Idea – Could Machine Learning Enhance The Solution To Differential Equations By Image Post Processing?

I was thinking. Our differential equations might not be able to generate the detail required for some situations. Or it might require to much memory and processing power to get that high detail. If at all possible?

Therefor I wonder if one could ”enhance” the solution of a differential equation by image post processing. Like we do in rendering ray traced photo realistic images.

Here it should be possible to post process an approximative solution to look like a ”photo” realistic solution.

I think this could be automated with machine learning.