BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260417T190107Z
LOCATION:West Building\, Rooms 118-120
DTSTART;TZID=America/Los_Angeles:20250812T104500
DTEND;TZID=America/Los_Angeles:20250812T123500
UID:siggraph_SIGGRAPH 2025_sess146@linklings.com
SUMMARY:Video Generation
DESCRIPTION:CineMaster: A 3D-Aware and Controllable Framework for Cinemati
 c Text-to-Video Generation\n\nA 3D-aware and controllable text-to-video ge
 neration method allows users to manipulate objects and camera jointly in 3
 D space for high-quality cinematic video creation.\n\n\nQinghe Wang (Dalia
 n University of Technology); Yawen Luo (The Chinese University of Hong Kon
 g); Xiaoyu Shi (Kuaishou Technology); Xu Jia and Huchuan Lu (Dalian Univer
 sity of Technology); Tianfan Xue (The Chinese University of Hong Kong); an
 d Xintao Wang, Pengfei Wan, Di Zhang, and Kun Gai (Kuaishou Technology)\n-
 --------------------\nLayerFlow: A Unified Model for Layer-aware Video Gen
 eration\n\nWe propose LayerFlow,  a unified framework for layer-aware vide
 o generation, enabling seamless creation of transparent foregrounds, clean
  backgrounds, and blended scenes. With multi-stage training and LoRA techn
 iques improving layer-wise video quality with limited data, it also suppor
 ts variants lik...\n\n\nSihui Ji (The University of Hong Kong); Hao Luo (D
 AMO Academy, Alibaba Group); and Xi Chen, Yuanpeng Tu, Yiyang Wang, and He
 ngshuang Zhao (The University of Hong Kong)\n---------------------\nGenera
 tive Video Matting\n\nLimited high-quality ground-truth data hinders tradi
 tional video matting's real-world application. This work tackles this by a
 dvocating for large-scale training with diverse synthetic segmentation and
  matting data. A novel generative pipeline is also introduced to predict t
 emporally consistent alpha...\n\n\nYongtao Ge (The University of Adelaide,
  Zhejiang University); Kangyang Xie, Guangkai Xu, and Mingyu Liu (Zhejiang
  University); Li Ke, Longtao Huang, and Hui Xue (Alibaba Group); Hao Chen 
 (Zhejiang University); and Chunhua Shen (Zhejiang University of Technology
 , Zhejiang University)\n---------------------\nInterspatial Attention for 
 Efficient 4D Human Video Generation\n\nWe introduce a novel interspatial a
 ttention (ISA) for diffusion transformers, which maintains identity and en
 sures motion consistency while allowing precise control of camera and body
  poses. Combined with a custom video variation autoencoder, our model achi
 eves state-of-the-art performance for photo...\n\n\nRuizhi Shao (Tsinghua 
 University), Yinghao Xu (Stanford University), Yujun Shen (Alibaba Group),
  Ceyuan Yang (ByteDance Inc.), Yang Zheng and Changan Chen (Stanford Unive
 rsity), Yebin Liu (Tsinghua University), and Gordon Wetzstein (Stanford Un
 iversity)\n---------------------\nMobius: Text to Seamless Looping Video G
 eneration via Latent Shift\n\nMobius is a novel method to generate seamles
 sly looping videos from text descriptions directly without any user annota
 tions, thereby creating new visual materials for the multi-media presentat
 ion.\n\n\nXiuli Bi, Jianfei Yuan, and Bo Liu (Chongqing University of Post
  and Telecommunications); Yong Zhang (Meituan); Xiaodong Cun (Great Bay Un
 iversity); Chi-Man Pun (University of Macau); and Bin Xiao (Chongqing Univ
 ersity of Post and Telecommunications)\n---------------------\nDiffusion a
 s Shader: 3D-aware Video Diffusion for Versatile Video Generation Control\
 n\nDiffusion as Shader (DaS) is a unified approach for controlled video ge
 neration that uses 3D tracking videos to enable versatile editing, includi
 ng animating mesh-to-video, camera control, motion transfer, and object ma
 nipulation, while improving temporal consistency.\n\n\nZekai Gu (Hong Kong
  University of Science and Technology), Rui Yan (Zhejiang University), Jia
 hao Lu and Peng Li (Hong Kong University of Science and Technology), Zhiya
 ng Dou (University of Hong Kong), Chenyang Si (Nanyang Technological Unive
 rsity), Zhen Dong (Wuhan University), Qifeng Liu (Hong Kong University of 
 Science and Technology), Cheng Lin (University of Hong Kong), Ziwei Liu (N
 anyang Technological University), Wenping Wang (Texas A&M University), and
  Yuan Liu (Hong Kong University of Science and Technology)\n--------------
 -------\nMotionCanvas: Cinematic Shot Design with Controllable Image-to-Vi
 deo Generation\n\nMotionCanvas enables intuitive cinematic shot design in 
 image-to-video generation by letting users control both camera movements a
 nd object motions in a 3D-aware scene. Combining classical graphics with m
 odern diffusion models, it translates motion intentions into spatiotempora
 l signals—withou...\n\n\nJinbo Xing (The Chinese University of Hong Kong, 
 Adobe Research); Long Mai, Cusuh Ham, Jiahui Huang, and Aniruddha Mahapatr
 a (Adobe Research); Chi-Wing Fu (The Chinese University of Hong Kong); Tie
 n-Tsin Wong (Monash University); and Feng Liu (Adobe Research)\n----------
 -----------\nVideo Generation - Interactive Discussion\n\nAfter the summar
 y presentations, attendees will participate in an interactive discussion. 
 Outside the room will be a series of poster boards for authors to gather a
 round with the audience. Authors are invited to bring any material related
  to their paper that could instigate further conversation such...\n\n\nInt
 erest Area: Research & Education\n\nRecording: Livestreamed, Not Livestrea
 med, Recorded, Not Recorded\n\nRegistration Category: Full Conference, Vir
 tual Access, Tuesday\n\nSession Chair: Li-Yi Wei (Adobe Research)
END:VEVENT
END:VCALENDAR