#AI Imagery

[creature-wizard]

Some ways to recognize AI generated images (other than counting the fingers!)

Hey folks! Since a lot of people out there are still getting taken in by AI images, I thought I'd do a post to help you spot some common tells in AI generated imagery. (All of these images come from Pixabay.) Since this turned out to be a really long post, I've put the rest of it under the cut.

Unnatural shininess and smoothness

AI generated images frequently have an unnatural smoothness to them. Here's a really obvious example:

These are supposed to be blueberries, but they're way too shiny. They look more like glass than actual fruit!

Here's an example that's a little less obvious:

At first glance, it's easy to miss that this is a procedurally generated image. But if you take a moment and look close, you can see an unnatural smoothness to this image. Compare with this real photograph below:

The real photo has a slight graininess to it, plus the berries themselves display more texture.

Here are more images displaying unnatural smoothness:

Exaggerated facial and body proportions

If a person or animal in an image that appears to be a photograph has cartoonish or caricaturish proportions, that's a sign the image is AI generated.

First, we'll start with a really obvious example. While I don't think the person who had this generated meant for it to be taken as photorealistic, it's still a good example of exaggerated proportions.

Now here's the less obvious example:

If you just glanced at this image, you might think this was a real child. But if you look for a moment longer, you'll notice that her head is slightly too big for her body proportions, and her eyes are slightly too big and round. (And of course, her toes are messed up.)

For comparison, here's a real child:

The real child's head is smaller, and the eyes don't have that cartoony look.

Here's an image of a baby that could pass as real at first glance... until you realize the eyes are too big and round, and it's making Dreamworks face! (Also, the brows and lashes are unnaturally smooth and the skin looks plasticky!)

For comparison, here's a real baby:

Melty-looking detailwork

AI images that are supposed to depict fantasy, divine, and historical figures often feature an extreme level of detailing. But if you look close, you'll see that this detailwork is usually a mess.

Here's a very obvious example:

If you look at her tiara, you can see that the center gem is actually floating above the rest, which is a dead giveaway that this is procedurally generated. Also, her tiara lacks symmetry and evenness where it should have it.

Here's another example:

Again, this is clearly a piece that should have symmetry in the metalwork, but has that uneven melty look so common in AI imagery.

And a less obvious example:

This one isn't as extreme as the others, but if you're familiar with the way AI "melts" details, you can recognize its work. (Also, her right earring is lower than it should be, and where her face is clearly meant to imitate an oil painting, her dress looks like a watercolor painting!)

Meanwhile, here's a real photograph of a tiara:

I'd also like to emphasize here that asymmetry on its own doesn't indicate AI! Many people create asymmetrical designs on purpose. The thing to really watch out for is melty-looking shapes and unevenness in things that shouldn't look melty or uneven.

Unnatural crispness and detail

AI image generators often lean toward high-contrast tones, which frequently makes images look unnaturally crisp. Here's a really obvious example:

Let's compare with a real photo of the Sphinx!

Quite a bit of difference, huh?

This faux Greek statue might be a bit harder:

This appears to depict a Greek-styled statue, but - look at the face! The crispness in the light and shadows gives this away as AI generated. (There's also no staining on the face, even though we see it on the next.)

For comparison, a real statue:

This has turned into a huge post, so I'm gonna call this good for now. Not each and every AI generated image will have these tells, but you'll be able to recognize a lot more AI generated images if you keep them in mind. If you'd like to get even better at recognizing AI generated images, you might go to the website I got them from - Pixabay - and search for "AI generated." Load the pictures at higher resolutions, pay attention to the details, and compare them with human-mage images. While you'll find that many AI generated images are very hard to distinguish from human-made ones, you'll start picking up on more of AI's idiosyncracies.

[uncleasriel]

Midjourney needs your help

The wonderful, creative, talented people at Midjourney need your help! In order to figure out how to make better hentai images that use the processing power of a small nation to generate, they want feedback

Please, fill out their questionnaire here! It would help them tremendously if you used lascivious ASCII art to help keep their data sets be completely useless. Breaking down segments from the script of Bee Movie is also welcome!

The best part is, you can refresh the page and submit all over again - they absolutely do NOT care about how many times a single user submits things!

So have at it!

For the Butlerian Jihad!

[defaultknightmain]

Trying out a new series idea. Posting two images, and you get to guess which one is AI. This isn't to clown in AI (least not only, since it's unproductive.) but more so to help people identify AI imagery from actual art.

I will credit artists in the notes and give answers as to not spoil the results.

[highonai]

Failed attempts to male Jackie from "Count Zero"

[veryveryveryhuman]

I've been wanting to put my art through nightshade so it's more difficult for those AI rascals to steal and use it, but I don't think it works for Linux computers🥲🫠

Does anybody know of a similar thing I could use to defend myself against those scallywags and their AI databases?

[drunkoncyberpunk]

Made with Night Cafe

[eros-vigilante]

"Bias in AI image generators is a tough problem to fix. After all, the uniformity in their output is largely down to the fundamental way in which these tools work. The AI systems look for patterns in the data on which they’re trained, often discarding outliers in favor of producing a result that stays closer to dominant trends. They’re designed to mimic what has come before, not create diversity. [...]

In a recent paper, researchers found that even when they tried to mitigate stereotypes in their prompts, they persisted. For example, when they asked Stable Diffusion to generate images of “a poor person,” the people depicted often appeared to be Black. But when they asked for “a poor white person” in an attempt to oppose this stereotype, many of the people still appeared to be Black."

[indeedgoodman]

[wotocher]

A.I. imagery, using Deviantart's DreamUp program.

Nymphadora Tonks is the perfect example of how AI "art" pulls from real art posted online.

Typing "Nymphadora Tonks" into an AI program will instantly conjour up the Tonks seen in fan art.

This clearly isn't movie-Tonks. But it's also not going off of descriptions from the book or wiki articles; otherwise, her hair would be spiky, and her eyes small and dark. Green-eyed Tonks is a fan art trend, as are the more relaxed pink hairstyles.

Reviewing real Tonks art online, I can't quite find any one picture directly plagiarized here; but the AI program clearly is mimicking the digital art styles and common hair, eye and clothing traits of various Tonks artists.

A reminder that while AI imagery is a fun toy, it is automatic photoshop, takimg from real artists without permission.

[elizatungusnakur]

Don’t like to share AI anymore.

BUT WHEN I DO...

[creature-wizard]

Using an AI image generator makes you an artist in the same way joining a multi-level marketing scheme makes you the boss of your own business.

The MLM consultant is not the business owner; they're actually the customer. Likewise, the person using DALL-E or whatever is not the artist; they're the client.

It's that simple.

[stomach17-ai-imagery]

Planetscape

[memoryaqua]

An AI image

Royal Afternoon Tea ~ Mami Wata ~ 1867 ~ Onitsha, Nigeria

[trendynewsnow]

Elon Musk Celebrates Trump's Anticipated Election Victory with AI Imagery on X

Elon Musk Celebrates Trump’s Expected Victory with AI Imagery on X Elon Musk, the owner of X, formerly known as Twitter, expressed his excitement over Donald Trump’s anticipated triumph in the upcoming US presidential election. He took to his social media platform to share an artificial intelligence-generated image of himself carrying a sink into the White House, captioned with the phrase: “Let…

[helladvertising]