Reproduce “Runway Characters: Serving a 57B Video Diffusion Transformer for Real-Time Conversation” 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
Rohan Agarwal, Robin Andeer, Alex Armbruster, Bennett Arthur, Michail Christos Doukas, Piero Esposito, Qiming Fang, Aron Filkey and 25 more
The authors build a conversational video system that generates four-frame chunks at 24 frames per second. It reduces repeated key-value cache work by moving operations to writes and overlaps tensor-parallel communication with computation. On four H100 GPUs, production sustains 24.5 frames per second, including resets; occasional slow reset steps mean this does not guarantee every chunk meets its deadline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.