Chenxi Song

Chenxi Song

Postdoc Researcher at AGI Lab, Westlake University. Research interests: 3D&4D scene generation, controllable video generation, and world models.

– visits

Publications

You can also find my articles on my Google Scholar profile.

WorldForge V2: Eliciting Camera-Trajectory Alignment in Video Diffusion Models

Published in arXiv preprint, 2026

Camera-trajectory control for video diffusion models through training-free guidance, efficient refinement, and data-efficient Trajectory LoRA adaptation.

Recommended citation: C Song, Y Yang, T Zhao, R Li, JT Zhou, C Zhang. (2026). "WorldForge V2: Eliciting Camera-Trajectory Alignment in Video Diffusion Models." arXiv preprint arXiv:2610.31640.
Download Paper

World in World: Explore the World with World Models

Published in arXiv preprint, 2026

A training-free inference-time interface that turns heterogeneous control evidence into camera- and time-labelled visual states, read through the native self-attention of a frozen causal video world model.

Recommended citation: C Song, Y Yang, C Zhang. (2026). "World in World: Explore the World with World Models." arXiv preprint arXiv:2609.11548.
Download Paper

DyWeight: Dynamic Gradient Weighting for Few-Step Diffusion Sampling

Published in arXiv preprint, 2026

A lightweight, learning-based multi-step solver that learns unconstrained time-varying parameters to adaptively aggregate historical gradients for efficient diffusion sampling.

Recommended citation: T Zhao, M Lei, L Yuan, Y Yang, C Song, Y Wang, B Zhu, C Zhang. (2026). "DyWeight: Dynamic Gradient Weighting for Few-Step Diffusion Sampling." arXiv preprint arXiv:2603.11607.
Download Paper

StyleAvatar3D: Leveraging Image-Text Diffusion Models for High-Fidelity 3D Avatar Generation

Published in IEEE Journal of Selected Topics in Signal Processing (J-STSP), 2026

A method for generating high-quality, stylized 3D avatars using pre-trained image-text diffusion models and GAN-based 3D generation.

Recommended citation: C Zhang, Y Chen, Y Fu, W Cheng, Z Zhou, W Jiang, Z Wang, B Fu, T Chen, G Yu, G Lin, C Song. (2026). "StyleAvatar3D: Leveraging Image-Text Diffusion Models for High-Fidelity 3D Avatar Generation." IEEE J-STSP.
Download Paper

Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration

Published in CVPR 2026, 2025

A training-free geometry-aware caching framework that accelerates 3D diffusion inference while preserving geometric fidelity.

Recommended citation: M Yang, Y Yang, C Xu, C Song, Y Zuo, T Zhao, R Li, C Zhang. (2025). "Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration." arXiv preprint arXiv:2511.22533.
Download Paper

Appagentx: Evolving gui agents as proficient smartphone users

Published in arXiv preprint, 2025

A novel evolutionary framework for GUI agents that enhances operational efficiency through action evolution.

Recommended citation: W Jiang, Y Zhuang, C Song, X Yang, JT Zhou, C Zhang. (2025). "Appagentx: Evolving gui agents as proficient smartphone users." arXiv preprint arXiv:2503.02268.
Download Paper

FewarNet: An efficient few-shot view synthesis network based on trend regularization

Published in IEEE Transactions on Circuits and Systems for Video Technology, 2024

An efficient few-shot view synthesis network that uses trend regularization for improved performance.

Recommended citation: C Song, S Wang, J Wei, Y Zhao. (2024). "FewarNet: An efficient few-shot view synthesis network based on trend regularization." IEEE Transactions on Circuits and Systems for Video Technology. 34 (10), 9264-9278.
Download Paper