titleFew-Step Streaming Video Generation via Condition Shifting
authorsDeyuan Liu, Jungang Li, Kaiwen Zheng, Kai Jiang, Jintao Chen, Yaole Wang, Ming Lin, Tao Lin, Jianfei Chen, Jun Zhu, Jintao Zhang
date2026-11-10
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.14033
affiliationTsinghua University
upvotes88 (Hugging Face Daily Papers, 2026-11-13)
linksabs pdf hf
cite
relatedVidu S2: Real-Time Interactive, Editable, and Spatial Video Generation
Vidu S1: A Real-Time Interactive Video Generation Model
AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning
LiveEdit: Towards Real-Time Diffusion-Based Streaming Video Editing
Transferring the Intelligence of VLMs to Robotic Control
abstractFew-step sampling reduces the cost of streaming video generation. Distribution Matching Distillation (DMD) supervises few-step generators with a separate fake-score network, which adds training memory and computation and must track the evolving generator. We extend APEX to streaming video by using condition shifting to estimate the velocity field of the generated distribution within the generator. The original text condition selects generation, while a shifted condition selects fake estimation using the same parameters. This removes the independent fake network while retaining S2-Avatar's frozen reference teacher, Self-Replay Forcing, and perceptual regularization. We evaluate APEX on 180-second videos using three backbone evaluations per block and a one-step spatial Refiner, with a 70-case human study and a 320-case StreamAV-Bench comparison. Compared with S2-Avatar, APEX achieves higher video instruction fulfillment on the Progressive track (3.856 versus 3.831) and higher visual aesthetics across both tracks (0.641898 versus 0.632606). Human raters favor APEX most strongly on instruction following, with +30.0% net preference, followed by consistency and video quality at +4.3% each. Motion quality favors S2-Avatar, audiovisual synchronization is predominantly tied, and usability is unchanged. Separate measurements under matched training settings show 4.4% lower peak allocated GPU memory per GPU and 9.2% shorter mean iteration time than DMD.

bibtex

@misc{liu2026few,
  title={Few-Step Streaming Video Generation via Condition Shifting},
  author={Deyuan Liu and Jungang Li and Kaiwen Zheng and Kai Jiang and Jintao Chen and Yaole Wang and Ming Lin and Tao Lin and Jianfei Chen and Jun Zhu and Jintao Zhang},
  year={2026},
  eprint={2611.14033},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.14033}
}