Tag
1 article
This article explains how Microsoft Research's World-R1 uses reinforcement learning and 3D-aware rewards to improve geometric consistency in text-to-video generation without changing the underlying model architecture.
