Reproduce “FastFrames: Video Generation at Interactive Latency” 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

Gene Chou, Guandao Yang, Fujun Luan, Lars Mescheder, Wei Dong, Stephan Richter, Vladlen Koltun

FastFrames accelerates autoregressive video generation by replacing costly cache refinement with a lightweight repair network and decoding pixels from transformer features, while also optimizing denoising. In a 1.3-billion-parameter Wan 2.1 comparison at approximately 480p on one B200 GPU, it reduces latency from 272 to 36 milliseconds, with similar aggregate benchmark scores; 720p results exceed 50 milliseconds.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.