Caleb Ward, a media industry professional, gained visibility in Spring 2023 for his production of two mock trailers — one of Star Wars, one of Lord of the Rings — which mimicked the recognizable style of Wes Anderson (“The Grand Budapest Hotel”, “Asteroid City”) using the generative visual AI program, Midjourney.
As a Polygram critic noted of Anderson, “Most cinephiles immediately recognize his distinctive look….He has plenty of familiar touchstones for humorists who are trying to mimic him: The use of chapter titles and other on-screen text, a love of elaborate fixed tableaus, characters with minimal emotional affect and a clipped, precise way of speaking.” Moreover, Anderson has a recurring stock company from project to project, so these AI parodies can cast, for example, Bill Murray as Gandolf in “Lord of the Rings” and Obi-Wan Kenobi in “Star Wars”. And what could be more fun that imaging Christopher Walken as Gollum.
Ward’s trailers were widely circulated as the most considered and most polished of the Wes Anderson parodies, which became the focus of critical anxiety. The same Polygon critic argued:
“No artbot is going to actually replace Wes Anderson: his work comes from a distinctive voice and artistic mindset. Individual still images aren’t going to replace entire movies, and Anderson’s films are much more than just the visual imagery….but even so, it’s easy to look at the images … and see how readily AI art generators can devalue an individual artist’s style and voice….All the signature stylings Anderson has been refining for more than 25 years can be reduced to a single repetitive joke, to the point where his own actual movie stills may not stand out much in the mix.”
Midjourney art, these critics argued, was “soulless” because it was produced by machines; these artists, with no mastery over the medium, might replace and devalue “accomplished” producers.
Ward said he wanted to test the capacities of these new tools, acquire new skills, and demonstrate his professional capabilities. For many fan artists, AI’s ability to mimic the style of particular artists, to effectively construct images using their favorite performers, and to apply them to beloved fan objects has enormous appeal. Such images display fan literacy — the capacity to identify codes from popular culture and apply them to new contexts. So we might compare the ways “Star Wars” might have been visualized by German expressionist filmmaker Fritz Lange or Japanese samurai film director Akira Kurosawa, or what a Chinese version of “Game of Thrones” might have looked like with Michelle Yeoh as Cersi. Such fan art depends on the machine’s recognition and application of each artist’s techniques to different contexts. So, we come to see the fascist undercurrents of “Star Wars” in the Lang interpretation or can retrace what George Lucas borrowed from Kurosawa in the first place.
As a long-time student of grassroots creativity, I see fandom as a group of early adopters and lead users of generative AI. The logic of fandom has historically involved appropriating and remixing resources drawn from mass media for other critical purposes. But fans have also been important early advocates for ethical practices surrounding machine learning. Some are seeing Midjourney as a tool which democratizes the capacity for visualization, allowing many who have not received art school training to express themselves in visual terms for the first time, while others see the repetition of Midjourney’s recycled images as a threat to individuality of expression.
Like other artists, many fan artists feel threatened by the ways that the skills they have acquired through hard work and practice may be less valuable in a world where others can use AI to replicate many similar effects. They argue for important distinctions between human and “machine” art, suggesting that the AI can only replicate but not really transform the works it scrapes. That argument, though, ignores the active role humans play in designing prompts and in curating and refining those images. The effective use of Midjourney requires the development of skills and knowledge just as any other artform does. Not all artists can make the sometimes balky software achieve what they envision.
Some of the fascination right now has to do with the comical and sometimes infuriating ways the AI fails to understand people’s prompts, such as when the term “Silence of the Lambs” produces images of actual lambs. These breakdowns in communication can also provide insights into how AI understands the human world. For example, at one point my son was unable to make Midjourney produce a picture of a female doctor and a male nurse no matter how clearly he explained what he wanted. This blind spot occurred because the AI looked at a vast numbers of images online and observed that typically men wore the white coats and women wore scrubs. Asked to depict Los Angeles without cars, Midjourney spat out images of Los Angeles without humans. Already a recent update has made it possible for users to highlight the section of an image that is wrong and explain how it needs to change, making as many small modifications as they want until they express their vision accurately. But there is something valuable, if sometimes painful, to learn from the computer's blunt honesty about its initial impressions.
Midjourney also depends on a community of digital artists sharing work online with one another, providing feedback on how to produce more effective prompts and how to exploit the full capacities of the program. It is no accident that the researchers chose to embed Midjourney in Discord, already a social network of gamers and fans. This platform ensures that the overwhelming majority of art is created in public forums where novices can observe precisely what experienced users are doing to achieve the best results.
Fan artists have also expressed concern about how artworks — including fan works — are scraped from the web as a basis for machine learning without authorization, ecognition or payment. They fear that work produced as part of the fan gift economy may well be used for profit by major corporations even if the fans themselves have chosen not to make money from their work. These fan artists often use an inflamed rhetoric of “stolen art.” We need to be clear on the parallels and differences between how machine learning takes “inspiration” from existing artworks and the ways any genre artist learns, by following — and deviating — from a formula acquired by studying pre-existing works.
Yet, guidelines need to be established about how these works get commercialized; how AI companies are marketing using images produced without permission that are the style of living artists; and whether there could be mechanisms — similar to the way radio stations pay recording artists — that might provide compensation for the appropriation and remix of artist works. Such ethical standards for generative AI would provide a greater balance between artist rights and fair use. Corporations are not going to walk away from the profits to be made from AI, so the question is whether fans walk away, leaving these developments entirely in the hands of commercial interests... or whether they collectively fight for a say in how AI might be used for social benefit.
As we work through the status of artists in a world being reshaped by AI, this fan community represents an important site of debate, as well as a set of stakeholders that may be too easily ignored amidst the struggles between tech and entertainment companies.
Henry Jenkins is the provost professor of communication, journalism, cinematic arts and education, and primary investigator, Civic Paths Research Group. Jenkins is the author or editor of 20 books on various aspects of media and popular culture. He writes extensively about cinema, television, comics, computer games, online communities, popular theater, and other forms of popular media, primarily in the American context.