[forsythe_]

I agree. This is one of the big factors that I haven't seen thoroughly addressed on this topic: no matter how capable AI tools are at interpreting broad declarative prompts there's still a wide degree of complexity in expressing detail and subtlety that I don't think they'll be able to match anytime soon, if ever.

The author gives the example of needing dozens of images to accompany scenes in his book. If you feed DALL-E something like: "create a panoramic view of a grey ocean with four mountains and an ominous dark sky", then sure, it can handle that prompt fine. It'll do a great job at broadly interpreting adjectives and phrases that are unambiguous. Mass produced art for advertising/media is going to change drastically because it doesn't need precision.

However, once you introduce specificity and detail the tools are going to suck. For example, telling it to generate a portrait of a specific person with subtlety in expression, body language, pose, and demeanor. It'll create something that looks aesthetically sane but not anywhere near what you had in mind.

And just for some extra counter-predictions of my own:

- The novelty of seeing thousands of "good enough" images in mass media is going to wear off

- AI generated images will become obvious to the average person

- Society will largely get tired of AI art due to the lack of novelty

- The utter collapse of the art profession some are predicting is never going to come