Current proprietary models struggle to generate complex human actions
Fig. 2 in the main paper
SORA-2
Your browser does not support HTML5 video.
VEO-3