The rationale is when you do not use 3D models to keep subjects consistent over multiple scenes it becomes too unpredictable and you would need way too many trial and errors. 3D helps with it. Then, if you even want to have that very same character starting from a specific pose it will more likely you will give up to the likes of the AI model rather than the other way around.
We’re actually finding that leveraging 3D reduces complexity rather than increasing it. Setting up a 3D scene is getting easier with tools like image-to-3D, and once it's set up, you can reuse it flexibly across different shots without having to recreate assets from scratch.
By passing a structured 3D-rendered image to the AI, you increase the chances of the video model generating exactly what you want, rather than relying on unpredictable, frame-by-frame AI improvisation.