Does Generative AI Have a Style?

 

The Myth of the “AI Art Style” 

There’s a persistent idea floating around that generative AI has a distinct, recognizable “style”. Something glossy, over-rendered, a little too perfect, a little too generic. You may have heard this yourself or you may believe it to be true based on what you’ve experienced directly from AI.

I don’t think this is entirely true though. Not in the way people mean it.

Most AI generated images have a 'tell' that identifies them as AI made, (AI doesn’t create the same way a human does) but in terms of actual style, AI is definitely not a one trick pony show. 

What most people are calling an “AI style” isn’t a built-in aesthetic or a default mode. It’s something else entirely; something shaped by trends, habits, and popular culture.

Let’s unpack that.

AI doesn’t choose a style as it's own default, nor is it programmed that way. It learns patterns from massive datasets of existing artwork and generates new images based on an average aesthetic.

What that means in practice is simple: AI tends toward what is common.

Not the weirdest work. Not the most experimental. Not the most niche. The most common.

Generative AI is averaging out the data based on what it knows, what it has been trained on. Collectively, this is what WE have fed the machine.

If AI outputs often do look similar, it’s worth asking then, who’s actually driving that?

Here's the thing, millions of users rely on the same prompt shortcuts using popular genres, trending artists or aesthetic buzzwords. These prompts get shared, copied, optimized, and repeated at scale. 

AI isn’t enforcing sameness. Users are reinforcing it.

We’re seeing a skewed version of what AI can do because users are choosing those styles on repeat.

Public perception is shaped by exposure, and right now, exposure is heavily filtered in a certain direction or two.

So when people say “AI art all looks the same,” what they really mean is, the AI art I keep seeing looks the same.

That’s not the same thing.

The following clipart examples show a few of the more popular styles that are seen much more often right now. While the styles are each very different, they are more representative of popular culture right now. 

 

 

In this next set of examples we have some maybe lesser used styles that are also very different. Each of these, above and below, was made with a relatively simple prompt and all with the same AI model. Style is therefore a choice selected by the user.

 

 

A couple things to note, any similar look you may be seeing is usually a result of prompt quality in combination with the choice of AI model used to generate the image. Different AI models have different strengths and outcomes can vary wildly.

Ironically, generative AI can stretch far beyond the stylistic averages it may tend towards, if the user prompts with more detail, more imagination and greater creatively from the human side of things. Lazy prompting from non-creative users generates more boring, same look images. Any tool in the wrong hands can produce sub-standard results.

I have more examples before but first I want to address one other point that comes up quite often.

That AI images lack something inherently human, that they are soulless. I'm not sure about that though I understand the point. I think it depends on what sort of 'art' you are comparing it to and in what setting. Would I pay $$$ for AI art to hang in my home? Not likely. I want the human touch there. But I'm not above using clever AI produced imagery for something like greeting card graphics.

I think a lot of people are reacting adversely to AI for good reasons. How these companies built their programs is ethically questionable and potentially a legal infringement of rights. This needs to be addressed and bad actors held accountable.

Now that AI is more embedded into society we are seeing so much 'AI slop'. This is the low value mass produced work that exists for the sake of it and usually not with good intentions. This is also a problem that needs addressing.

And then there is the AI replacing humans conversation. Which is utterly insane and we should all definitely reject any such ideas. 

But a thought I'm having lately on that point is - can human endeavor include AI as a tool in ways that enhance our creativity and outputs without actually replacing us or corrupting our humanity?

As importantly, could we start to judge AI usage based on how/when/why it's used and not an absolute yes or no to its usage full stop? I'll explore these ideas more at a later date. I think this is going to be a big conversation in the years to come.

For now, let's look at what kind of range AI has stylistically. This isn't comprehensive, just a quick snapshot within the illustrative area I've specifically been working.

This set of examples were all made using Night Cafe's Flux 2 Klein model which happens to be pretty good at character and composition consistency. It enabled me to take the same image idea and simply swap out the style part of the prompt to create different looks. 

The following represent MY choices and not the AI's choices for theme, topic and colors.

 

 

These were fun to make because it challenged me to explore how different styles are applied by this particular and where I might be able to take certain ideas further for something more unique.

After the above set I hopped over to Midjourney and created with a different kind of prompt because I wanted to show you what some AI models do by default if you use a really simplistic prompt. The next group was created with the prompt "a painting of a fox in a suit". I'm letting AI make almost all the choices for me. There’s no human creativity here. 

Midjourney does offer a more diverse range of options than some models when given simple prompts, because it's programmed to do that. But because it wasn't given any direction there's not a really big stylistic difference here. This would the aesthetic average from its training dataset and user popularity.

I think there is definitely some sameness happening here.

 

 

If I adjust my prompt to add a very simple tweak adding just an art style I get a completely different result.

 

Each of these had a variation of the prompt "an impressionist painting of a fox in a suit". I just swapped the art style for each image to get different looks. This is a great next step but it's still a very simple prompt. You can definitely work the image into a variety of style options, mash styles together and get specific on details to create something less homogenized.

 

 

Maybe something more like this? 

Midjourney again, this time describing the image in just a bit more detail.

"a fox, sitting in an armchair, wearing a suit like an old fashioned gentleman. a rough abstract sketch in editorial cartoon style, on a white paper background"

 

 

 

 

 

Right now I think everyone is using the label "ai style" as more of a shorthand for stuff that feels generic, mass-produced, over-polished and maybe lacking personality. Because that's what we're seeing a lot of.

If everything AI generates looks the same, it’s worth asking if it's really a limitation or a reflection of the culture around AI use.

Can we do better? Yes, I think we can.

 

Let me know in the comments what you see in these examples. These are not highly original or anything particularly creative, I just threw together an idea to quickly show range. So try to critique based on that.

Do you still think AI has a "look" or can you see beyond mass production to something that has more to offer?