Why The Internet Is Raving About DALL-E Mini, A Free AI Tool That Converts Any Text To Art
Artificial intelligence (AI) is often as amazing as it is terrifying. It has nearly boundless potential — potential which also makes some people feel uneasy. Whether you love AI or you find it scary, DALL-E mini is a tool that will most likely only reinforce your opinion. Using artificial intelligence, data scraping, and billions of images from all over the internet, DALL-E mini is able to turn text into images, and this includes even the weirdest requests that you could ever come up with. It's no wonder that the internet at large loves it right now — but how exactly does it work? And what can it do?
On the example shown above, we typed "dog driving a car" into DALL-E mini's prompt engine. The service is so popular right now that it's often busy, but it can be used free of charge, so after a few minutes, we were able to receive our images. Nine squares, each with what the AI believes is best fit to picture a dog driving a car. When you look at these images closely, they look all kinds of wrong — and it only gets worse if you enter the name of a real person instead of an animal. However, when seen from afar, they are pretty much what you'd expect to see: different types of dogs in different types of vehicles.
The internet is hyped up on DALL-E mini
Although DALL-E mini can certainly create some pretty pictures, it works based on math and algorithms with no artistry involved. The AI simply looks at countless images that best correspond to your query and then identifies patterns that are found in the majority of them. This includes shapes, colors, and captions. Through that, it learns how to create an image that fits your description, and voilà, there's your dog driving a car. But of course, the internet has even better (often disturbing) ideas of what to do with DALL-E mini than your everyday average dog driving a car.
You heard it here first: “NASA to research UFOs†#dalleminihttps://t.co/1mWiVG1naa pic.twitter.com/VOTmk0IUnz
— Dall-E News (@DallENews) June 10, 2022
In the last few days, Twitter has been flooded with screenshots of DALL-E, showing just how much the software is growing in popularity right now. Some queries are innocent enough, and to DALL-E's credit, they are fulfilled very well. As shown in the example above, "NASA to research UFOs" returned great results that look like a mix of the actual artwork and "The X-Files." However, when it comes to showing actual faces, DALL-E falls a little flat, and the results can be disturbing as well as hilarious.
Uh oh #dallemini #weezer pic.twitter.com/85oVVK4Y3I
— Milkiiyo (@fromhitTVshow) June 6, 2022
The tweet above marks a good example of how disturbing this tech's output can be. The user searched for "Weezer in the backrooms." If you stand very far from your screen and squint, it might look fine, but upon closer inspection, it's complete nightmare fuel. Even in such cases, it's certainly impressive what DALL-E is capable of. However, it's worth noting that, although the tech is certainly advanced, it's not as innovative as it might seem.
DALL-E mini is one of many
The version of DALL-E shown above, hosted by Hugging Face and called "DALL-E mini," is based on an open-source project from OpenAI. Google has its own working version of its user-facing software called Imagen, also using OpenAI to output next-level imagery based on textual prompts. Unfortunately, Google doesn't have as much of a demo as what we can get from DALL-E mini, but you can see some of what Google suggests are AI-generated results here. Google's creation is extremely photorealistic, though we can't be sure how much processing was done behind the scenes before each creation was ready for public viewing.
Perhaps the most interesting example of this text-to-images AI demonstration trend is NVIDIA's GauGAN2, a software that utilizes artificial intelligence and machine learning to create photos based on textual prompts. The resulting images, shown off by NVIDIA, are so lifelike that it's hard to believe a machine made them based on a few lines of text. GauGAN2 also has another kind of lead over DALL-E — it also offers Draw plus Text-to-Image that lets you draw something super simple and see it become something akin to a photograph. Much like DALL-E, NVIDIA's GauGAN2 is free to try right now.
If you're more into the abstract side of things, you might like StarryAI. It works in a very similar way to DALL-E, but instead of trying to create something plausible, it creates wildly unnatural images that make no pretense about being mistaken for photos. NightCafe makes dream-like images based on your prompts, and you can choose from over a dozen of different styles such as "cyberpunk" and "fantasy." There's also AI Art Maker that seems to dip into a different side of art by creating slightly bizarre images, once again based on textual prompts. As you can see, even though DALL-E mini is swiftly becoming a Twitter favorite, there are plenty of other options, too.