Getty Images Sues Stability AI for Copyright Infringement
The Briefing by Weintraub Tobin
Do You Know if Using Copyright Protected Works to Train an AI Platform Is Fair Use?
The Fair Use Doctrine permits the use of copyright protected work in certain contexts to promote freedom of expression and for other specific purposes. It may be the case that using copyright protected works to train a platform would constitute fair use, but using it to generate new content may not be. For example, if millions of photos were provided to an image generating platform and it generated a novel image, it seems unlikely that would constitute copyright infringement. On the other hand, if the platform had only a small sample of photos to work from and if perhaps those photos were from the same creator, the resulting product could constitute copyright infringment.
Getty Images filed a lawsuit against startup tech company Stability AI for allegedly scraping more than 12 million photographs from Getty Images’ portfolio without consent. Scott Hervey and Josh Escovedo discuss this case on this episode of The Briefing by the IP Law Blog. Watch this episode on the Weintraub YouTube channel here.
Cases discussed:
- Getty Images (US), Inc. v. Stability AI, Inc.
- Visual Arts v. Goldsmith
Show Notes: Josh: Getty Images, a preeminent global visual content creator and leading source for visual content, has filed suit against startup technology company Stability AI for allegedly scraping more than 12 million photographs from Getty Images’ portfolio without consent or compensation. Getty Images also alleges that Stability removed or altered Getty Images’ copyright management information, such as watermarks, and provided false copyright management information. Accordingly, Getty Images’ complaint includes claims for copyright infringement, providing false copyright management information in violation of 17 U.S.C. section 1202(a), removal or alteration of Copyright Management Information in violation of Section 1202(b), trademark infringement, unfair competition, trademark dilution, and deceptive trade practices under Delaware law. In order to understand the nature of the claim a bit better, it’s important to understand what these parties do. Scott, do you mind providing a brief overview of what the parties do for our listeners? Scott: Sure. Getty Images is a visual media company and a supplier of stock images, editorial photographs, video, and music for businesses and consumers. It has over 477 million assets, and it generates revenue by licensing the right to use these assets to creatives, the media, corporate entities, and general consumers. Many of you have probably seen their images online, complete with the Getty Images watermarks. They’re all over the place. On the other side of the case, we have Stability AI. Stability AI is a startup technology company that has created an image-generating platform called Stable Diffusion that uses AI to generate computer-synthesized images in response to text prompts. There is an open-source version of Stable Diffusion, as well as a revenue-generating version known as DreamStudio. To simplify the description of what Stability AI does, we can refer to the example used by Getty Images in its Complaint. If a user wanted to generate a photo of a cat wearing a scarf, the user would type “cat wearing a scarf” into Stability AI’s platform, and the platform would generate an image of a cat wearing a scarf. Of course, the returned image would not be a real photo of a cat wearing a scarf, it would be a computer-generated version of a cat wearing a scarf. And it is the process of how the AI platform generated that image that is at the heart of this Complaint. Josh: That’s exactly right. In order to generate images in response to prompts submitted to Stability’s platform, the platform relies on various images that Stability utilized to teach its platform what a cat is and what a scarf is. The platform then utilizes those images to generate a synthesized version of what the user has requested, so in the example above, a cat in a scarf. With that understanding, we can discuss the specifics of Getty Images claim. According to Getty Images, Stability copied over 12 million photographs from its portfolio and used those images to train its Stable Diffusion model. Getty also claims that Stability used associated text and metadata, which it contends are also protected as copyrighted expressions to train its model. This, according to Getty Images, constitutes copyright infringement. Scott: Getty explains in its complaint that its assets are highly desirable for use in connection with AI and machine learning because of its high quality and because the assets are accompanied by content-specific, detail captions, and rich metadata. Apparently, Getty Images has licensed millions of digital assets to technology innovators for various purposes related to AI and machine learning. Of course, here, Getty Images alleges that Stability didn’t even attempt to negotiate a license with Getty Images for the content, but instead copied the images without Getty’s consent, in violation of the Getty Images terms of use and allegedly United States Copyright Law. Josh: But according to a variety of AI technology companies, the practice of using copyrighted materials for the purpose of training an artificial intelligence platform constitutes fair use under United States Copyright Law. Whether that’s true or not remains undecided. The rise of technology has created this issue of first impression, so it remains to be seen how the Courts will apply the fair use doctrine in this context, which, as many of you know, permits the use of copyright-protected work, in certain contexts, to promote freedom of expression and for other purposes. In my opinion, this isn’t an issue that will be decided through just one case. Fair use is a multi-factor analysis that takes a number of items into consideration when determining whether a use was fair under the doctrine. Two of the most significant factors are the purpose or nature of the use and whether the allegedly fair use has an effect on the market for the protected work. With that said, it may be the case that using copyright protected works to train a platform would constitute fair use, but using it to generate new content may not be. In other words, it may be fair to use someone else’s IP to train your platform, but if you use it to generate a new product that may not be fair. Scott: And its even a bit more nuanced than that. For example, if millions of photos were provided to an image generating platform and it generated a novel image, it seems unlikely that that would constitute copyright infringement. On the other hand, if the platform only had a small sample of photos to work with, and if perhaps those photos were from the same creator, the resultant product may constitute infringement. Josh: Exactly. And as to the latter, exactly how different the resultant product would need to be to be considered transformative in nature and therefore protected by fair use is a question that is currently before the Supreme Court in the matter of Andy Warhol Foundation for the Visual Arts v. Goldsmith, which has already been argued and should be decided this term. As many of you know, that case deals with how different a work must be from the original protected work to be considered transformative and not infringing. That decision will have consequences in the AI-IP disputes as well. Scott: There are so many interesting IP issues involved in this dispute, and it’s just getting started. I do not foresee this case resolving right away, so we’ll be watching very closely. Thanks for sharing, Josh.
The AI-powered Podcast Player
