> These flaws would not exist (to such a degree) when a professional artist paints the same scene.
The flaws in AI generated art were 100x as obvious in systems like this only a few years ago. In 5 years I doubt anyone will be able to tell the difference between AI and human art.
When the photograph displaced most portrait painters, we invented a new type of artist - the photographer. I hope we’ll see the same thing here - artists who specialise in using stable diffusion (and friends) to make new art in a new way. This blog post is like one of the world’s first photographers saying “hey look how the photo changes when I move the subject relative to a light source!”. I can’t wait to see what results we get with deep expertise (and better algorithms).
How long before we have filmmakers using AI to cast, direct and shoot their films?
Maybe, maybe not. It didn't happen with CGI, the uncanny valley hasn't been surmounted, car chases aside.
The mulchers, as Bruce Sterling calls them, have a fresh meat problem. They've consumed all the words, and all the pictures, and we already know that feeding them their own mulch gives worse results.
We're not at the scale limit for data but we know where it is. It's not clear that refinements to the mulching process will create mulch good enough to tell apart from creation. It might. But it might not.
> with CGI, the uncanny valley hasn't been surmounted
I'm not so sure about this. While some scenes are still obviously using CGI, I think a lot of CGI in movies now passes unnoticed, even entirely digital characters.
We certainly notice when those characters do things humans can't do, of course, and when budget or schedule or both result in things being pushed out too early, but how would we know when digital characters look natural on-screen? We wouldn't!
I wonder if anyone is gonna take advantage of this point in time where the average person isn't aware of these breakthrough AI models, and sets up themself an account as a professional-grade artist on Fiverr, offering to draw highly detailed landscapes or whatever where the client can provide reference material and ideas.
You generate the image in 30 minutes (maybe less if you get the process down to a science), then wait around for a few weeks to keep up the illusion that you're actually doing the drawing by hand, and send it off to your satisfied client. You could be charging hundreds of dollars for your "artistic services," and have dozens of clients going on simultaneously.
The flaws in AI generated art were 100x as obvious in systems like this only a few years ago. In 5 years I doubt anyone will be able to tell the difference between AI and human art.
When the photograph displaced most portrait painters, we invented a new type of artist - the photographer. I hope we’ll see the same thing here - artists who specialise in using stable diffusion (and friends) to make new art in a new way. This blog post is like one of the world’s first photographers saying “hey look how the photo changes when I move the subject relative to a light source!”. I can’t wait to see what results we get with deep expertise (and better algorithms).
How long before we have filmmakers using AI to cast, direct and shoot their films?