Can I use depth channel of an image to help AI somehow?

This is my typical image to material (AI)  experience  vs   depth channel  generated by ortho projection export from  Reality capture put in Designer .

I wonder can I use and improve the captured depth which is in fact a bit blurry and has lots of photogrammetry artifacts  to help AI do better thing.   Because in my experience  I get mostly nothing like in Sampler youtube videos  but rather hardly recognisable mess.