IP Frontiers: Artificial Intelligence’s Copyright Battle in Documentary Filmmaking

By: Arjay Parhar

Image generation is one of the most popular AI tools. By exponentially lowering the requisite skill, high-quality content can now be generated remarkably quickly. Last Week Tonight perfectly showcased the ease, power, and creativity of AI generated images by showcasing host John Oliver’s passionate yet mariticidal affair with a cabbage, all generated with simple user prompts. With these tools’ increasing pervasiveness, legal issues will inevitably impact several industries. This article will examine how AI generated images could disrupt at least one legal barrier that many documentary filmmakers must overcome to distribute their films.

Documentary filmmakers are often required to purchase insurance on their films to limit film festivals’ liability from a copyright infringement suit. Entering films in festivals is not only important for the distribution of a film, but it can also be a dream for many. Insurance carriers, or sometimes just the film festivals themselves, require a letter from an attorney explaining whether the film’s use of copyrighted materials is protected under fair use. A filmmaker fills out a timestamped sheet with the source and owner of the clip to facilitate the attorney’s recommendation. The attorney will then analyze the film and the copyrighted material to determine the fair use strength of each clip and counsels the client both on the efficacy of their fair use and recommends what steps can bring that film into greater fair use compliance.

Documentary filmmakers commonly rip images to comment on and weave a narrative. Legally, these filmmakers are allowed to use these ripped images if they make fair use of the images. Making fair use is ultimately a judicially determined doctrine determined by a four-factor test: (1) the purpose and character of the use, (2) the nature of the copyrighted work, (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole, and (4) the effect of the use upon the potential market for or value of the copyrighted work.

AI generated images disrupt this process between an attorney and a client because AI images conceal the clip source and thus the owner. AI tools do not disclose how its algorithm arrives at its final product, such as what images the algorithm is using, learning from, and/or combining. Without knowing these sources, an attorney cannot determine fair use factor 3, the amount and substantiality of the portion used in relation to the copyrighted work as a whole. Even a small portion of a borrowed clip could weigh against this factor if the borrowed portion goes to the heart of the original source’s purpose. Additionally, the now unknown image’s owner usually helps an attorney counsel which owners are particularly litigious and which owners are open to reasonable licensing fees. Without this knowledge, an attorney’s advice will be limited and could impede an attorney from making any recommendation.

Even if some attorneys, recognizing the risk of AI generated images, decide to greenlight AI images, insurance carriers may increase the costs of insurance for AI created images to balance out the uncertain risk for how courts will handle this new technology. This could create a chilling effect on independent and low-income filmmakers from producing content with AI images, leaving the AI film landscape to larger corporations who have a strong legal infrastructure to support their films.

However, fair use is a holistic analysis that allows for the strengthening of some factors over the weaknesses of others. The ease with which these tools can transform works provides a simple workaround to transform copyrighted images and clips into fair use. Although the use of the image in relation to the copyrighted work as a whole (fair use factor 3) may be obscured, filmmakers can focus on and strengthen transformative uses (fair use factor 1), i.e., uses that add something new with a further purpose or different character. If an attorney finds that an AI clip is not strong fair use, an attorney can counsel a client to further use that AI tool or other tools to add elements to illustrate their point. Take for example, a documentary filmmaker following the Sacramento Kings this year. Asking an AI image generator to show a sequence of images that illustrate De’Aaron Fox’s clutch shot making this year could produce a series of images that closely resemble the NBA’s copyrighted material. Since buzzer beaters and clutch shots would likely go to the heart of the NBA’s copyrighted material for that game, this would run against the substantiality of the portion used in relation to the copyrighted work as a whole. However, using AI tools to further transform the images such that the crowd is surrounded by a desert that transforms into an oasis upon the basketball slipping through the orange rim, illustrating that the King’s 16-year playoff drought has ended, would make a stronger case for fair use because the filmmaker would be ensuring that additive creative elements are illustrating the underdog growth narrative of the Kings. Recognizing the potential for copyright infringement, filmmakers should be able to make better fair use of their films by focusing on adding transformative elements.

Viewing the AI legal complications from the other side, this AI black box will also frustrate the owners of a copyright from exercising their rights. Artists are already complaining that AI image generators have been training their algorithms on their art without their permission. Early this year, three artists and Getty Images separately filed copyright lawsuits against Stable Diffusion, citing evidence that the algorithm could output images that closely resembled already existing images. A user’s belief that their AI generated image is uniquely generated combined with the inability to check reference images incentivizes a willful blindness to copyright infringement, endangering the original market. This disruption of the original market runs against fair use factor 4 and further cautions against simple uses of the algorithm. The separation between the algorithm’s decision-making and the user’s interface makes turning a blind eye to copyright infringement easier than ever.

How these algorithms arrive at their final products will likely be pivotal in the algorithms’ legality. Although the class action lawsuit filed by the three artists may oversimplify the AI algorithms as a “complex collage tool”, the truth may be closer to a drawing algorithm that learned to draw based on copyrighted images. If this latter representation is more accurate, the precedent for cases of copyright infringement on novel technologies that privately train an algorithm on massive amounts of copyrighted material have weighed in favor of the algorithm’s legality. For example, in Authors Guild, Inc. v. Google, Inc., 721 F.3d 132 (2d Cir. 2015), an appeals court found that Google was not infringing copyright when it scanned books in libraries to create a book search engine but did not publish the books. In A.V. v. iParadigms, L.L.C., (4th Cir. 2009), an appeals court rejected a copyright lawsuit against TurnItIn, where the lawsuit claimed that the company infringed students’ copyrights by internally archiving essays without permission. The court noted that since the students’ essays had not been published, the service was not a substitute for the essays. Both cases focus on the disruption of the original market and found that the cases did not run against fair use factor 4. Like in those cases where the courts found that leveraging large amounts of copyrighted material to train a program was not copyright infringement, perhaps the courts will similarly find for the legality of these AI image generators because the AI is not publicly publishing a “collage” of images, but rather privately using images to train a unique drawing algorithm.

In conclusion, while AI tools’ black box decision making can frustrate both artists attempting to make fair use and the owners of the copyright from accessing existing legal processes, transformative uses of these AI tools will bolster filmmakers’ ability to transform art into their own unique works. Hopefully the AI tools’ owners or others in the industry will recognize the legal difficulties the technologies bring and create other tools to bolster transparency of the algorithm and assist the rights of copyright holders.