Reproduce “Wave Forcing: Towards Speed-of-Light Streaming Video Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yijie Jin, Yuxuan Liu, Jiajun Xu, Ye Huang, Chuzhi Wang, Jiayang Li, Ruiyang Chen, Chenling Hong and 2 more
Wave Forcing overlaps video chunks at different denoising stages, allowing newer chunks to use cleaner predecessors’ cached attention context without skipping denoising updates or the clean-history store pass. After pipeline fill, it completes one chunk per concurrent forward round. In a 5-step comparison on 100-second videos, Dynamic Degree is 91.45 versus Rolling Forcing’s 36.63; this evidence is limited to the tested model and evaluation setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.