AI Art Generator Raises Questions About Digital Copyrights

Type in a prompt like "email," and several entirely original stock images of computers and messaging will be created instantaneously. A few more clicks and inputs will generate a host of text snippets suggested for ad copy. That's how a new tool from a startup called Omneky lets brands—usually growth-hungry apps and ecommerce sites—pair artificial intelligence-generated photorealistic images with passable machine-made copy and post the results directly to social and video platforms. The platform, which taps a popular open-source AI art generator called Stable Diffusion, is one of the first to tailor this cutting-edge technology specifically to brands in this way, but it likely won’t be the last. As AI-powered image generation technology has advanced by leaps and bounds in the past year, its output has transformed from garbled abstractions to sometimes-realistic renderings. In turn, a host of new creative software tools have put the power to instantaneously create lifelike images through a simple text directive in the hands of virtually anyone. AI creations are winning local fine art contests, flooding major stock image banks and providing imagery for email newsletters. The sudden ubiquity of this kind of technology is already raising new questions about digital ownership and what constitutes an original work of art. For instance, do human artists have a case against a machine that copies their style of work because it happens to be trained on a massive data set that might have included the artist’s own images? And can feeding prompts to an algorithm and building on the machine's work with human touches ever qualify as an art form in itself? As these tools inevitably begin to present challenges for copyright law, experts told Adweek it’s only a matter of time before those questions are litigated in court, which could have wider implications for how copyright works on the internet. The legal liabilities around AI-generated art are likely to become even more pressing as platforms like Microsoft-backed research group OpenAI attempt to sell developers and brands on these technologies through new commercial products. "And so if we're going to talk about this in any way, from a copyright perspective, it really should be considered a work of joint authorship, where me as the human has had some input into what the AI is actually going to create at the end of the day," said Suzanne Hengl, partner at Baker Botts, which focuses on branding and copyright law. Copyright requires human authorship As it stands now, works created solely by artificial intelligence are not applicable for protection under U.S. copyright law. Legal authorities have repeatedly ruled there must be some proof of human authorship, which could present major hurdles for brands looking to protect campaigns created with AI. “It's a pretty dicey proposition for brand owners and for ad agencies to rely solely on AI content for their campaigns because there's really very little that they can do to protect those campaigns from getting out into the public domain and allowing folks to use them at will,” Hengl said. Getty Images has banned AI-generated images for legal liability reasons. While Shutterstock has removed some of the AI-generated art posted to its platform, the stock image bank posted a statement from CEO Paul Hennessy on Sept. 22 that said the company plans to chart a path for allowing such content once it grapples with "open questions on the copyright, licensing, rights and ownership of synthetic content. "We need to do all that we can to not only protect the intellectual property rights of our contributors alongside the advent of this technology, but also ensure that they’re empowered to take advantage of this new creative medium," he added. A Getty Images spokesperson echoed concerns about copyrights around AI generation. "There are open questions with respect to the copyright of outputs from these models, and there are unaddressed rights issues with respect to the underlying imagery and metadata used to train these models," the spokesperson said in a statement. Heinz has experimented with research group OpenAI’s DAll-E 2 image generator. Major brands and ad agencies have been careful about treading into these technologies in part for this reason. The few projects advertisers and media companies have undertaken with this technology have been largely experimental, such as a campaign where Heinz used research group OpenAI’s Dall-E 2 program to generate images of ketchup bottles. Because that campaign tasked the AI with mimicking Heinz's existing trademark, the brand was able to circumvent any legal issues around the project. “There are ongoing cultural conversations—and maybe future legal questions—around who has true ownership of the output of certain prompts,” said Kai Tier, vp of technology at R/GA. “If I am creating something with a prompt that is in the style of an existing artist, and the output is based on their previous work, is the work I created all mine, or is it really that of the artist who created the original references?” What constitutes a novel creation? Whether or not brand output is applicable for protection is less important for the kinds of companies Omneky’s AI platform caters to, which are mostly looking for stock art to display in performance marketing-style digital ads. Hikari Senju, founder and CEO of Omneky, said while the output images of the tool might not be up for copyright protection, they are different enough from the training data to constitute a unique creation. “My dad is actually an artist … and the thing with art is that artists copy each other all the time,” Senju said. “Even before generative art, there was this question of what constitutes original content; what is sufficiently original versus what is copyrightable or copyright defensible. Generally, as long as you can prove that the concept or idea is sufficiently novel or original, even if you're using other people's assets, that is a piece of artwork.” It's (almost) alive! A variety of tools have popped up that now offer just about anyone the ability to create lifelike AI images from simple text prompts. Dall-E 2: Released in January, Microsoft-backed research group OpenAI’s Dall-E 2 is the program that almost single-handedly advanced AI art to its current state. The software is trained on hundreds of millions of images pulled from across the web. Midjourney: This tool is the product of a small, independent research lab that operates through a bot in a Discord group chat that anyone can join. Stable Diffusion: This open-source software was created by a startup called Stability AI and is accessible to anyone through an API program. Forrester analyst Will McKeon-White said brands are also likely wary of liability issues if an AI output does happen to hew too closely to an image already under copyright, such as an existing work of art or another brand asset. But some of the legal questions could come down to whether or not a brand is intentionally trying to replicate a copyrighted image, as evidenced by the text prompt fed to the machine, he said. For instance, if the prompt used to generate an image was "McDonald's Happy Meal," the resulting copyright breach might appear much more intentional than if the input was a more generic "fast-food combo." “It's really up to the humans providing the training data to think of these problems beforehand, or as people using them to then leverage them after the fact,” McKeon-White said. “If you're effectively trying to use one of these things as a mechanism to intentionally create the image of a specific copyright in a specific scenario, that will likely lead to some legal ramifications, but unintentionally … it’s likely a little bit more of a gray area.” Potential legal challenges Stephen Thaler, president and CEO of AI creativity startup Imagination Engines, has been on the front lines of the push for an expansion of copyright beyond human authorship. The research scientist has been attempting to secure a copyright for his AI-created art since 2018. But each time, the U.S. Copyright review board has shut down his applications and appeals based on its established precedent around human involvement in the creative process, most recently in a decision last February. But Hengl thinks Thaler's efforts could soon push the issue of AI copyrights to a higher court or even Congress if he decides to invoke a law called the Administrative Procedures Act, which would force a judicial review of the U.S. Copyright Office's decisions. Based on how Thaler has pursued a previous patent case, she believes he will likely do so in the next few months. Agency 10 Days used the Midjourney AI tool to create new logos for iconic brands, including these surreal images for Gucci. “We're going to see some pushback on this idea of you have to have human authorship in order to have something that's protectable under copyright,” Hengl added. In any case, Hengl thinks the advent of AI art will also lead to an expansion of the terms of fair use, the legal exemption made for use of a copyrighted work without permission under certain conditions, in much the same way the doctrine was widened with the dawn of the internet age. “I do think you're gonna see the fair use doctrine expand,” Hengl said. “We're going to be taking principles that have been laid out for a very long time and apply them to a new medium, which is always what the law does. It's always about 10 steps behind where technology is.”

AI News

AI Artwork's Growing Spread Raises New Questions About Digital Copyrights

Emerging photorealistic image generators could have major implications for creative ownership

Patrick Kulp

About

Subscriptions

Events

Publications

WORK SMARTER - LEARN, GROW AND BE INSPIRED.

Subscribe today!

Patrick Kulp