by Olivia Hunter Willke
“A collaboration between a Human and AI,” read the credits at the end of Per Bifrost’s HYPNOSUGGESTION (2023). I wondered how the images were created as I watched this short film unravel at the Onion City Experimental Film Festival in Chicago. An uncanny haze surrounded each new flick of a frame, a sense of detachment in the seemingly imprecise blur that permeated each depiction of whatever object, person, or entity appeared. The synopses pitched it as “a hypnotic sci-fi thriller consisting only of still images generated by an AI.” A somewhat rhythmic pasting together of smeared, mildly engaging “sci-fi” images, a blurred infrared human brain floating in a forest, or a massively oversimplified view of space from a window.
Moments before, I had the pleasure of witnessing Federica Foglia’s stunning analog film Glitter for Girls (2023), which uses found materials in the form of temporary tattoos applied directly to the 16mm film—an organic exercise in blurring the line between the artist’s skin and the film itself. Foglia described the painstaking process of applying temporary tattoos frame by frame to form a colorful and innocent animation accompanied by random sounds of children’s toys taken from sites like freesounds.com. It’s a film, like most experimental and analog films, that is deepened and enhanced by its processes. It is often these processes in which I find meaning or a thread of meaning that connects to the already deliberate images on the screen.
The less evident contribution was Prapat Jiwarangsan’s Parasite Family, described as “constructed from film negatives discovered in an out-of-business film lab, the film traces a journey from analog to digital, to the world of AI-generated images.” The reason this is a less obvious example is that as the film morphs into AI-generated images, and one could easily mistake the faces displayed, mutilated, and morphed for computer-generated renderings that were then manipulated digitally by hand, through a few clicks and drags of a mouse. A cheek could be enlarged, an eye could drag, etc. But it would take longer, much longer than telling a computer to do this for you in rapid sequence.
Parasite Family (2022) by Prapat Jiwarangsan
Observing these works so closely aligned with other analog and digital experiments was troubling. The difference in quality and attention to detail was glaringly obvious, not to mention the discernible misalignment of what the computer was supposed to portray vs. what it actually portrayed. A clear example is a sky with sharp edges where the image morphs from one mined likeness of a cloud to another. The uncanniness of such a sky does not invoke surreality; it feels heartless, without soul or human touch or even a human eye to catch such a distinctly confused depiction.
The Ann Arbor Film Festival, a home for underground film, showcased an AI-generated music video. In SOMNIA (2023) by Farbod Khoshtinat, an AI-generated “woman” “sings” a track by duo Skycabin in front of shifting blue-and-white-patterned backgrounds. Again, the same smeared visuals and texturally vacant swaths of image present themselves. Although the “light source” is fixed in front of her “face,” the algorithm can’t quite keep up with a human eye, casting shadows and light where they shouldn’t be. In short, it is a messy void. It is a work I wouldn’t expect to be accepted by a music video festival, much less a film festival renowned as a beacon for underground film. The festival’s lack of disclosure in its program listing is even more disturbing. Not one mention of AI was made. This work was presented on par with all the rest, with no distinction, no statement on the difference of process, or even a nod to the use of computer imagery, artificial or not.
Art created with generative artificial intelligence uses algorithms that mine the internet for images, text, and works created by artists, then process those through a system that, in short, smashes it all together to create something “original.” Not only is AI-generated work not indebted to a creation and assembly process in the same realm that experimental film and video occupy, but it is also often artistically disingenuous. Unlike people paying homage or wearing influences on their sleeve, AI can not credit its influences. Furthermore, the people using these systems can not see what work was mined, from where, or whom, making it impossible to check for plagiarism or stolen copyright material. AI can not evolve independently; it must be taught, and it must be taught with others’ work without the ability to distinguish between influence and duplication. The ethical concerns regarding intellectual and artistic property are one thing, but the environmental concerns are another. It is estimated that a search driven by AI uses four to five times the energy of a conventional web search. Generative AI also uses massive amounts of fresh water to cool its processors and generate electricity, increasing some districts’ water usage by up to 34%.
Artists have always experimented with new technology, testing the limits of what they can accomplish with a new tool as a means to expand the medium. Once that technology is old hat, discarded for something new and better that has come along, the so-called faults become endearing and enduring: the grain of film, the static of VHS, the pixelation of early digital. Those qualities are caught up in nostalgia and used for homage. But there is no aspect of generative AI art to single out, no quality that stands out, nothing that can evolve. Because it is all borrowed, there is nothing to borrow. When a prompt is run through the system and spits out an image, it’s a cheap mock-up of what a computer algorithm “thinks” these qualities are and how they operate within a work. AI misses the mark of what makes art feel so connected and tactile. Art born from a human mind can only be done so because each mind is unique and the tools they possess to execute that idea are also unique. Not everyone has the same skills, access, mobility, privileges, and utility. You can adjust a code, add, detract, but tweaking a system is not the same as going through the thought processes that lead to creation. That internal search for meaning, a “why?” and “how?”, those decisive pauses are what make the act of creation so beautiful. Art lives and breathes in those in-between moments, those pauses to think.
Olivia Hunter Willke is a film writer, analog filmmaker, and programmer based in Chicago by way of Texas. Her work blends political urgency, formal analysis, and emotional revelation.
Header image: Glitter for Girls (2023) by Federica Foglia
If you enjoyed this article, please consider tipping the author and/or supporting Ultra Dogme on Patreon or Ko-fi, so that we may continue publishing writing about film + music with love + care.
Nice article. I also recommend anyone interested in this AI art non-sense also read Wesley Goatley’s series of critical articles at https://wesleygoatley.substack.com/.
The music video was a great example of everything that is wrong with this style of art. The tell is the constantly shifting, smeared, jittery images. It’s obvious that the artists do not want their videos to have this effect, but they do not know how to avoid it. The tools are not giving them what they want, but they have this overwhelming desire to use the newest tools and so they accept these aesthetically heinous products. Then they submit this stuff to film festivals! Just compare this video to a random music a music video from the 90’s! It boggles the mind. We have cheap, ubiquitous cameras and totally free video synthesis, compositing, and non-linear editing software. The only reason to make art that looks this bad is laziness or mindless novelty seeking.
I also cannot stand when “generative AI” users try to rationalize the instability as “dream-like”. My dreams do not vibrate or morph randomly!!! All the humans in my dreams have 5 fingers!
All that being said, soon enough someone will make something actually decent with these tools. I just haven’t seen it yet.