Reproduce “Guidance Is a Field, Not a Scalar: Interval-Integrated Classifier-Free Guidance for One-Step 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
Zijie Meng, Xixin Cao
AGAVE enables one-step text-to-video generation with guidance that varies across noise-time, spatial regions, and frame positions. It incorporates this schedule into MeanFlow’s average-velocity prediction, allowing one inference evaluation instead of repeated guided steps. On the 14B Wan2.1-distilled backbone, AGAVE scores 85.2 on VBench Total, versus 84.3 for CFG-Zero⋆ at 50 steps, while matching its UCF-101 FVD of 372.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.