Getty Images has banned people from uploading AI-generated pictures to its massive stock image collection, citing concerns over copyright.
Text-to-image tools, such as DALL-E, Midjourney, Craiyon, and Stable Diffusion, have opened the floodgates for machine-made artwork. Anyone can either pay a small fee or use a free model to create images from text descriptions.
All you have to do is tell, in writing, the AI system what kind of scene you want it to make, and the software will generate it for you. The quality of these images has got so good they are now being used by professionals to make magazine front covers, adverts, win art competitions, and so on.
You can see them as interesting tools to generate pictures, or as the end of art as we know it.
The copyright on these machine-made pictures remains unclear. The neural networks trained to generate images are trained on photos and art scraped online from sites such as Pinterest or Artstation. Netizens can easily create digital art in the style of any living or dead artists included in the training dataset in just a few seconds.
That puts a question in some people's minds: if an AI closely apes - or rips off - an artist, how legally safe is that? If a computer is trained from other people's pictures using someone else's software and that output is then sold by another party, how does that affect ownership, rights, and liability?
Getty has, amid this uncertainty, updated its policy to now ban submissions created by AI software to its stock libraries; it will not host and sell these types of images anymore. If there's one thing stock libraries like, it's well-defined ownership and copyright of material in its libraries - without these, it's not prepared to license work to other to use. It's too much of a legal mess.
"There are real concerns with respect to the copyright of outputs from these models and unaddressed rights issues with respect to the imagery, the image metadata and those individuals contained within the imagery," CEO Craig Peters told The Verge.
"We are being proactive to the benefit of our customers," he added.
Peters declined to answer questions on whether Getty Images had been threatened with legal issues from people challenging AI-generated content.
He said the changes were made to "avoid risk to [customers'] reputation, brand and bottom line." A quick search on the company's iStock site for keywords such as "AI-generated" or "Midjourney" shows that thousands of pictures have been removed. There are still many lurking on the platform that are less obviously produced from a computer's imagination.
Peters said Getty Images is going to rely on users to identify and report AI-generated images and the company is currently working with the the Coalition for Content Provenance and Authenticity) to create filters that may be able to automatically flag problematic content.
Other stock image giants, such as Shutterstock, also appear to be curtailing AI-made artwork. Motherboard noticed Shutterstock had quietly been removing images that are described as "AI-generated" or directly associated with tools like Midjourney.
The Register has asked Shutterstock for comment. ®
They can't be any worse than some human developers
Meanwhile in China, Alibaba runs 500 delivery-bots and they've delivered 10 million items to Easy Street
Kernel live-patching and a full decade of software updates
How meta, or should that be, Meta?
Google-owned company claims another scalp with applications in computer science
And it wouldn't be a Redmond OS update without printing issues
Plus: It wouldn't be a Windows update without printing issues