Comic Strips

Oops, just wanted to write a quick comment but it evolved into me giving some of my thoughts on AI gen as a means of artistry. Oh well, not deleting this now.

I’m still waiting for someone to make art that requires machine learning and is obviously creative by our standards, instead of using AI to recreate old art.

Most self proclaimed AI artists just type a prompt, maybe do a bit of "prompt engineering" (Read: putting the name of a good artist on the prompt) and then in-paint (Read: re-prompting, but only affects a specific area). That does not give you enough control over the drawing to do anything interesting.

I say this from personal experience. Even small differences is facial expressions, too small to be described with words, can make a big impact. The no. reason artists don't use AI and dislike it is because it doesn't given enough control over the final image, because it does not let them put in details which cannot be described through words. You might say we might someday have an AI that (somehow) gives you more control, but that would nullify the whole "advantage" of AI: Not having to spend time worrying about the details. If you are going to spend 4 hours prompting in details... you could have just gotten a better result by just drawing it yourself.

Think of it like making a level in Mario Maker VS making a game in a game engine. Sure, making things in Mario Maker is faster than making a game yourself, but it doesn't give you the same fine grain control that making a game from scratch would. (But even this is not a perfect analogy has, in Mario Maker you actually get to choose where the blocks go, instead of with AI, where you can only describe how the blocks go and hope the AI gets it right with little hope of editing it yourself.)

Actually, about that "editing it yourself". In this hypothetical AI Mario Maker scenario, you could go into Mario Maker's editor mode and edit the level with the same amount of detail a normal, handcrafted, Mario Maker level would, but with AI image gen, you get the image and... Ya, about has useful as any other downloaded image. Artists typically create layers to do their art thing, but AI output puts everything in one layer, making hard to edit. I could go on this, but I don't have all the time in the world to write this. Someone posted this video on [email protected] , where an AI "artists" quit AI because of these problems of lack of control. (Don't judge me based on the video, I found it on the aforementioned community here (lemmy.ml link))

I know it’s possible to use this tool in a way that’s revolutionary, but the users and developers seem to have little interest in pushing art beyond replacing the artists.

That's the multi billion dollar the AI companies are trying to solve, having to pay wages. The far right loves this as they feel like those who worked hard to develop artistic skills are below them somehow. Part of the conservative rhetoric. AI: The New Aesthetics of Fascism by Gareth Watkins.

I want to see someone develop an original ML model with an original training set that can generate something impossible by any other method.

I feel like people who want talk and argue about AI should know how the training works at a mathematical level. I swear the number of people who act like it's magic is way too much. I say this because it would give you a really good idea of how specialized training won't solve the lack of originality problem. I haven't had a refresher on this so I might be misremembering some things... Any who, this playlist is pretty good I think.