I'm trying to figure out exactly how AI texture enhancement would work. Typically, you'd have (at least) diffuse, roughness, and normal for each texture map. Multiply that by, say, 5 material tiles (like in La Femme) and you have 15 textures to recalculate. And that is just for one character.

Once that is done, then you have the scene render, which would have to calculate all the light bounces for each item in the scene. And the number of bounces would have to be calculated for each light in the scene. And it wouldn't make sense to do a 4K render for a scene where you enhanced the textures by four times the amount, so render times would be increased, yes?

So, wouldn't that end up being much slower a process than just having AI do a post-process on a finished render to make it larger?