Some shots are impossible to put into words. "From behind her left shoulder, waist height, tilted slightly up, with a railing in the foreground" — write that ten times and the model still may not give you the angle; run it again for the next shot and the people and the set have drifted apart again. DOTV's Previz lets you lay it out instead: place the people, the props and the camera in a 3D space, then take a picture from the camera's point of view. That picture is a gray-and-white blockout, not a finished frame — its job is to be a composition reference: wire it into an image node so the model paints from that angle, with that blocking (see the reference images in 3.2). The camera positions and blocking that prompts can never quite pin down in AI video creation get set exactly right here, once.