The DALL-E 2 AI text-to-image generator is now available to the general public, parent company OpenAI announced Wednesday.
While DALL-E 1 launched in January 2021, DALL-E 2 launched over a year later, initially to only 200 beta testers, which included artists and researchers hand-selected by OpenAI. By May, OpenAI was introducing 1,000 new users a week from a waitlist of users who added themselves, hoping to be approved by the company. 1.5 million people have so far been granted access to DALL-E and were creating, OpenAI claims, 2 million images a day.
OpenAI restricted access during the beta testing phase to receive feedback, the company said, and, crucially, to find out whether their safety systems were effective at blocking violent, sexual, or misleading images that could be used to spread false information. Users could be banned for using certain triggering words, like riot, warrior, battle, fighting, and the names of a real person, one user found.
According to OpenAI’s announcement, it has made their safety systems more robust and, thanks to feedback, it has added new features like Outpainting, which allows users to create on larger canvases in a wider variety of aspect ratios as opposed to the default square.
That DALL-E was restricted may come as a surprise given how many people seemed to be posting DALL-E type creations, but copy-cat applications like Craiyon, formerly known as DALL-E Mini, provided the public with a generator similar to the DALL-E system.