Getty Images is so confident its recent generative AI model is freed from copyrighted content that it’ll cover any potential intellectual-property disputes for its customers. 

The generative AI system, announced today, was built by Nvidia and is trained solely on images in Getty’s image library. It doesn’t include logos or images which have been scraped off the web without consent. 

“Fundamentally, it’s trained; it’s clean. It’s viable for businesses to make use of. We’ll stand behind that claim,” says Craig Peters, the CEO of Getty Images. Peters says corporations that wish to use generative AI want total legal certainty they won’t face expensive copyright lawsuits. 

The past yr has seen a boom in generative AI systems that produce images and text. But AI corporations are embroiled in quite a few legal battles over copyrighted content. Outstanding artists and authors—most recently John Grisham, Jodi Picoult, and George R.R. Martin—have sued AI corporations similar to OpenAI and Stability AI for copyright infringement. Earlier this yr, Getty Images announced it was suing Stability AI for using thousands and thousands of its images, without permission, to coach its open-source image-generation AI Stable Diffusion.

The legal challenges have sparked many attempts by others to learn from generative AI while also protecting mental property. Adobe recently launched Firefly, which it claims is similarly trained on copyright-free content. Shutterstock has said it’s planning on reimbursing artists whose works have been sold to AI corporations to coach models. Microsoft recently announced it’ll also foot any copyright legal bills for clients using its text-based generative models. 

Peters says that  the creators of the photographs—and any those who appear in them—have consented to having their art utilized in the AI model. Getty can also be offering a Spotify-style compensation model to creatives for using their work. 

The incontrovertible fact that creatives might be compensated in this fashion is nice news, says Jia Wang, an assistant professor at Durham University within the UK, who focuses on AI and intellectual-property law. However it is likely to be tricky to find out which images have been utilized in generated AI images as a way to determine who needs to be compensated for what, she adds. 

Getty’s model is barely trained on the firm’s creative content, so it doesn’t include imagery of real people or places that could possibly be manipulated into deepfake imagery. 

“The service doesn’t know who the pope is and it doesn’t know what Balenciaga is, and so they can’t mix the 2. It doesn’t know what the Pentagon is, and [that] you’re not gonna have the ability to blow it up,” says Peters, referring to recent viral images created by generative AI models. 

For example, Peters types in a prompt for the president of the USA, and the AI model generates images of men and girls of various ethnicities in suits and in front of the American flag. 

Tech corporations claim that AI models are complex and might’t be built without copyrighted content and indicate that artists can opt out of AI models, but Peters calls those arguments “bullshit.” 

“I believe there are some really sincere those who are literally being thoughtful about this,” he says. “But I also think there’s some hooligans that just wish to go for that gold rush.”