Reproduce “Kandinsky 5.0: A Family of Foundation Models for Image and 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
Vladimir Arkhipkin, Vladimir Korviakov, Nikolai Gerasimenko, Denis Parkhomenko, Viacheslav Vasilev, Alexey Letunovskiy, Nikolai Vaulin, Maria Kovaleva and 17 more
Kandinsky 5.0 is a family of models for image generation and editing, text-to-video, and image-to-video. Staged training, NABLA sparse attention, and distillation target video-generation costs. In human comparisons, Video Lite scored 0.73 versus 0.27 for Kandinsky 4.1 on motion dynamics, but 0.50 on prompt following. Flash cut 10-second generation time from 224 to 61 seconds, with moderate quality loss.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.