OpenAI, a company founded by Silicon Valley heavyweights like former YC president Sam Altman and Elon Musk, has unveiled an update to its AI system that generates original images in varying styles and formations based on text queries.
DALL-E 2, a name blending artist Salvador Dali and the Pixar robot WALL-E, is the second iteration of the system and has the power to create an image from erratic combinations of commands like "a bowl of soup that looks like a monster knitted out of wool." It can even create different variations of that concept, you know, for people that need that kind of thing.
DALLE-2 is not yet available in OpenAI's API or available to the public to fool around with, but there is a waiting list for those eager to try it out.
"We don’t have a set timeline for when we will release this broadly," an OpenAI spokesperson told me over email. "We’re still in the discovery phase and are previewing the research with a limited number of trusted users to continue understanding DALL·E 2’s capabilities and limitations."
The product could one day be used by artists, architects, product designers and magazine cover designers that could use it for either brainstorming or creating finished products, the spokesperson said.
Below are some images created by DALL-E 2 matched with the text used to create the wacky images.
Open AI can also add edits to existing images, like taking a picture of a monkey and making it do taxes or adding a corgi into a photo of a museum gallery.
But with such great power comes great responsibility: The system has a number of guardrails to prevent abuse or the spread of disinformation. It bars offensive imagery from being uploaded into the system's machine learning program and also prevents images of real people from being manipulated.
"We share people’s concerns about misuse and it is something we take very seriously," the spokesperson said. "This is one of the reasons why we are previewing this research to a limited number of trusted users who are helping us learn about DALL·E 2’s capabilities and limitations, and why we currently do not make it available in our API. And we always conduct a safety analysis of our technology prior to release and our slow release approach allows us to respond quickly to unforeseeable risks while the stakes are relatively low."