Last year, Getty Images banned AI-generated content on its platform, citing potential legal concerns for customers.
Getty also sued Stability AI, accusing it of copying 12m+ of its photographs.
But now, Getty is getting into the game on its own terms, partnering with Nvidia on a text-to-image generator. Pricing will be based on prompt volume, separate from a standard subscription, per The Verge.
What’s different about Getty’s?
Getty’s editorial content includes photos of real people, places, and events; its creative section consists of stock imagery and illustrations. Getty’s model only trained on the latter, per MIT Technology Review.
That’s key because:
- The model won’t create bogus images of public figures or include brand logos — so, no pope in a Balenciaga jacket.
- Getty’s images are already licensed, and those who created them or appear in them have agreed to have their work train Getty’s AI.
- Thus, any content created with Getty’s model can be used commercially. Getty is also indemnifying its customers — meaning that, should legal challenges arise, Getty will be responsible.
Elsewhere…
… The legality surrounding AI-generated work remains in flux.
- A judge recently decided that work entirely generated by AI can’t be copyrighted.
- Authors including George R.R. Martin and Jodi Picoult have joined the Authors Guild in suing OpenAI, accusing its AI model of training on their novels.
- Three artists filed a class-action lawsuit against Midjourney, DeviantArt, and Stability AI, alleging their models illegally trained on their work.
Getty may be an image powerhouse with 477m+ assets, but being able to generate images for free threatens its subscription model. This move both protects Getty from getting left behind and its customers from all that legal murkiness.
Similarly, Photoshop maker Adobe’s new AI tool did not train on copyrighted materials either, and Adobe also promised indemnification for its customers.
