Quality of generation

I think the greatest problem of AI is the quality of the training images. They are all generic and over-saturated.  The themes are the tired,  monsters, inflatable babes and lurid processed photography... the training images need to be more neutral, and then the users can apply their bad taste afterwards. I have been training models in Alpaca and the best ones are not made of knock-out images but of more subtle ones which are not too busy. These respond better to prompts. You can always prompt for drama etc but it is harder to remove it if it is a default due to the training.