An artist who infamously duped an art contest with an AI image is suing the U.S. Copyright Office over its refusal to register the image’s copyright.
In the lawsuit, Jason M. Allen asks a Colorado federal court to reverse the Copyright Office’s decision on his artwork Theatre D’opera Spatialbecause it was an expression of his creativity.
Reuters says the Copyright Office refused to comment on the case while Allen in a statement complains that the office’s decision “put me in a terrible position, with no recourse against others who are blatantly and repeatedly stealing my work.”
The basic legal test has to do with control over the output. A prompt is not control. If you tell Stable Diffusion “draw a dog playing chess” then you do not control the creative choices made in the image. Thus, they are not protected.
That’s why Pollock paintings can be copyrighted: the key creative choices were controlled by Pollock. He wanted some blue streaks in one area and some red streaks in a different area.
To the extent that AI output can be controlled, it can be copyrighted. If you take a photo and tell an AI, “desaturate this photo” then there is only one possible outcome. The lack of color in the product was fully under your control. Likewise if you say, “Copy dog.gif from my Documents folder to the bottom left corner of the image”.
On the other hand if you say, “Add a dog to the image”, then not so much. Who determined what the dog would look like? Not you. So the dog is in the public domain.
And once in the public domain, it will likely remain there even if you iterate your prompts, like “Elongate the snout and widen the eyes”. For the same reason that you generally cannot copyright an image of the Mona Lisa even with minor alterations.
So if I tell the ai - create a canvas 12" x 12" and put 00FF13 colored paint via a 1" brush aligned 13 degrees in the upper left quadrant 2 inches down and to the right, then another in the bottom right quadrant 3" from the bottom and 2" from the right of 1200FF paint, is that not way more specific than buying paint at a store, doing some guestimation mixing of colors and splattering some on a canvas haphazardly?
Just trying to understand where the line of ML assisted art becomes non-artistic/non-copywritable. Is there a technical criteria? what about artists woh use Photoshop with ML derived filters for their final product? are those also not owned by the artist the moment that the ml tool is applied?
What if you create the model yourself with only your own training data? e.g. you are the only sole creator of the entire tool/model itself.
So many possibilities that create questions…
The cases where copyright was denied involved prompt-driven generative AI. At some point the artist admitted that some of the creative decisions were made by the AI.
In your first example, you made all the creative decisions. The same is true of many filters, for example if I apply a “50s cinema filter” then I know exactly what it will do. The AI doesn’t get to make any decisions.
On the other hand, if you tell the AI to “add some clouds to the image” and the algorithm decides where they go, then the sky it produces is not protected. Only the elements you controlled are.