This AI Can Generate An Image From Any Prompt, and the Results Are Hella Weird
Jun 15, 2022 • Meryl Medel
Jun 15, 2022 • Meryl Medel
The idea of artificial intelligence can be scary. Even just seeing it on fictional media could probably already incite a whole bunch of existential horror in many of us (more Black Mirror, anyone?). But sometimes, an AI program can keep things simple and funny — such as in the form of meme-y photos. If you have been seeing an influx of weird, freaky images on social media, it’s the work of an AI called DALL·E mini.
Created and developed by Boris Dayma, DALL·E mini is an artificial intelligence program that can generate a set of images from any prompt you give. DALL·E is a play on the names of renowned Spanish surrealist artist Salvador Dali and the Disney trash-cleaning robot WALL-E.
The original iteration of DALL·E was created by OpenAI. “DALL·E mini is an attempt at reproducing those results with an open-source model,” says the developer.
Let’s be real, DALL·E mini probably has a lot of complicated programming involved that only developers might be able to understand. But in simple terms, the AI essentially looks at millions of images on the internet and their associated descriptions and captions and draws a new image from those millions of images based on the text prompt it receives. The AI continues to learn and is expected to improve over time.
A lot of images generated from the DALL·E have gone viral on social media, which isn’t surprising, seeing as many prompts are downright ridiculous. Here are some of the silliest we’ve seen so far:
Grimace at the Capitol Riot
— Weird Dall-E Generations (@weirddalle) June 13, 2022
Thanos looking for his mom in a Walmart
— Weird Dall-E Generations (@weirddalle) June 11, 2022
A bottle of Ranch Dressing testifying in court
— Weird Dall-E Generations (@weirddalle) June 11, 2022
Demogorgon as a guest star on Friends
— Weird Dall-E Generations (@weirddalle) June 13, 2022
🧑🎨 @max_elbo a garden full of flowers #elbo_ai #ai #not #dalle #digitalart pic.twitter.com/LAgrGJNOuO
— 🧑🎨 max (@max_elbo) May 10, 2022
According to the developer, DALL·E Mega is the “largest version of DALL·E mini trained.” There is a lot of programming jargon involved, but based on the sample predictions the mega version has run, it would provide more accurate, more sophisticated images compared to the mini. You can test it through @max_elbo on Twitter.
Yes, it does! Since the AI uses associated descriptions and captions in generating images, it seems DALL·E mini can generate images from prompts in any language — as long as that language is on the internet. Here’s a prompt in French:
One of the things people don’t know is possible pic.twitter.com/H0vNz4Q7MA
— Boris Dayma 🥑 (@borisdayma) June 13, 2022
The exact answer is unclear. According to the developer, the images “come from public datasets (Common Captions and YFCC100M) with their own license”, so some images the AI uses may be free to use, but others may have licenses that come with it.
All training images come from public datasets (Common Captions and YFCC100M) with their own license but I’m not knowledgeable enough to give any advise
— Boris Dayma 🥑 (@borisdayma) August 5, 2021
Do you want to try the AI program yourself? You can head on to their website. But you should expect to encounter errors on the website. DALL·E mini and its generated images have gone so viral that the website frequently crashes since there’s “too much traffic.” If you’re getting that error, you simply need to refresh the website every few minutes. Or you can also wait for the larger mega version.
Take a screenshot of your set of generated images on the AI and share it with your friends on social media. Your prompt might just be the next viral DALL·E image.
We now have a screenshot button 🔥 pic.twitter.com/vlxoHRgVXC
— Boris Dayma 🥑 (@borisdayma) June 14, 2022
Follow us on Facebook, Twitter, Instagram, Tiktok, and Youtube for the l8est entertaining, useful, and informative lists!
Input your search keywords and press Enter.