new styles? you can train on existing styles, then just add some random noise to the latent space. Or you can mix existing styles like you can mix various amounts of red, green, and blue to create new colors.

But I’d say humans would still be involved because even though an ai can generate a new style, it’s ultimately up to humans to decide If the art style is worth while to look at. But you can automate that ranking of art style a bit, indeed ai was used to filter on images worth training stable diffusion on.