Presentation

<
UniVidX: A Unified Multimodal Framework for Versatile Video Generation via Diffusion Priors
SessionVideo Generation
DescriptionUniVidX is a unified multimodal video generation framework that repurposes video diffusion priors to flexibly generate and translate across visual modalities. Instantiated as UniVid-Intrinsic and UniVid-Alpha, it supports tasks such as inverse rendering, relighting, matting, and inpainting, while achieving strong temporal consistency and data efficiency with limited training videos.
Authors
Houyuan Chen
HKUST
Hong Li
Beihang University
Xianghao Kong
HKUST
Tianrui Zhu
Nanjing University
Shaocong Xu
Beijing Academy of Artificial Intelligence
Weiqing Xiao
Nanjing University
Yuwei Guo
The Chinese University of Hong Kong
Chongjie Ye
The Chinese University of Hong Kong
Lvmin Zhang
Stanford University
Hao Zhao
Tsinghua University
Anyi Rao
HKUST
Event Type
Technical Paper
TimeMonday, 20 July 20262:40pm - 2:50pm PDT
LocationRoom 408 B
Digital Library PDF
Session Time & Location
Sunday, 19 July 20266:00pm - 8:45pm PDTHall K
Monday, 20 July 20262:00pm - 3:30pm PDTRoom 408 B
Keywords
Generative AI
Rendering
Registration Categories
Full Conference Supporter
Full Conference
Next PresentationNext PresentationVideo Generation - Interactive Discussion
Similar Presentations
LongE2V: Long-Horizon Event-based Video Reconstruction, Prediction, and Frame Interpolation with Video Diffusion Models
BodyReLux: Temporally Consistent Full-Body Video Relighting
Just-Dub-It: Video dubbing via Joint Audio-Visual Diffusion