we should continue to talk about the ethics of web scraping. but it's very possible to train a model only on public domain images - CC0 photos, works by long dead artists - and still get interesting, beautiful output. this is an uncomfortable fact, but it's also a hopeful one.
it's not enough to put a modern subject in the style of someone ancient. good! it's not enough to juxtapose two contrasting concepts for shock value. good! it's not enough to 'blaspheme' an established art tradition by making a small alteration. good! we can no longer rely too heavily on the prominence of other artists without having anything much new to say ourselves. good!
new developments in art always come of saying "the old ways are no longer enough". this is no different - possibly, it is easier. the old masters we challenge today are neural nets. stacks of linear algebra in a trench coat.
we're better. we can 'train' ourselves by mindfully consuming art which influences (with far more nuance and warmth) what we create. we can choose which of the 'datasets' we've been fed to draw from and which to ignore.
every artist is asking themself - what can I do that an AI cannot? how can I infuse my work with humanity? and I'm hopeful for what that will bring about.
will there be increased interest in physical media, where the limitations of the materials and the artist's body are more tangible in the work?
will we lean on longer forms - series of paintings, illustrated novels, comics - with complicated storylines and a requirement for consistent representation of characters and concepts?
will we begin to make, display and consume art in ways that emphasize the process of making it?
will we reject mass produced items in favor of forming connections with artists and makers?
will we choose to express incisive messages - precise, complicated ideas that go beyond vague associations between metadata tags sifted from a sentence prompt?
god. I hope we will.