By Katie Paul
NEW YORK (Reuters) – A “responsible” AI image generation startup backed by stock photo provider Getty Images has raised $24 million in a Series A funding round, including from top ad agency Publicis Groupe, it said on Wednesday.
Israel-based BRIA, which licenses more than a billion images from stock providers for its system, said in a press release that it would use the cash infusion to expand globally and build out text-to-video generation capabilities.
The investment, while small relative to others in the white hot AI space, makes BRIA an early test case in how licensing deals can provide a path forward in disputes raging over the use of copyrighted content to train generative AI models.
Music labels, artists and other content owners have pointed to licensing as a more palatable approach to generative AI, after big AI companies built their systems using content scraped off the open internet for free.
BRIA said its license-only approach heads off the risk of trademarks and other fraught elements appearing in its generated images, attracting customers in the advertising, marketing and media world including Publicis.
“There is a way to do this that is both responsible and commercial,” CEO Yair Adato told Reuters in an interview.
Getty, which has backed BRIA since 2022, has figured prominently in the disputes around training data.
Last year, Getty sued leading AI image generation company Stability AI for using its photos without a license, which it alleged was “brazen” infringement of its intellectual property “on a staggering scale.”
It also teamed up with chip maker Nvidia last year to build an image generation service trained only on the images in its own library.
Rivals Adobe and Shutterstock likewise offer AI image generators, while all three photo companies license their visuals to Nvidia for its own service.
BRIA fed its model visuals from 18 stock providers, including Getty, Alamy, SuperStock and Envato. It charges customers for access to the platform and shares revenue with the photo companies.
Each time BRIA generates an image, it displays the original licensed images that contributed data toward that output and accounts for how much came from each.
Adato likens the approach to a Spotify model for the AI era, which compensates content owners for their role in enabling the technology.
Controlled training also guards against the technology being used to create misleading visuals that could tarnish brands or influence elections, he said.
BRIA’s model, for instance, is not capable of generating an image of U.S. President Joe Biden because it has never been taught what he looks like, said Adato.
(Reporting by Katie Paul; Editing by Christopher Cushing)