiple Agents](https://pawa.siki.moe): Siqi Yang, Yimu Wang, Ziqi Cai, Chengxuan Zhu, Shuchen Weng, Erwin Wu, Zhaofei Yu, Kaipeng Zhang, Boxin Shi. arXiv:2610.20240. Paper: https://arxiv.org/pdf/2610.20240; Project: https://pawa.siki.moe; Code: https://github.com/yousiki/pawa-pipeline.
- [StreamAV-Bench: A Comprehensive Benchmark for Streaming Audio-Video Generation](https://liukqchoco.github.io/StreamAVBench/): Kaiqi Liu, Haoxuan Zeng, Jingqi Liu, Jiacong Fang, Ziqi Cai, Yunyao Mao, Henglin Liu, Yu Sheng, Shuchen Weng, Boxin Shi. arXiv:2608.26336. Paper: https://arxiv.org/pdf/2608.26336; Project: https://liukqchoco.github.io/StreamAVBench/; Code: https://github.com/liukqchoco/StreamAVBench-Code.
- [DirCast: Live Esports Directing and Commentary with a Single Streaming Video-Language Model](https://zixiangao.github.io/DirCast/): Zixian Gao, Lixin Xiu, Penghao Luo, Xiangcheng Cao, Ziqi Cai, Zhen Li, Erwin Wu, Kaipeng Zhang. arXiv:2610.31734. Paper: https://arxiv.org/pdf/2610.31734; Project: https://zixiangao.github.io/DirCast/; Code: https://github.com/ZixianGao/DirCast-code.
- [PhyS-EdiT: Physics-aware Semantic Image Editing with Text Description](https://openaccess.thecvf.com/content/CVPR2025/papers/Cai_PhyS-EdiT_Physics-aware_Semantic_Image_Editing_with_Text_Description_CVPR_2025_paper.pdf): Ziqi Cai, Shuchen Weng, Yifei Xia, Boxin Shi. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. Paper: https://openaccess.thecvf.com/content/CVPR2025/papers/Cai_PhyS-EdiT_Physics-aware_Semantic_Image_Editing_with_Text_Description_CVPR_2025_paper.pdf.
- [Unified Reconstruction of Static and Dynamic Scenes from Events](https://openaccess.thecvf.com/content/CVPR2025/papers/Gao_Unified_Reconstruction_of_Static_and_Dynamic_Scenes_from_Events_CVPR_2025_paper.pdf): Qiyao Gao, Peiqi Duan, Hanyue Lou, Minggui Teng, Ziqi Cai, Xu Chen, Boxin Shi. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. Highlight. Paper: https://openaccess.thecvf.com/content/CVPR2025/papers/Gao_Unified_Reconstruction_of_Static_and_Dynamic_Scenes_from_Events_CVPR_2025_paper.pdf.
- [Real-time 3D-aware Portrait Video Relighting](http://www.geometrylearning.com/VideoRelighting/): Ziqi Cai, Kaiwen Jiang, Shu-Yu Chen, Yu-Kun Lai, Hongbo Fu, Boxin Shi, Lin Gao. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. Highlight. Paper: https://arxiv.org/pdf/2410.18355; Project: http://www.geometrylearning.com/VideoRelighting/; Code: https://github.com/GhostCai/PortraitRelighting.
- [DreamPolish: Domain Score Distillation with Progressive Geometry Generation](https://arxiv.org/pdf/2411.01602): Yean Cheng*, Ziqi Cai*, Ming Ding, Wendi Zheng, Shiyu Huang, Yuxiao Dong, Jie Tang, Boxin Shi. arXiv:2411.01602. *Equal contribution. Paper: https://arxiv.org/pdf/2411.01602.