From Search Engine Journal:
In a lawsuit filed on February 3 and made public on Monday, Getty Images alleged that artificial intelligence company Stability AI, Inc., infringed on the visual media company’s intellectual property.
The suit, filed in a Delaware US District Court following a separate Getty lawsuit against Stability in the UK, accused the London-based generative AI company of copying more than 12 million photographs and associated captions and metadata without permission.
According to the filed document, Stability used Getty’s IP in violation of trademark law to train its Stable Diffusion model.
A latent text-to-image model used to generate photorealistic images based on text input, Stable Diffusion is viewed as a direct competitor for creative imagery by Getty.
Additionally, the suit alleges trademark-protected Getty IP was used as training data for the AI platform, and “its infringement of Getty Images’ content on a massive scale has been instrumental in its success to date.”
Getty has claimed that Stable Diffusion’s output often contained a modified version of a Getty Images watermark. Getty’s complaint states this usage, particularly in low-quality images, creates confusion about the image source and falsely implies an association.
. . . .
This is not the first AI-based copyright infringement that has been filed. Last month, a trio of San Francisco-based artists, Sarah Andersen, Kelly McKernan, and Karla Ortiz, filed a suit against Stability AI and DeviantArt, alleging their platforms have infringed on the rights of artists with images scraped from the web.
“This latest lawsuit continues the debate and argument that OpenAI does not have the right to use or learn from publicly posted content, at least not without some sort of credit and compensation,” said Brent Csutoras, digital marketing expert and co-founder and managing partner at Alpha Brand Media, Search Engine Journal’s parent corporation.
. . . .
“Artists have taken inspiration from others for as long as we can remember, but I think it is fair to say that there is a difference between what a human takes as inspiration in their own art, versus what an AI can do by taking inspiration from millions of publicly available [works of]art.”
Link to the rest at Search Engine Journal
As PG thinks he has mentioned previously, this type of copyright infringement suit was certain to be filed by one or more entities who have collected a huge number of images they did not create, but, presumably licensed from the actual creators who presumably had valid copyrights.
PG thinks the last excerpted quote from the OP demonstrates the problem. Why is an artist who takes inspiration from dozens of paintings to create her/his own works of art different from a machine which is programmed to take inspiration from millions of publicly available paintings?
Can a very long-lived human artist eventually commit copyright infringement by taking her/his inspiration from looking at too many copyright-protected, publicly available images?
Another quote in the OP caught PG’s attention, “OpenAI does not have the right to use or learn from publicly posted content.”
PG is not aware of any provision of US Copyright laws that prohibits humans from learning from copyright-protected content they examine. If this were the case, every library in the world and each of its patrons would be violating millions of copyrights.
If the computer is just “reading” the images to transform them into millions or billions lines of code, where’s the copying and where’s the violation of the rights of a holder of the copyright to those images?
As a side note, Stable Diffusion’s management and engineers were monstrously stupid not to have scrubbed the Getty watermarks from the images during their digitization processing. PG doesn’t think this effects the ultimate question regarding copyright, but it certainly gives Getty’s attorneys something to harp about to a suspicious technophobic judge and/or jury.
2 thoughts on “Getty Images Sues Stability AI For Copyright Infringement”
Well, Getty’s lawyers are actually exhibiting some intelligence here (after so many times in the past that they have not).
They do have the two elements that are absolutely necessary to a trademark infringement lawsuit: use of a registered mark, and in such a way that it can easily confuse the consumer.
What they did was acknowledge that those fuzzy, unreadable fake watermarks that sometimes appear in text-to-image-generated images in no way mean that the generated image belongs to Getty Images and by extension, not to the artist who created the image which was allegedly copryright-violated by the generator. Thanks for clarifying that point, Getty. It’s now safe to assume that the generated images are indeed unique and original.
Comments are closed.