That's because the people putting in the prompts, aren't even good at envisioning.
I'd imagine that they say things like "uh, put guy here and make them look strong. also add some aggressive font about aspiring to weight lift" something like that and they accept it.
I think it's the default style. The style formed when you merge all styles on earth. And most people don't care enough/feel confident enough/know enough to tell the AI to do it different.
They're trained on the same information
Good creative and artistic design obviously wasn't part of it, though.
That's actually because people doing AI art aren't artists, they have absolute no artistic vision and most probably lack basic art knowledge, and the very reason they are doing AI art is because they are mediocre. Their prompts are "give me [actor] doing [action] in [location]" and that's it, they don't talk about style influence, color palette, scene composition, etc etc, so it all comes out the same.
This. That's like vibe coding. Vibe arting get you slop as well.
Now, give an AI to a good coder, and you don't get slop code. Same with artists
Do you still get AI stuff? Yes, but a whole better result.
Except good artists and coders enjoy the creation process, so they won't use it like that.
I need to change every instance of this function to another? Sure, let an LLM do it, that's tedious to do and easy to verify. Implement a new system? Hell no, that's interesting to do and hard to verify. I imagine artists have similar feelings on the matter, but I don't know what's a tedious task that's easy to verify for an artist.
if you are just doing a photo composing or a design, I guess "remove fake png background" "remove watermark" hehe
(although content aware tool already does a great job)
Yes indeed, I work in printing and use AI regularly. But I use it as a tool to enhance things, not to start something from scratch. Square picture needs to be A4? I let it generate the extra space around. Foot cut off on the bottom? AI adds it (after a couple of attempts sometimes). Or upsaling can also do miracles nowadays. But for text you really can't use it. If there is text generated, I generate background over the text, and later add proper text in indesign.
Came here to say exactly this. Well put.
It’s the most statistically average art.
It's literally that.
Trained on everything, biased towards middle-of-the-road, AI is going to produce the most generic "art" possible, unless specifically prompted to do something else.
That's what really gets me about everyone who touts AI as the future of creativity. All it does is give you what it determines to be the most likely response to your prompt. It's never going to be clever or subversive.
Even if you wade through all the slop it shits out and only go with the results closest to what you asked for, the absolute best you can hope for is predictable and mediocre. If you're not interested in making something at least halfway decent, why bother at all?
ClipArt
Generative AI tools are by definition averaging machines. They take lots of data and give an average output based on input.
Not all training data is as valuable as others. The most valuable are distinct and highly associated with a certain tag or description. So a picture that is just a dog with over exaggerated dog features is better training data than a whole landscape with tons of stuff including a dog. This makes cartoonish stuff a better fit for training data.
Similarly, people using these tools tend to be less detail oriented and just want something that has an obvious, center stage, flat message. They want something fast and cheap so far less thought goes into it than traditional art.
Given these two, you end up with a huge bias towards the bland cartoonish style.
Beyond that, there is also the fact that as time goes on, generated images find their way into training data, so it keeps getting more closer to that bland style.
Seriously, I feel like the first generation of stable diffusion models were superior to what we have now.
I feel that "generative AI is by definition an an averaging machine" explains AI better than any other definition.
It kills out the extremes of creativity
The organic meatbags seem to enjoy it.
“AI ain’t got style.” Lol
It doesn't - you only notice the low-effort ones that default to the generic style where as the rest fly under your radar and go undetected. It's called a toupee fallacy: "All toupees look fake, I've never seen one that didn't"
Is that like inverse survivor bias?
It's like confirmation bias which I think survivor bias is a type of.
So is it still possible to give a style prompt? I haven't dabbled with it in ages. When it started out, you'd tell it you wanted, say, the style of Rembrandt but with a dash of horror or something. Can you still do that?
it's a language model. You can give it any kind of prompt you want. How effectively it interprets those instructions is a separate discussion, but in general newer models expand on those sort of capabilities, they don't strip them out.
Aight. I just tried again for science. I gave it very detailed instructions and it produced a person with three arms. I did not specify three arms.
Did you specify not three arms
I like how this sounds kinda like a bratty response, but is legitamte advice.
If you're not aware, there is the concept of "negative prompts" with image generation where you basically tell it not to draw something ugly or wrong, but also can try to supress certain things too like three arms.
"Tried again" is pretty vague. There are millions of different text-to-image models and adapters and millions of different ways of configuring and running them and thousands of different services providing these technologies at different service tiers. I'm going to go out on a limb and assume you're using a free service, which is obviously going to be the worst possible option using the cheapest possible model, there are likely self-hosted free models that can do better. You can't just assume you're using a newer model if you don't even have access to describe what model you're running.
Because that’s the data average. LLMs don’t think, they just algorithmicly process data. It is literally impossible for them to create, let alone create something original.
I don't know where y'all seeing all this ai adverts. Social media? YouTube ads?
But I hardly see any of it.
I see it primarily in real life actually, many paper flyers and sandwich boards have a clearly ai generated style, and if I go to a farmer's market in the city at least 20% of the graphics are AI.
These are more neat than the ones I see often, but if you want examples: https://redlib.catsarch.com/r/graphic_design/comments/1w1xve1/my_neighborhood_has_been_taken_over_by_chatgpt/
I see what you're saying thank you for the examples that definitely clears up a lot.
I guess I just kind of tune these out in public but you're absolutely right they do have a very similar styling.
IDK, I started blocking ads on every platform 20 years ago and haven't looked back. Nothing gets through in any app on every device I own (PC, phone, watch, TV, etc.). I will go out of my way to make sure that an ad can never get to my eyeballs using every available tool at my disposal: browser extensions, alternative DNS servers, patched apks, service menu tweaks—you get the idea.
I can't wait until we have AR glasses that can adblock billboards and other ads in public too. We're less than 5 years away from that tech being mainstream IMO.
This was for general diffusion model outputs a few years ago, but a lot of the speculation of why the outputs are similar likely hold here
There are also only so many good data sets available for people to use to build image models, Phillip Isola, a professor at the MIT Computer Science & Artificial Intelligence Laboratory, told me, meaning the models might overlap in what they’re trained on. (One popular one, CelebA, features 200,000 labeled photos of celebrities. Another, LAION 5B, is an open-source option featuring 5.8 billion pairs of photos and text.)
[...]
Five years ago, he explained, image generators tended to create really blurry outputs. Researchers realized that it was the result of a mathematical fluke; the models were essentially averaging all the images they were trained on. Averaging, it turns out, “looks like blur.” It’s possible that, today, something similarly technical is happening with this generation of image models that leads them to plop out the same kind of dramatic, highly stylized imagery—but researchers haven’t quite figured it out yet. Additionally, “most models have an ‘aesthetic’ filter on both the input and output that reject images that don't meet a certain aesthetic criteria,” Hany Farid, a professor at the UC Berkeley School of Information, told me over email. “This type of filtering on the input and output is almost certainly a big part of why AI-generated images all have a certain ethereal quality
[...]
The third theory revolves around the humans who use these tools. Some of these sophisticated models [this is not as sophisticated as the author is implying] incorporate human feedback; they learn as they go. This could be by taking in a signal, such as which photos are downloaded. Others, Isola explained, have trainers manually rate which photos they like and which ones they don’t. Perhaps this feedback is making its way into the model
[...]
This could be intentional: If such imagery has a market, maybe companies would begin to converge around it. Or it could be unintentional; companies do lots of manual work in their models to combat bias, for example, and various tweaks favoring one kind of imagery over another could inadvertently result in a particular look
That's interesting. I always thought it was a conscious choice by model trainers.
They all trained on the same info. For generative models, "more data is better" is pretty much a global truth. And when you have a few dozen providers all consuming as much data as possible, they eventually all process all the available data with the same algorithm and all land on roughly the same product.
And that's before you consider that in practice, people are only generating assets from a couple providers. So it's literally the same product, not just two similar products.
I would imagine that it's the easiest for the model to produce. Bold colours, clear and distinct shapes, that sort.
It doesn't help that the companies have scraped the entire internet (which is a can of worms in terms of both legality and ethics) and then some, so it will mimic the advertising strategies employed everywhere (again, bold colours, clear shapes, etc.)
I would assume, they even explicitly tell the model to use this default style, if the user doesn't specify anything else, because it is so easy to generate.
At least, this started early last year and there were fairly distinctive points when each chatbot's outputs started looking all the same.
You are asking: "why are all AI users the same kind of stupid and want it this way - at least all the ones that I am watching"
!nostupidquestions is a community dedicated to being helpful and answering each others' questions on various topics.
The rules for posting and commenting, besides the rules defined here for lemmy.world, are as follows:
Rule 1- All posts must be legitimate questions. All post titles must include a question.
All posts must be legitimate questions, and all post titles must include a question. Questions that are joke or trolling questions, memes, song lyrics as title, etc. are not allowed here. See Rule 6 for all exceptions.
Rule 2- Your question subject cannot be illegal or NSFW material.
Your question subject cannot be illegal or NSFW material. You will be warned first, banned second.
Rule 3- Do not seek mental, medical and professional help here.
Do not seek mental, medical and professional help here. Breaking this rule will not get you or your post removed, but it will put you at risk, and possibly in danger.
Rule 4- No self promotion or upvote-farming of any kind.
That's it.
Rule 5- No baiting or sealioning or promoting an agenda.
Questions which, instead of being of an innocuous nature, are specifically intended (based on reports and in the opinion of our crack moderation team) to bait users into ideological wars on charged political topics will be removed and the authors warned - or banned - depending on severity.
Rule 6- Regarding META posts and joke questions.
Provided it is about the community itself, you may post non-question posts using the [META] tag on your post title.
On fridays, you are allowed to post meme and troll questions, on the condition that it's in text format only, and conforms with our other rules. These posts MUST include the [NSQ Friday] tag in their title.
If you post a serious question on friday and are looking only for legitimate answers, then please include the [Serious] tag on your post. Irrelevant replies will then be removed by moderators.
Rule 7- You can't intentionally annoy, mock, or harass other members.
If you intentionally annoy, mock, harass, or discriminate against any individual member, you will be removed.
Likewise, if you are a member, sympathiser or a resemblant of a movement that is known to largely hate, mock, discriminate against, and/or want to take lives of a group of people, and you were provably vocal about your hate, then you will be banned on sight.
Rule 8- All comments should try to stay relevant to their parent content.
Rule 9- Reposts from other platforms are not allowed.
Let everyone have their own content.
Rule 10- Majority of bots aren't allowed to participate here. This includes using AI responses and summaries.
Our breathtaking icon was bestowed upon us by @Cevilia!
The greatest banner of all time: by @TheOneWithTheHair!
