If you want to see a good example of human-in-the-loop visual augmentation, check out this tool that I worked on at one point:
https://www.vizcom.ai/
They now have the capability of taking a sketch, making a 2d image with prompt guidance, and then generating 3d objects from that, and creating 2d images from composable 3D objects. I think it's just about the workflow and the control we wish for. I'm an Artist myself, and I find this an excellent tool. We can choose what parts we want to automate, and what parts we want machines to do.
Do you have a link to this paper? It would be great to read, thanks!