While I’ve never had a problem with LLM’s in most actually productive and reasonable use cases I’ve still never really seen the point in generative AI images and of course am constantly annoyed at all the slop online.

But in testing some locally run models as well as frontier models I’ve realized that there’s not only actual utility but also lots of interesting potential creative and artistic applications of these tools that are not slop.

Prompting is not remotely comparable to actually putting in the effort to bring something to life from scratch, but I do seriously think that the process of doing many small adjustments in your instructions over and over again until you get something that resembles your vision takes mental effort, taste, and creativity. It is not a substitute for art, but I don’t think that makes it bad, I think it should be understood as its own thing separate from art but still a creative outlet, maybe comparative to programming in some way.

And of course I am obviously not talking about just whipping out ChatGPT and just using whatever you get, that is completely different from the method of AI “art” generation I’m referring to. If you’re not actually envisioning the specifics of a scene then that is just slop, because the model is doing the “creative” work for you instead of you doing it yourself.

Another use case I can see value in is using it for editing and visualizing things. My friend has a daughter with aphantasia, so since she could not mentally picture things she was indecisive about how to decorate, so my friend took a photo of the space and asked for the decorations to be shown in that space and her daughter ended up liking the idea. Thats one of I’m sure thousands of potentially millions of cases where these tools can actually be used for accessibility.

But visualization really is my main point here, it’s the main thing I see as being worth while in these tools, because it allows you to quickly get a reference for what something would look like that’s not just your own brains neural networks confabulations, and as these tools continue to get better they will be more and more appropriate in professional workflows.

So I just really don’t see a reason to outright reject an entire technology, yes the slop is exhausting, no its not good for the environment if you use cloud/frontier models, no you shouldn’t act like you’re a real artist because you just prompted some stuff, but there’s still so many current and potential use cases that are objectively useful to many people, so instead of complaining about how others use these tools I attempt to find ways I might be able to use these tools for myself and in a way that’s aligned with my morals and ethics, of which are very similar to the anti-AI crowed. So I am not pro or anti ai, I am not blindly rejecting or blindly accepting of it, it is a tool, and so like any other tool it should be used responsibly and where appropriate. And particularly in regards to AI it should be kept to yourself.

I am still very much against polluting the internet with slop, people don’t really care about what you prompted, I think it’s fine and reasonable to use it as a tool in your creative process. But it should not be your final product, ever.

And of course the environmental impact is a whole other can of worms, but you shouldn’t act like real artists are like eco friendly saints, paper requires trees, digital artist draw on a device made with minerals acquired through child slave labor in the Congo, and I’m sure you can think of many more examples.

And it is a bit of a clickbaity title, but it’s true, I do understand AI “art” now, it’s just very specific use cases not as a whole.

And no LLM’s were involved in writing this, I never “write” with AI on public posts. That’s weird.

And for further context I’m a photographer and occasional graphic designer, neither of which involve generative AI at any step of my creative process.

And I’m sorry to the person who just posted about how AI debates are exhausting, lmfao. I’m trying to bring a more nuanced take to this debate and I hope that is reciprocated; I’m not interested in arguing about this stuff I’m just sharing my perspective.

I would make the “in this essay I will” joke again but I basically did write an essay, on the toilet too, truly the throne of introspective pondering. And as the writing environment implies this really is not a genuine essay or an attempt to convince anyone of anything, it’s just my yapping so I do not want nor expect it to be treated otherwise.

Ffs I literally now just dropped my phone on the tile floor💔😭😭 Mustve been the ghost of Marx trying to take it away from me while writing this BS since as we all know communism is when no iPhone

you are viewing a single comment's thread
view the rest of the comments
[–] 8 points 2 weeks ago (30 children)

I’ve realized that there’s not only actual utility but also lots of interesting potential creative and artistic applications of these tools that are not slop.

“Slop” is in the eye of the beholder. I try to avoid using the term given its anti-AI baggage.

  • source
  • hideshow 30 child comments
  • [+] 11 points 2 weeks ago* (last edited 1 week ago) (29 children)
  • [–] 11 points 2 weeks ago (18 children)

    Some of the problems with the "slop" term off the top of my head:

    1. It implies an objective metric of image quality that is not being met by AI generated images, but doesn't provide one. This already has some knockoff effects at the offset. It's effectively saying that images are only good if they look good by [arbitrary metric that usually comes down to whether people on the internet say they look good or not]. This easily excludes a shitload of images drawn by humans that could now easily be called "slop" because somebody felt it looked too much like their idea of "AI", which must be bad because it's AI. There's something of a circular logic going on there in how people use the term.

    2. It doesn't account for what happens when an image generation model can produce images that aren't easily categorized as "slop". Which can already happen. Some models are good at producing images that you'd never guess are AI at a glance because they look hand-drawn. This ends up making people look stupid if their whole definition boils down to something that AI can already overcome.

    3. It gets overused to the point of losing all meaning. If just about everything gen AI makes is "slop", then it's just a pejorative for saying "AI generated output", not a word with real meaning about the characteristics of the composition. And if that's all it means, then as stated in point 2, it looks increasingly silly the better AI output gets in quality.

    What would be more grounded? Pointing to what seems wrong with a specific AI generation, even if that takes more effort to analyze and describe. For example, AI generated audio can have audio artifacts; that's something that can make it sound less pleasing to the ear than human created audio. But if an AI generated audio clip does not have noticeable artifacts, then it won't necessarily register as less pleasing. So I could not say that AI generated audio is intrinsically bad for this reason, but that when it has this problem, it is a noticeable reduction in listening quality.

  • source
  • parent
  • hideshow 18 child comments
  • [–] 17 points 2 weeks ago (5 children)

    This easily excludes a shitload of images drawn by humans that could now easily be called “slop” because somebody felt it looked too much like their idea of “AI”, which must be bad because it’s AI.

    As a Lemmy admin, I’ve already seen people report “AI slop” that wasn’t AI-generated. Many folks are not as good at discerning AI-generated images as they think, and it will only get harder as the technology improves.

  • source
  • parent
  • hideshow 5 child comments
  • [–] 14 points 2 weeks ago*

    Reminds me of that one tumblr post of an AI generated Monet style painting that got shit on by everyone on how obviously shitty and fake it is.

    OP then revealed the obvious: it was an actual Monet painting and people were trying to row their boats back so hard that the water had skid marks on it.

  • source
  • parent
  • [–] 13 points 2 weeks ago

    It would be funny to me if it wasn't so sad and painful. Too many people are in denial about the fact that machine generated stuff just isn't as obvious as they want it to be and I guess they can't cope with what that means about purely human created stuff, so they'd rather insist they can figure it out at a glance.

  • source
  • parent
  • [–] 6 points 2 weeks ago*

    A lot of it is just artists (wishing to make it their main career or already doing it) anxious or mad at AI for doing exactly that what „they“ told people in CS/Science fields AI would do to them-

    They deluded themselves that they are somehow exempt from the contempt of the markets, that they are special and safeguarded, if only they reach that mythical ladder step in their careers.

  • source
  • parent
  • [+] -5 points 2 weeks ago* (last edited 1 week ago) (10 children)
  • [–] 10 points 2 weeks ago (8 children)

    You just did a great example of everything wrong with the witchhunt anti-AI stuff. I wrote that from my own head and because of vibes, you think an LLM wrote it.

  • source
  • parent
  • hideshow 8 child comments
  • [+] -8 points 2 weeks ago* (last edited 1 week ago) (7 children)
  • [–] 11 points 2 weeks ago (6 children)

    Please stop being so confidently illiterate about how the technology works. It's the reverse and is the reason that people can't actually reliably tell what is AI generated at a glance. AI gen is trained on human works, so naturally it's going to look like human works sometimes and, in fact, the better the training process gets, the more it will look like them. Often the things that people most recognize as AI tells are AI tells because they are especially common in human works and so end up getting overtrained on. People then become overly sensitive to noticing these things on account of AI overusing them and go on to start hating perfectly normal literary structures and phrasing.

  • source
  • parent
  • hideshow 6 child comments
  • [+] -8 points 2 weeks ago* (last edited 1 week ago) (5 children)
  • [–] 5 points 2 weeks ago (4 children)

    So you don't even have to use AI, having looked at its output even once makes you "tainted" and everything you write "AI".

    One drop rule 2.0 — farce edition

  • source
  • parent
  • hideshow 4 child comments
  • [+] -5 points 2 weeks ago* (last edited 1 week ago) (3 children)
  • [–] 6 points 2 weeks ago (2 children)

    Where's the problem? You act like a fascist, I compare you to fascists.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 7 points 2 weeks ago (8 children)

    Also, here are a few examples since I mentioned how far some models have come. Are these all "slop"? Keep in mind I did some minor targeted correction on one (what is called "inpainting") but otherwise they are straight generated as is. Someone who is an artist by trade (not me) could I'm sure do more with this; better composition design, detail tweaking where necessary, etc.

  • source
  • parent
  • hideshow 8 child comments
  • [–] [S] 3 points 2 weeks ago

    I agree, these are images I didn’t even write the prompts for, I had an LLM create the prompts based on my worldbuilding and I changed literally nothing about the prompts or resulting images.

  • source
  • parent
  • [+] -2 points 2 weeks ago* (last edited 1 week ago) (4 children)
  • [–] 7 points 2 weeks ago (2 children)

    So are you telling me there is no merit as propaganda in the second image because of the quality problems in it? Or are you leaving that aspect out entirely. I notice you don't seem to have rated the 1st and 3rd at all with specific criticisms, yet saying you are "rating them Slop/10".

  • source
  • parent
  • hideshow 2 child comments
  • [+] -4 points 2 weeks ago* (last edited 1 week ago) (1 child)
  • [–] 8 points 2 weeks ago

    I'm sorry, what? Do you know anything about Chinese? The first one is not a meme, it's showing a grammatical rule/convention about 没有 and negation. It describes the rule itself on the whiteboard and then the teacher uses 有 in the process of asking "Do you have any questions?" To which the student replies, using 没 to negate 有, "I don't have any questions."

  • source
  • parent