Fighting Against Generative Art is Futile
The Islands of Human Supremacy Are Shrinking Fast
If your argument against AI art is "it's not as good as the masters," you are fighting a losing battle. The tide of AI competence is rising, and while there are still islands of human supremacy, they are shrinking fast.
The thing I can't get over generative art is that the number of variations it creates is infinite. The attached images are two memetically identical sets -- that is, the same prompt was used to create them -- individuals and couples. Given a model, a seed number, and the prompt, the model will render an identical result every time, but the number of seeds is infinite.
What does it say about the value of individual variations when the possibility space is infinite? The value of the work shifts from solely the end product, to being split 3-ways between the model, the prompt, and the judgment used to pick a particular seed to share with you. Negotiating the value allocation between these three elements will be a major point of cultural struggle in the coming decades.
Aside: Why Is AI So Bad At Drawing Hands?
Why is AI so bad at drawing hands? The root of the problem is that fingers are complex dynamic systems, far more so than the rest of the body. There is an infinite number of positions a hand can be in, especially in relation to other hands and objects. Btw, human artists also struggle with hands, which is why many cartoon characters wore gloves.
To correctly render a hand requires awareness of the underlying physical space, which generative models are not built to do. I believe that this is being solved by creating models large enough to contain internal representations of hand physics. This is done indirectly -- by training the models on more hands with GANs, they increase the latent space for hands, where the AI develops the underlying physical simulation on its own.
Generative Art Is Not Inherently Derivative
Given a particular prompt, generative AI will generate an infinite stream of original variations. Some people criticize AI art for remixing the work of other artists, but that's not necessarily true -- while you can tell it to imitate specific artists, it can also create original work. Yes, its knowledge derives from the images used to build the model, but only in the same sense as a new novel reuses English words the author learned in other works.
Here's an example of Shakespeare-inspired art I made. I can't get my mind around the fact that I can produce an infinite stream of this, and soon the art will get better and more creative than any human artist. (Btw, I know nothing of medieval dress, and my memory of Shakespeare is foggy - I had ChatGPT pick the scenes and write the prompts. Stable Diffusion can't render some scenes well yet, so I picked creativity over accuracy for some of them.)
