For many years and many thousands of dollars I searched for the box of software that included talent. I found that I could use the toys in a lot of the boxes to make very pretty results, but each looked as imbalanced and unimpressive as the last. At the end I was forced to conclude that the creative spark, the difference between interesting and not-interesting, would require a massive neural network trained on all manner of inputs over a period of years.

I'm working on training up a neural network now, but I don't have any interesting results yet. Until I can get good enough results, I'll just rely on humans for the things that need to appeal to other humans and I'll rely on software for the other things.