Let me introduce my project called MATISSE AI - it is toolset of support template-based AI-assisted image generation. It is introductory video fully describing all features done on moment of publication.
MATISSE AI [www.youtube.com]

It can help you build and control image from shapes, images and ROP-renders or previously done generations. Then automatically get generated imagery form different kind of working processes - like custom direct API call, ComfyUI server or even some canvas-based web-tools integrated directly into MATISSE tool. Also you can transfer generation results further to Copernicus.
MATISSE features complex working with curve-based shapes, some of them are unique and related to my decades of experience with artistic computer graphics tools.

Yes, it is all about control and prediction, as main purpose of tool. Of course, it highly depends on image generation model you use for rendering and it's harness and environment. But this tool is an independent facade for different complex setups which you may implement for your needs, and you may easily switch your backyard generation mechanics. I decide to build it with high degree of integrations in Houdini, but it is not strict bind, just successful collaboration. Tool now is in active developing phase, I have some milestones in it. It is developed not in "vibecoding manner" - all features are controlled, it would be strange to develop generation controlling tool without high degree of controlling its development.

Thank you