Best episode I’ve heard in years. And the subject reminds me of two (almost three) things:

1) In an Art History class taken around 2000, I remember reading an anecdote about the late post-modern artist Robert Rauschenberg. According to the story, at a show opening an attendee approached Robert and asked him what the difference was between him (Robert) putting junk together and calling it a “Rauschenberg” and the her (the attendee) doing the same thing and calling it a “Rauschenberg”, to which Robert replied, “Only I can create a bad ‘Rauschenberg’.”

2) The late Harry Harrison’s short story “Portrait of the Artist” (published in 1964) addressed the idea of programmable machines intruding upon the the art world-comic books specifically-and one artist’s reaction to that.

3) In an Intro to Photography class also taken in 2000, we discussed whether digital photography-which wasn’t ubiquitous back then-qualified as “art” compared to film photography, and how even film photography experienced the same labeling of “NOT ART” by painters in the 19th century. Sculptors had no opinion because sculpture is what you trip over when stepping back to look at the paintings ;)

An important distinction that I think at least Dan understands, but Wayne misunderstood, is that the majority of AI art generators start from random noise, not an existing image, and then tries to interpret that noise into the prompt based on its understanding of how to create the elements of the prompt (which it gets by being given a training image with relevant keywords, noise is applied over top of that, and it tries to recreate the original image).

Strictly speaking, the AI doesn’t have the original images in its data set, just its knowledge of how to create a similar image based on the same keywords (though some recent research has shown that it can sometimes memorize and reproduce images from the original data, so that’s a problem).

But. Regardless of that, AI art at this point should either be for personal use only, or for commercial use only in the case that the art is generated with training data that you own the rights to (video game studios could potentially use their huge databases of assets they made to create new assets).

I’ve been playing around with it for use in my campaigns for things like unique monsters that don’t have art on the internet (and I’m not going to commission someone for) or for making a bunch of bandit tokens to provide some consistent variety to their looks on the VTT.