Coda-DaVinci adds agent-guided control to Code2Video generation
Researchers introduced Coda-DaVinci, a task-adaptive agentic framework for Code2Video that selects between proxy-guided and direct generation, adjusts modeling granularity per entity, and coordinates physics simulation with proxy-aligned prompting. The authors report stronger instruction following, scene consistency, physical plausibility, and visual realism in their experiments, while noting limitations in transition stability and adherence to scene settings.
The paper presents a concrete approach to controlling AI video generation through agent-authored 3D proxies, physics simulation, and multi-agent review, backed by controlled experiments and blinded human evaluation.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.26595 (cs)
[Submitted on 19 Nov 2026]
Title:Coda-DaVinci: Reimagining Video Generation with Code2Video
Authors:Yiyan Ji, Qixun Wang, Yang Shi, Zhuoran Zhang, Yifan Xu, Tengfei Liu, Yuqi Tang, Xiaohan Zhang, Shujie Li, Jiafu Tang, Xuanyu Zhu, Xinyu Liu, Yan He, Yushuo Guan, Zihao Wang, Xianghua Ying, Shixuan Liu, Haotian Wang, Wenjing Yang, Zhaoxiang Zhang, Yuanxing Zhang, Pengfei Wan, Jiaheng Liu
View a PDF of the paper titled Coda-DaVinci: Reimagining Video Generation with Code2Video, by Yiyan Ji and 22 other authors