Why the Furor Over AI Image Generation Copyright is Misplaced
Right now AI art generation is making the news all over the internet — it started with the incredible images produced by Dall-E 2, then MidJourney hit, and it was a game changer. AI art started showing up everywhere. The next thing to appear on the scene was StableDiffusion, an open-source AI image generator. Dall-E 2 has fairly stringent restrictions on use and is fairly expensive, which hurt adoption. MidJourney is more accessible and less restrictive. While Dall-E 2 mandates G-rated content and restricts all political content or real-world faces, MidJourney is good with PG and allows real-world faces and political content.
StableDiffusion is open source. It’s unrestricted… if you are running on your own computer you can do whatever you want. In addition, the open-source nature has allowed for massive innovation in the space, with new tools coming out every day since the release.
That’s the backstory, now to the meat. One thing AI image generators can do is mimic the style of existing artists. This has led to a lot of talk about copyright. Most of the controversy comes from not understanding how diffusers (the method used for the modern family of AI art generators) actually work. Now, as I said, this isn’t a legal perspective, first I don’t have the knowledge to give a legal perspective, second at this point it’s still up in the air, and third, it will be different in different places. This is why the controversy shouldn’t exist from a technical perspective.
The photo above is a marble tile. It’s essentially random noise. Now, when you look at it you might start to see shapes — a face, a dog, whatever.
That’s one of the many things I saw in the pattern. Now, if I was a better artist than I am I could take that inspiration and keep transforming it, adding details, transforming it into a better drawing.