Inside Getty Images’ GenAI Transformation Strategy

It’s rare for a company to get confronted by a new technology that is so powerful it poses an existential threat to its business.

That’s arguably what happened to Getty Images when text-to-image generators like Stable Diffusion first captured the public’s attention a few years ago.

Three years into the mass adoption of artificial intelligence (AI) image generation, Getty, which is known for its high-quality photography, illustrations and stock images, has emerged stronger.

It has embraced generative AI; it’s in the midst of acquiring its largest competitor, Shutterstock, in a $3.7 billion deal and in the process is carving out a strategy to position it well for a generative AI future.

In an exclusive interview with PYMNTS, Grant Farhall, chief product officer at Getty Images, explained that the company’s competitive strength lies in its library of licensed images. This comes at a time when companies are shying away from using content that infringes on a creator’s copyright.

Getty believes that trust — backed by decades of content curation — sets it apart, since customers care less about flashy AI tools and more about what will enable them to create faster and smarter content — without getting into trouble.

“Customers are using our tool largely because of the fact that it’s commercially safe,” Farhall said. “That’s one of the main reasons why customers have been hesitant to use these [image generation] tools. … The number one hesitation is around the legal risks and the brand risks.”

Several artists have sued the makers of AI image generators like Stable Diffusion since 2023, alleging that these AI models were unlawfully trained on their copyrighted works.

Last August, in a landmark case, artists got a win when a Northern California federal judge ruled that their lawsuit against Stability AI (co-maker of Stable Diffusion), Midjourney, DeviantArt and Runway AI, could proceed.

Getty itself sued Stability AI in 2023, alleging that it infringed on the intellectual property rights on millions of photos and their metadata in building its image generator. The case is ongoing.

Earlier this month, Getty urged the Trump administration — which is drafting the U.S. AI Action Plan — to require that AI model developers fully disclose what data they use for training the models. Getty also said the government should not deem as “fair use” under U.S. copyright law if the outputs of AI models compete with the owners of the content, among other stipulations.

Read more: U.S. Appeals Court Upholds Ruling Denying Copyright for AI-Generated Art

Farhall said users of Getty’s content library are protected in three ways. First, the assets are licensed content. Second, Getty will indemnify users against lawsuits in the case of alleged violations. Third, users of Getty’s AI image generator can rest assured because the model was trained only on its legally sanctioned content library.

“Our [AI] model is only trained off our creative library,” Farhall said. “There’s nothing in there that is scraped. There is nothing … from public domain sets. There is nothing in there that is synthetic, because we don’t take any AI content into that creative library. So it’s a clean dataset.”

Bill Bon, senior director of creative operations at iStock, a Getty business unit, said years of work went into also ensuring their dataset reflects authentic, inclusive and global imagery.

“We really do direct and guide the artists that we work with on a global scale to make sure that we’re representing not only authentic-looking imagery but also representing a range of regions across the world,” Bon said.

One of the most popular use cases is the ability to edit Getty’s stock images to fit a particular brand’s voice. “That’s where we see a lot of our customers using the tool and licensing content because they can get to something that is more exactly what they’re looking for,” Farhall said.

As for who owns the AI-generated image made by a customer, Farhall said Getty grants customers full rights “to the extent the law recognizes it.”

Beyond legal and ethical considerations, Getty is pushing hard on user experience and customization.

One area of growing interest is personalized brand content. Through custom fine-tuning, Getty helps companies build brand-specific models trained on proprietary data and only accessible to them.

A fun example Farhall mentioned was a “Venomize My Pet” campaign, in which a foundational model was fine-tuned with proprietary Marvel character data to generate pet portraits in Venom’s likeness.

However, there is one type of content that is off-limits for AI: Editorial content like those related to news, sports and celebrities.

Getty does allow some use of AI for retouching editorial content, since photographers have been retouching photos using other tools for decades, but restricts changes to less than 10% of the image.

“When you get more than 10%, you’re really starting to change the very essence of the image,” Farhall said.

Getty also does not watermark AI-generated images, but Farhall said their metadata tags do mark them as such. As for Getty’s customers, it’s up to them to decide if they want to reveal an image is AI-generated, he said.

Looking ahead, Farhall said the company will continue to find ways to pair AI with Getty’s extensive creative library — legally.

“Everything we do will still be commercially safe and respect the rights of creators and IP holders,” Farhall said. “Fortunately, that’s something that makes us unique.”

Credit: AI-generated image from Getty Images

The post Inside Getty Images’ GenAI Transformation Strategy appeared first on PYMNTS.com.