World-R1

World-R1 is a framework developed by Microsoft that uses reinforcement learning with feedback from pre-trained 3D foundation models to align text-to-video outputs with physical 3D constraints.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.