Ai photo2material is a nice idea but so far a thing you toy a few times with and then never use .
Would be more helpful to have something Ai helping to fix what ortho projection from Reality capture gives you: photo and depth :
1. ability to work with big arbitrary sized original first . like 16547x5346 where you could extend the photogrammetry chunk to image borders first and fix bad patches and photogrammetry artifacts using content aware and a special "content aware move" tool that has a small 2x2 deforming mesh to match the patches precisely. Working properly in depth/height without "steps"
1a Something like super convenient Frequency separation tool in Affinity Photo to get rid of height gradients and low frequency things with actual manual input instead of just simple hi pass.
2. Crop tool that would let u chose what part of the extended chunk would be actual texture where same move/patch tool coudl be used to cover seams . Same working in depth with adaptation without sadden abrupt height riffs.
3 . Ai producing roughness , removing tiny and big shadows in color layer , making depth and normal map less blurry looking you often get from photogrammetry untill you are using super hi res camera ,gazillion shots and special albedo producing flash light eating batteries like crazy .
That said current ai shadow removing is not very good in Sampler killing all if every colour nuances. So perhaps this part is most hard to do one.
ps. Or maybe just add this 2x2 mesh to Photoshop patch/con aware move tools please.