Today, DALL-E 2, OpenAI’s AI system that can generate images with an application or edit and refine existing images, is increasingly available. The company announced in a blog post that it will speed up customer access to the waiting list with the goal of reaching approximately one million people in the coming weeks.
With this “beta” release, DALL-E 2, which had been free to use, will move to a credit-based fee structure. First-time users will get a finite amount of credits that can be used to generate or edit an image or create a variation of an image. (Generations return four images, while editions and variations return three.) Credits are reloaded each month up to 50 in the first month and 15 a month later, or users can purchase additional credits in $ 15 increments.
Here is a chart with the details:
Image credits: OpenAI
Artists in need of financial assistance will be able to apply for subsidized access, according to OpenAI.
DALL-E’s successor, DALL-E 2, was announced in April and was available to a select group of users earlier this year, and has recently surpassed the 100,000 user threshold. OpenAI says broader access was made possible by new approaches to mitigating bias and toxicity in generations of DALL-E 2, as well as developments in the images that govern the policies created by the system.
An example of the types of images that DALL-E can generate 2. Image credits: OpenAI
For example, OpenAI said this week it deployed a technique that encourages DALL-E 2 to generate images of people that “more accurately reflect the diversity of the world’s population” when a message describing a person with a race is given. or an unspecified genre. The company also said it now rejects image uploads that contain realistic faces and attempts to create the likeness of public figures, including prominent political figures and celebrities, while improving the accuracy of its content filters.
In general terms, OpenAI does not allow DALL-E 2 to be used to create images that do not have a “G rating” or that can “cause damage” (for example, self-harm images, hate symbols, or illegal activity). ). And previously it did not allow the use of images generated for commercial purposes. However, as of today, OpenAI grants users “full usage rights” to market the images they create with DALL-E 2, including the right to reprint, sell, and market, including the images they generated during preview. initial.
As demonstrated by DALL-E 2 derivatives such as Craiyon (formerly DALL-E mini) and the unfiltered DALL-E 2 itself, image-generating AI can very easily detect biases and toxicities embedded in millions of images. of the web used. to train them. Futurism was able to incite Craiyon to create images of crosses in flames and demonstrations of the Ku Klux Klan and found that the system made racist assumptions about identities based on “ethnic” names. OpenAI researchers noted in an academic paper that an open source implementation of DALL-E could be trained to make stereotyped partnerships such as generating images of white men in business suits for terms such as “CEO “.
Although the OpenAI-hosted version of DALL-E 2 trained in a filtered dataset to remove images that contained obvious violent, sexual, or hateful content, filtering has its limits. Google recently said it would not launch an AI generation model it developed, Imagen, because of the risks of misuse. Meanwhile, Meta has limited access to Make-A-Scene, its art-centric imaging system, for “prominent AI artists.”
OpenAI stresses that the hosted DALL-E 2 incorporates other safeguards, such as “automatic and human control systems” to prevent things like the model from memorizing faces that often appear on the Internet. Still, the company admits there is more work to be done.
“Expanding access is an important part of our deployment of AI systems responsibly because it allows us to learn more about real-world usage and continue to iterate about our security systems,” OpenAI wrote in a post to block. “We continue to investigate how AI systems, such as DALL-E, can reflect biases in their training data and different ways of approaching them.”