LogFAQs > #969372375

AI "art" is not itself art, but it might be the basis for art with further work applied. People comparing the outcry against AI art to the development of other new mediums are being disingenuous. Photography was hated originally because it endangered the livelihoods of landscape and portrait painters. Photography did take some of those jobs, but it still requires a good deal of skill and work on the part of a photographer to get good images. Moreover, portrait paintings and landscape paintings were and are still sought after because human artists can inject something special to the interpretation of the subject matter.

Tools like Stable Diffusion are different. You type in a sentence (which you can find on lexia or a similar gallery and make small adjustments to), move a couple of sliders (or not---the defaults work great 90% of the time, and for the remaining 10%, you just move them 'til prompt looks good) and then have the thing spit out as many pieces of art as you want. If none of them look anywhere close to good, you move a couple of words around the prompt or poke the sliders until something decent gets pushed out. You can even add some kind of basic scribbles and have the software "finish it" for you. Where, exactly, is the supposed creativity of the user coming from when creating these "works"?

Stable Diffusion was trained on billions of images---that isn't how a human learns to draw at all and it is very different than using a reference. It isn't a tool like a digital or physical brush: it does all the work for you, and can spit things out in any form or style you like. You can even further refine the model using specific images if you want something very specific that it doesn't do well out of the box. It is a phenomenal achievement of computer science, and it relies on billions of extant, tagged images in order work its magic. I've used almost all of them at this point, and I'll fight anyone who declares that people using Stable Diffusion are working hard to get their results. Some things it just won't do (like recreate likenesses of specific anime characters with very few references available) and for some things you need a different model. But if you earnestly sit down with Stable Diffusion and try and make something and aren't getting in the right ballpark in under 30 minutes, something has gone horribly wrong.

There is something fundamentally different about "regular" art mediums. Even stuff like making complex materials in Substance Designer requires a real negotiation between the artist and their medium, and you grow as a person the more work with a medium. Stable Diffusion is just a really fast Fiverr artist you direct and don't pay (or pay a monthly fee for work), not an art medium.

People know this, too. They want to desperately claim that they made their work, just like a painter made a painting. In fact, there has a been a flood of people who claim stubbornly that they did paint or draw something that is obviously AI produced.