The thing that took me longest to work out: you cannot prompt a camera move. You describe it, the model invents its own, and you burn a run every time. So I stopped describing it. I block the move in Blender first. Grey boxes, no textures, no lighting, nothing in there except where the camera goes and when it gets there. Then I hand that grey clip to Seedance as a movement reference and it copies the path frame for frame. The prompt then does one job only, which is saying what things look like. The previz already decided where the camera goes. This F1 one is thirteen shots and twelve hard cuts in a single generation. The cuts are baked into the previz, not into an edit. The .blend files are attached at the bottom of the lessons in the Classroom. Open one, press play, watch it through the camera.