Current AI image/video generators lack precise spatial or angle control — you can't reliably prompt for an exact camera tilt (e.g., 20°) or height change (e.g., 1-2 feet) and get a consistent result. A workaround demonstrated in Figma Weave's node-based workflow (see Node-Based Prompt→LLM→Image→Video Pipeline Chaining) is to feed a generated or reference 2D image into a 3D-reconstruction node (Houdini-style) that converts it into a 3D object/model. Once in 3D, camera position, tilt, and height can be adjusted precisely, and the scene is then re-rendered back to 2D from the new camera position. This sidesteps the generator's lack of native spatial control by moving the camera-angle decision into an explicit 3D intermediate step rather than relying on prompt wording alone. Related: Blocking Lock via Position Reference for locking camera/subject position via reference image; this technique instead lets you actively change the position with precision before re-rendering.
Apply: When a shot needs an exact, hard-to-prompt camera angle or height, reconstruct the reference image into 3D first, adjust the virtual camera there, then re-render — rather than iterating on prompt language to coax the generator into the right angle.