📝 Publications

GigaWorld-1: A Roadmap to World Models for Robot Policy Evaluation
GigaAI Team (Xuancheng Xu)

[Paper] [Project page] [Code]

• Responsible for the training and evaluation of the video foundation model in GigaWorld-1, as well as the fine-tuning of downstream tasks.
• GigaWorld-1 is a 1.3B/5B-parameter autoregressive diffusion-transformer world model designed specifically for robot policy evaluation. It generates simulated robot rollout videos and is adapted with parameter-efficient LoRA fine-tuning so that it can better predict whether different robot policies are likely to succeed or fail.

SMRABooth: Subject and Motion Representation Alignment for Customized Video Generation
Xuancheng Xu, Yaning Li, Sisi You, Bing-kun Bao

[Paper] [Project page] [Code]

• IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026
• We explore the application of representation alignment in video customization and demonstrate significant performance improvements across various baselines.

Disco-LoRA: Disentangled Composition of Content, Style, and Motion for Multi-concept Video Customization
Xuancheng Xu, Gengyun Jia, Bing-kun Bao

[Paper] [Project page]

• ACM International Conference on Multimedia (MM) 2026
• We pioneer the task of multi-concept video customization by systematically categorizing content, style, and motion, and establish a comprehensive benchmark comprising four distinct tasks to evaluate this capability. Furthermore, we introduce a novel framework to effectively disentangle and recombine diverse concepts from data.

CLGC: Continuous layout guidance for consistent text-to-video editing
Xuancheng Xu, Ming Tao, Bing-kun Bao

[Paper]

• IEEE International Conference on Multimedia & Expo (ICME) 2025 [Oral]
• We propose CLGC, a novel framework that leverages layout guidance for consistent and efficient video editing.