Nightmare fuel is everywhere, particularly online. The latest source: Dall-E Mini, an AI tool capturing attention on social media thanks to the weird, funny and occasionally disturbing images it creates out of text prompts.
Dall-E Mini lets you type a short phrase describing an image, one that theoretically exists only in the deep recesses of your soul, and within a few seconds, the algorithm will manifest that image onto your screen.
Odds are you've seen some Dall-E Mini images popping up in your social media feeds as people think of the wildest prompts they can -- perhaps it's Jon Hamm eating ham, or Yoda robbing a convenience store.
This isn't the first time art and artificial intelligence have captured the internet's attention. There's a certain appeal to seeing how an algorithm tackles something as subjective as art. In 2016, for example, actor Thomas Middleditch made a short film based on a script written by an algorithm. Google has produced more than a few tools tying art and AI together. In 2018, its Arts & Culture app let usersin famous paintings. Or will figure out what you're trying to doodle, and fix it up for you.
There are other text-to-image systems, like OpenAI's Dall-E 2, as well as Google's Imagen and , which the tech giant isn't releasing to the masses.
Here's what you need to know about Dall-E Mini and its AI-generated art.
What is Dall-E Mini?
Dall-E Mini is an AI model that creates images based on the prompts you give it. In an interview with the publication I, programmer Boris Dayma said he initially built the program in July 2021 as part of a competition held by Google and an AI community called Hugging Face. Dayma didn't immediately respond to a request for comment.
How does it work?
Anyone can type in a prompt and hit the "run" button (though you're likely to get an error message about traffic to the tool and have to try again). Dall-E Mini will spit out its results in the form of a 3x3 grid containing 9 images. A note about the tool on its website says it was trained on "unfiltered data from the internet."
How good is the AI?
Unsurprisingly, Dall-E Mini is a little hit or miss. In the interview with I News, Dayma said the AI is better with abstract painting, less so with faces. A landscape of a desert is quite pretty. A pencil sketch of Dolly Parton looks like it might steal your soul. Paul McCartney eating kale will take years off your life.
Dayma did say, though, that the model is training (that ability to learn is one of the things people love – and fear – about AI), which means it can improve over time. And with the viral popularity of Dall-E Mini, the point is to stumble upon the most bizarre image you can think of, not necessarily to get a perfect impressionist rendering of a Waffle House. The fun is more about dreaming up with most outlandish images that don't exist -- that perhaps shouldn't exist -- and bringing them into cursed existence.
Dall-E also has a note saying that image generation could have a less fun side and could be used to "reinforce or exacerbate societal biases."
Is Dall-E Mini related to Dall-E 2?
No, they're not associated. Dall-E 2 is also a tool for generating AI images that was launched as a research project this year. It was created by the AI research and deployment company OpenAI and is not widely available.
What kinds of images are people creating?
On social media, you can find an abundance of strange Dall-E Mini creations, from Thanos in a Walmart looking for his mother, to Jar Jar Binks winning the Great British Bake Off. Here are some other highlights.