Reproduce “ReferStream: Interactive Subject Injection for 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
Shuolin Xu, Guanyuan Li, Boyu Li, Kelu Yao, Xiaogang Xu, Hongchuan Yu, Shuicheng Yan, Xiaobin Hu
ReferStream enables users to add a referenced subject during streaming video generation while preserving established subjects and emitted frames. It uses Joint Latent to contextualize reference features, then Cache Restoration to restore global memory after generating the event chunk. Against online LongLive-2.0, it scores 52.86 New-ID, 73.18 Existing-ID, and 82.33 Count accuracy; this whole-system comparison does not isolate Cache Restoration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.