I wanna show you what I mean, since I just finished working on exactly this today.
Here's a 3D Sketchup model that I made. Unlike the skybox, it's an actual 3D object that allows me to rotate at any angles I want
Which I use as the foundation for my my backgrounds.
A model takes me ~30-60 minutes to build on a free software. It's quick and rough because mainly use it to get the perspectives and foundation of the scene; I can add details on my own. Some people take much longer because they add all thee details they need to their model. Some people import fully rendered scenes that others have made to speed up the process.
For step 2 of your suggestion: Im not sure how AI would help find premade 3D scenes. 3D assets are categorized by their creators. Sketchup has the "Warehouse" for example, where you can just search for premade asset. I also don't think AI is at a point where they can recognize 3d objects, let alone full scenes?
For step 3: I don't think AI can generate 3D models right now either. To do so would require a massive database of 3D assets. Unlike 2D images, which can be swipe on google, 3D assets aren't openly available. It would require partnering with company like Sketchup to allow you to take premade 3D assets and put them into a database. And correct me if I'm wrong but AI can't process 3D models. The way Art AI work in very simple term is is by processing images as code, scrambling said code, and produce a new code that results in an new image. 3D assets are vastly more complicated than images.
Step 1 and 4 is simple. Storyboarding takes a few minutes a panel but also require knowledge of composition in relation to your story's narrative (i.e which camera angle will make a character seem more powerful vs weak). AI could help with generating ideas perhaps, but it seems unnecessary. Getting proper camera angles also doesn't take much work. It's just spinning things to the way you want it.
Step 5 is probably the most useful and possible, as others have point out. As you can see. the the screencap of my 3D model is very bare. II would be helpful if Ai programs can render details (i.e line works, wall texture) using my screenshot of the sketchup. BUT that's basically what AI already do and it require data, which, as of now, are stolen.