Achieving controllable camera movement requires the model to internally construct a three-dimensional representation of the scene (such as point clouds, meshes, voxels, or neural implicit representations) to ensure geometric consistency across different viewpoints