| title | CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion |
|---|---|
| authors | Kai Zhu, Yintianrun Zhang, Shuwei Jiang, Ruolin Zuo, Yixin Chen, Duomin Wang, Miao Liu, Yin Li, Daquan Zhou, Yiwu Zhong |
| date | 2026-10-21 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2610.29915 |
| upvotes | 17 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | Diffusion-based video generation models (VGMs) can produce visually plausible videos, yet their camera ego-motion often drifts over long horizons. Camera ego-motion determines how the viewpoint changes between frames and is essential for maintaining consistent geometry and trajectories. To address this problem, we propose CausalAlign, a causal alignment framework with two components. First, we diagnose how precise ego motion is represented, used, and retained within VGMs using linear probes and causal interventions, revealing a rise-and-fall pattern during denoising. Second, we strengthen ego-motion state in VGM by aligning the representations at the identified layers to a 3D foundation model through delta-token space. This space encodes the changes between consecutive frames, allowing the alignment to focus on ego-motion dynamics rather than static scene content. With Wan2.2 fine-tuned on DL3DV and RealEstate10K, CausalAlign can largely improve motion accuracy, geometric consistency, and temporal smoothness, while preserving visual quality. For example, it improves geometric consistency by more than 20% relative to the base model. Moreover, with the fine-tuned VGM, CausalAlign improves the success rate of world action models (WAMs) by 10 points on mobile manipulation tasks of RoboCasa365 benchmark. Overall, CausalAlign improves ego-motion consistency in diffusion-based VGMs and yields noticeable gains in embodied models. |
cite
@misc{zhu2026causalalign,
title={CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion},
author={Kai Zhu and Yintianrun Zhang and Shuwei Jiang and Ruolin Zuo and Yixin Chen and Duomin Wang and Miao Liu and Yin Li and Daquan Zhou and Yiwu Zhong},
year={2026},
eprint={2610.29915},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2610.29915}
}