Yes, one of the main goals of this model was to keep 3D consistency as the camera moves around; it can do this even without relying on an explicit 3D point clou

Yes, one of the main goals of this model was to keep 3D consistency as the camera moves around; it can do this even without relying on an explicit 3D point cloud or Gaussian splat representation of the scene.