Comment Re:AI "art" on Etsy == basic fraud and violating T (Score 1) 87
AI produced objects cannot, by definition, do this-- they're un-copyrightable to start with.
This is frequently misrepresented, and I strongly encourage people to go back and read the US copyright office ruling on the subject rather than bad internet summaries.
The copyright office determined that, based on their current understanding of the tech at the time of writing (which they stated will require changing stances as the nature of the technology changes), people using AI to generate media do not control the details of the composition, only generalities, and thus raw outputs are not copyrightable. Note the boldface: raw outputs. The ruling, however, explicitly spells out that further postprocessing on a work (which is extremely common with AI art) or even selection / compilation (a near-universal thing with AI art) can meet the human control and creativity requirement.
Now, personally, I think it's a weak ruling. We'll ignore that it was already out of date (things like ControlNet give you precise control over the composition of your scene). Was Ansel Adams controlling the exact composition of his scenes? "Okay, I'd like the river here to turn more to the west, add a few more mountains in the back, let's make these trees bigger, let's get rid of those clouds... okay, now cue the flock of geese!? No. He was "exploring the latent space" of his world looking for aesthetically favorable scenes, not hand-positioning them like someone in a studio - but the latent space in AI diffusion models is far more vast. He controlled his camera/film configuration, of course - but the amount of configuration options in something like Automatic1111 and its extensions are far more vast. I simply don't see how someone pointing a camera out their window and pressing a button to take a picture of whatever happens to be out there warrants protection, but someone spending hours exploring the latent space of an AI model does not. Especially if they're forcing a specific appearance onto a scene with ControlNet (or even img2img).
But that's all an aside. Because the simple fact is, good AI art is rarely "just type in some text, click a button, grab one of the first things it spits out, and post it". It's generally a long process of refinement, not just on the prompt / settings side, but often going back and forth between Photoshop for hand edits and img2img for AI refinements until you get it just how you want. And this, under the copyright office's ruling, is able to be protected by copyright.