CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion

Hugging Face Daily Papers Papers
video-generation diffusion-models ego-motion 3d-vision robotics representation-alignment

Summary

CausalAlign 论文提出一种因果对齐框架，通过将扩散视频生成模型的内部表征与 3D 基础模型在 delta-token 空间中对齐，改善长时程相机自运动的一致性，在几何一致性上提升超过 20%，并提升了具身操作任务中世界动作模型的成功率。

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.
Original Article
View Cached Full Text

Cached at: 10/28/26, 01:56 AM

Paper page - CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion

Source: https://huggingface.co/papers/2610.29915 Published on Oct 21

·

Submitted byhttps://huggingface.co/j-zhu

KaiZhuon Oct 27

Abstract

CausalAlign improves long-horizon camera ego-motion consistency in diffusion video generators by aligning internal representations to 3D dynamics via delta-token space.

Diffusion-based video generation models(VGMs) can produce visually plausible videos, yet theircamera ego-motionoften drifts over long horizons.Camera ego-motiondetermines 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 usinglinear probesandcausal interventions, revealing a rise-and-fall pattern duringdenoising. Second, we strengthen ego-motion state in VGM by aligning the representations at the identified layers to a3D foundation modelthroughdelta-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 improvesgeometric consistencyby more than 20% relative to the base model. Moreover, with the fine-tuned VGM, CausalAlign improves the success rate ofworld 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.

View arXiv pageView PDFProject pageGitHub7Add to collection

Get this paper in your agent:

hf papers read 2610\.29915

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.29915 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.29915 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.29915 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Video Generation Models: A Survey of Post-Training and Alignment

Hugging Face Daily Papers

This paper presents the first comprehensive survey of post-training and alignment methods for video generation models, organizing existing approaches into implicit and explicit alignment categories including supervised fine-tuning, self-training/distillation, preference- and reward-based methods, and inference-time methods, while reviewing datasets, benchmarks, and open challenges.

Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video Generation

Hugging Face Daily Papers

The paper introduces Ego2Act, a goal-directed benchmark of 2,640 egocentric videos across 110 real-world tasks for evaluating whether video generation models can simulate multi-step physical manipulation, along with Ego2Act Judge, a reference-free evaluation pipeline that aligns better with human judgments of task completion and physics plausibility.