OpenAI, the company behind DALL-E 2, did away this week with the invitation-only barrier to its artificial intelligence image generation tool, opening the growing platform to anyone interested in using it.
So far, roughly 1.5 million users have used DALL-E to generate a wide array of imagery, from photorealistic faces and landscapes to painterly renderings of characters and scenes that didn’t exist until a user typed in a string of descriptive prompts to guide the AI’s output. Each AI image generator starts with free access, with pricing set according to usage.
In short order, a number of users have taken to using DALL-E, and tools like it, including Midjourney and Stable Diffusion, to flood stock image sites with new content in hopes of profiting from the fees paid to creators selling images and photos.
But if the AI software user didn’t draw, paint, or photograph an image, do they own it? Or does the company behind the software own it?
Not even the AI generator sites agree on this. Midjourney’s terms of service states that “you own all assets you create with [Midjourney],” with the caveat that the user grants the company a broad license to use not only the AI-generated images the user produces, but the user’s text prompts used to generate the images.
DALL-E’s terms, on the other hand, asserts that “OpenAI owns all generations” created on the platform, and only grants users the right to use and claim copyright on their AI-generated imagery.
Midjourney’s user-friendly terms are helpful to AI image producers like Kristina Kashtanova, who set a precedent for other AI users by successfully copyrighting an AI-generated graphic novel called Zarya of the Dawn on Sept. 15.
Under Midjourney’s terms, Kashtanova owns the work outright.
But if users of DALL-E violate OpenAI’s terms or content policy, the company says “you will lose rights to use [the AI-generated images].”
That seems straightforward enough, but it’s an exception to the way other popular digital art tools are managed. Serious artists would likely avoid using Photoshop, for example, if it meant giving Adobe the right to own or revoke the copyright on the work they produce in the app.
The rules on different stock image sites are no less variable. Getty Images has banned AI-generated images from its service. Shutterstock had a similar ban but recently reversed course on it, now allowing anyone to upload and profit from their AI-generated imagery.
Kashtanova tells Quartz that images she owns were initially deemed problematic by Shutterstock, but then suddenly allowed back on the platform.
“My [Shutterstock] images were 100% AI-generated using Midjourney. They were deleted, and then restored five hours later,” she says.
The first message Shutterstock sent to Kashtanova informed her that Shutterstock “does not accept machine-generated content,” warning her that “[w]hile the AI-generated content space continues to develop at a rapid pace, there are currently multiple copyright-related implications to machine-generated content that do not meet our Contributor Terms of Service.”
Hours later, Shutterstock told Kashtanova that “the AI-generated content that was removed from your portfolio has been restored.”
We asked Shutterstock to clarify its policy.“Our current stance is that we are accepting AI-generated content onto the platform for commercial and editorial use,” a company spokesperson tells us.
“At the moment, AI-generated content will be reviewed no differently than any other type of digital illustration submission… This may change at a moment’s notice as we learn more about synthetic imagery.”