"Modular" models

Would it be possible for some kind of “modular” based model system to be developed in future?

So for example, models could be more specialised to perform a more targeted enhancement, and could then be grouped together by the user to suit their target content more closely… as it stands the models are very generic and I’ve found cope very well on certain scenes, but not others… especially in regards to recovering or enhancing details. An example where all the models trip up for me is some beach/sea/cliff shots that introduce some weird artifacts and the picture looks unnatural.

For example, you might have a model that specifically enhanced detail on faces and skin, leaving the rest of the image untouched. A model that specifically detected and enhanced details on textures like carpets, walls and brickwork. A model specifically for enhancing trees, foliage etc. Then you could basically tick/untick which specific modular model enhancements you wanted to be applied, depending on the content.

As these individual models would have more targeted training, they should inevitably work much better on that specific enhancement, than the generic models currently do.

User’s having the ability to train and share their own modular models would also be a game changing enhancement… ie give some kind of training utility some high quality 4k example footage containing elements that you’d want to enhance elsewhere in low quality footage (eg footage of animals), and you could train and use a new modular model to enhance similar content.