Generative artificial intelligence (AI) can create a wide range of output, including images, text and videos. This generative art is increasingly discussed, due to the recent popularity of tools such as DALL-E, Midjourney and Stable Diffusion for images, and OpenAI’s Chat GPT-3 for text. Generative AI makes it possible to generate new images in diverse styles. Since generative adversarial networks (GAN) were introduced in 2014, these tools have developed fast. In earlier blogs, I wrote about AI fan fiction (2017) and AI games and text adventures that use earlier versions of Chat GPT. This is by no means a new trend, but a broader audience now uses these tools. And it’s also making its impact in fandom.
An AI image can be a beautiful and interesting revision of what we know. In 2022, The AI Star Wars Project by Oren Sved gained critical attention. Generated in Midjourney, the images depict different scenes from the original Star Wars trilogy in a style similar to the Russian director Andrei Tarkovski. The AI Star Wars Project resembles Tarkovski’s classic Soviet science fiction art film Stalker (1979). The scenes feature austere landscapes, junkyard technology and Storm Trooper outfits that blend with Soviet militarism.
AI art can surprise us and be an interesting combination of styles. At its best, a human prompter gets the most out of the AI through different iterations and something unexpected is created. A related example of generative art is this AI trailer of The Galactic Menagerie, or Star Wars in the style by Wes Anderson. Complete with narrator, this nostalgic, colorful and quirky adaptation of Star Wars drew millions of viewers already. They both appreciate and fear what AI can do.
Generative Art as Transformative Art
This generative art is clearly a form of transformative art, since it’s derived from existing data, styles and artworks. Within fan and media studies, we usually say transformative art to refer to pastiche, homage and fan fiction, but I think the term also captures this new data-driven art very well. It is derivative on purpose as well, because of the prompts we use (“in the style of Escher, in the style of Van Gogh, in the style of Wes Anderson’). AI is remix, and purposefully created as such by human prompters. But it is not remix in the sense that it has clear traces of the original, which makes it difficult to say where we stand in terms of copyright.
As one of my favorite AI critics, Roland Meyer, rightly points out, not every style is that easily mimicked. AI for instance excels in the style and moods of Wes Anderson. His art direction is highly saturated, filtered and nostalgic. It’s a highly Instagramable style. No wonder that models trained in vast amounts of user generated data excel in this.
Generative art is about moods, styles and vibes more than it is about specific characters. The AI makes an estimation, based on the huge amounts of data that it has on “Baby Yoda” or “Darth Vader”. Sometimes it comes close, sometimes it completely misses the mark.
Text is also prone to this automation. In 2017, I was surprised by Botnik’s Harry Potter and the Portrait of What Seemed Like A Large Pile of Ash, which I wrote about in this entry. This is one of the first AI-driven fan works that really made an impact on fandom. It was a hit with many users, and many loved to create fan art based on the work. The AI converted Harry Potter into a surreal horror (“The sky outside was a great black ceiling, which was filled with blood”). By now, generative fan fiction is steadily uploaded on AO3 and other platforms. I have no doubt that’s only the tip of the ice berg of what’s generated. My guess is that a lot of users who first use Chat GPT-3 prompt it for fan fiction, because using known characters and settings is an easy and fun way to experiment with the tool.
It’s worthwhile to note that AI is also trained in specific genres of fan fiction. Wired wrote a very interesting article on how AI recognizes specific terms from the omegaverse. This brand of slash fanfiction originated in Supernatural fandom and assumes there is a hierarchy within sexual partners (e.g. alphas, omegas). “Because the Omegaverse has such specific terms and phrases associated with it, ones that are found within fan fiction and nowhere else, it’s an ideal way to test how generative AI systems are scraping the web. Determining what information has gone into a model like ChatGPT is almost impossible.”
The article was inspired by the post of a Reddit user, who prompted the GPT-3 tool Sudowrite for consistent omegaverse fiction. This data most likely comes from the CommonCrawl dataset, which relies on public internet data. Archive of Our Own, where many of this fan fiction is hosted, is public domain. They have been subject to scraping, which is how this data became part of these tools and platforms.
Archive of Our Own recently released a statement that they are taking measures against scraping and its results: “Once we became aware that data from AO3 was being included in the Common Crawl dataset — which is used to train AI such as ChatGPT — we put code in place in December 2022 requesting Common Crawl not scrape the Archive again. We cannot go back in time to stop data collection that already occurred, or remove AO3’s content from existing datasets, as much as we may dislike that it happened. All we can do is attempt to reduce such collection in the future.” (May 13, 2023)
It’s great that we are seeing more regulation in this space, but companies have scraped the open web for over a decade. They have tons of art, fiction and content to draw from and this is a problem.
Concerns from Artists and Fandom
These are just a few examples of how fan platforms and fan conventions are slowly speaking out about AI. Different organizations for instance already banned AI art from being sold at artist alleys, to make sure human artists receive profit and stay in the lead.
While AI allows for many playful interactions and inspiring outcomes, users are especially critical of the data used to train these networks as well as the unwanted monetization of AI art. Some tools spark more debate than others in fandom. For instance, an AI artwork generated in NovelAI circulated in Genshin Impact fandom which had clearly been stolen from the original artist, as reported by Kotaku. During a Twitch stream an artist was drawing Genshin Impact fan art of the character Raiden Shogun. A user took the in-process image, and created a similar image of Raiden Shogun with Novel AI, and then uploaded it six hours before the artist’s stream ended. This person claimed to be the original creator of the image.
The image generated in NovelAI clearly resembles the original artwork. To make things more problematic, the person who generated the image claimed their image to be the original creator. Such incidents contribute to the bad reputation of AI art. The varying quality of AI tools also adds an extra layer to these discussions. Some of these tools have little to nothing to do with generative art, and lead to the direct tweaking and filtering of existing art. That clearly should not happen. Other tools, such as Midjourney, lead to more nuanced discussions. They draw from large databases and generate art with unclear sources.
These black boxes also lead to many questions. How do we know if our art or fan fiction is used in these tools? Can we not opt in or out, and stay in control of our data somehow? Should content creators such as fans not get something in return if they contribute training data? Should their art not be licensed if it’s used in AI tools? What are the extends to which AI art should be sold at all? I could also envision a future where we treat it similar to fanfiction and fan art, as a form of remix that we allow and experiment with, but where making large profit is not endorsed. I would say that this type of transformative art should fall under fair use, but the problem is that this technology gets better and better. At some point, AI art will be hard to distinguish from human art. This makes regulation extra complex.
If AI art is the next phase of fan art, we need to seriously consider how to regulate it. We should also make sure that human artist get the profit and support they deserve, so please keep supporting the artist that you love!
Further reading
Nicolle Lamerichs 