WorldFlow Jointly Predicts Video and 3D Motion via Rectified-Flow
October 28, 2026
WorldFlow is a generative world model that uses a Mixture-of-Transformers (MoT) backbone to jointly predict future video and 3D trajectories. It parameterizes 3D motion as continuous displacement tokens, coupling explicit geometric dynamics with visual evolution.
HOW THIS AFFECTS YOU
●
builderThis could be a core component for developing more capable embodied AI and robotics simulation environments.
●
researcherThe unified MoT approach provides a more robust learning signal by combining visual and geometric supervision.