A picture is worth a thousand words, but according to DALL-E 2, an artificial intelligence application, you can get a professional-looking image with many less.
DALL-E 2 is a new neural network method that lets you make an image from a short word or text. The initiative, which was launched in April 2022 by the artificial intelligence research facility OpenAI, has not yet been made public. But a small but growing number of people, including me, have been given permission to try it out.
Using DALL-E 2 is similar to looking for a picture on the internet: you put a brief word into a text field, and it returns six photos.
Instead than using photos from the internet, the computer generates six entirely unique images, each of which reflects some variation of the input word. (Until recently, the application generated 10 photos per each question.) For example, when some friends and I gave DALL-E 2 the word prompt “cats in devo hats,” it generated ten photos in various styles.
Almost all of these may be mistaken for professional images or drawings. While the algorithm could not exactly understand “Devo hats,” the unusual helmets used by the New Wave band Devo, the headgear in the photos it generated came close.
Over the last several years, a tiny group of artists has been employing neural network algorithms to create art. Many of these pieces have particular characteristics that make them nearly seem like actual photographs, but with strange distortions of space—a kind of cyberpunk cubism. The most contemporary text-to-image technologies often generate dreamy, surreal images that may be enjoyable but seldom seem genuine.
DALL-E 2 represents a considerable improvement in picture quality and realism. It may also imitate certain styles with incredible precision. It will generate six lifelike photos if you desire images that seem like genuine photographs. If you desire ancient cave paintings of Shrek, it will create six images of Shrek that seem like they were made by a prehistoric artist.
It’s mind-boggling that an algorithm can do this. Each picture collection is created in less than a minute. Not all of the photos will look good, and not all of them will be what you were thinking of. Even if you have to filter through a lot of outputs or experiment with various text prompts, there is no other way to get so many amazing outcomes so rapidly—not even by employing an artist. And sometimes the most unexpected outcomes are the greatest.
In theory, anybody with sufficient money and experience may create a system like this. HuggingFace, a company, is openly building its own version that anybody can test right now on the web, however it isn’t as good as DALL-E or Google’s system.
It’s easy to see how these technologies could change how people make and share pictures, whether they’re memes, greeting cards, ads, or even works of art.