The AI-Generated DALL-E Images Keeping Us Awake at Night

The AI-Generated DALL-E Images Keeping Us Awake at Night
DALL-E. Image: Zachariah Kelly/Gizmodo Australia

My god, you guys are doing some messed up things with the DALL-E image generator.

If you’ve been living under a rock, DALL-E is an AI image generator, able to create almost nightmarish (and often, humorous) images from a string of text. You can type out any assortment of text that you’d like and DALL-E will have a go at it. Gizmodo in the U.S. had a lot of fun creating a Seinfeld-fuelled nightmare using the tool, but there’s a lot more that people have done with the tool than recreate Seinfeld in a dream state.

Take, for example, the “Weirddalle” subreddit, dedicated to the most unhinged creations from the tool. This is the kind of subreddit that makes me worried for internet users, however a lot of it is incredibly funny.

We’ve decided to bring you some of the funniest DALL-E images going around.

Willem Dafoe Pez dispenser

Yeah, that’s creepy. Also, kind of accurate, right? Like, that’s Willem Dafoe’s head and that sure is a pez dispenser, positioned like heads would be on the toy-cross-snack dispenser.

Let’s talk a little bit about how DALL-E works. DALL-E is a trained neural network that translates the input words into characterisable, associated images. From these images, the neural network churns out nine pictures based on the prompt, all stemming from its vast database of word-to-image associations.

“The model is trained by looking at millions of images from the internet with their associated captions. Over time, it learns how to draw an image from a text prompt,” the webpage for DALL-E mini (the one everyone’s using) states.

“Some of the concepts are learnt from memory as it may have seen similar images. However, it can also learn how to create unique images that don’t exist such as ‘the Eiffel tower is landing on the moon’ by combining multiple concepts together.”

That’s the basics of it. It’s a very smart tool, but we’re using it to create these shitposts (well, technically, a miniature version of the tool called DALL-E Mini). Hence, Willem Dafoe Pez dispenser.

CCTV of PS1 Hagrid stealing a washing machine

Yeah, you read that right. The DALL-E shitposting community has this obsession with PS1 Hagrid, the Harry Potter character, specifically from the PS1 game Harry Potter And The Sorcerer’s Stone. It’s seemingly half because of the obscurity of having PS1 Hagridof all characters, in obscure scenarios like “stealing a washing machine”, but I reckon another part of it is how low-poly PS1 Hagrid was.

You could put PS1 Hagrid into found footage or CCTV footage and, because of the low quality and pixel resolution of those associated filming types, he’d kind of look natural there. Or, at least, according to DALL-E. Here’s another one.

dall-e image
Image: Knowyourmeme

November 5, 2024

So uh, yeah… Huh.

Eery, for sure, but to be honest I’m not quite sure how or why DALL-E generated these pictures.

It’s more than likely that this one’s a fake, actually. As said by one reddit user: “What I dislike most is that it’s probably not really the prompt the OP gave to get this image. That input pretty much always gives nice images of Fall. OP should be ashamed of themself.”

Debunked.

But, if it was legit, November 5, 2024 (the date of the next U.S. presidential election) appears to spit out quite the anomaly. I don’t have an explanation for this one.

I’m creeped TF out.

Duolingo trail cam

Everybody’s afraid of the Duolingo owl, and now your fears have been confirmed with this totally real trail cam footage.

“Trail cam footage” is a common prompt to use in DALL-E, which generates in low enough quality that it being from a “trail cam” makes sense. Users like to pair trail cam footage with monotonous objects, like washing machines or even PS1 Hagrid as earlier mentioned.

But pairing it with the Duolingo owl feels right. The owl is coming for you. Say “mercy” in Spanish, or you’ll be owl food.

1993 Renault Twingo in Skyrim

Able to identify the aesthetic and shape of the (not-so) classic 1993 Renault Lingo appears to be something DALL-E does with ease, and plugging it into the artstyle of Skyrim is done just as easily.

Again, following on from what we were saying about polygons and resolutions, a Skyrim filter appears to be done quite easily. If you squint your eyes, it looks like maybe the city of Riften.

What’s your favourite DALL-E image?

Feel free to send me your favourite DALL-E image or creation.

I love AI-generated horror.