WorldCrafter's approach: no explicit 3D reconstruction. The camera pose queries an implicit memory of earlier views; the video generator uses that context to render the next shot. The idea is that…