I. Slop
One thing to know about Gothic horror is the way it moves: It starts with a surface — a veil, a veneer, or a skin. Then, like a poorly-popped pimple, it tunnels, growing deeper and darker in its irritation. It burrows, swelling into a cyst that grows more archaic, psychoanalysable, and harder to historicize as it takes its monstrous form. Rare is the narrative that describes why the Gothic monster stalks, why it hungers or why it spooks. It doesn’t chase or pursue. Its work is done by the time it breaks through. All we can do is watch it creeping and inching, always emerging like a heavy and slothful ooze.
This emergent motion is doubled over by the latest technologies and economies to play host to this drama of monsters: 19th century industrialization and imperialism, 20th century telecommunications and multinational capitalism, and now, Artificial General Intelligence accelerationism and technocapitalism. If the vampire was born from industrial smoke and the blob crawled out of the mid-century imagination, the last three years have seen streams of AI-generated slop give life to new figures in the shapes of mythical creatures, aliens, humanoids and beasts. Accounts like Instagram’s @catsoupai and TikTok’s @shadecore have reached millions with their phone-sized Gothic horrors: Siren bodies sit half-encased in ice, lizard faces with fixed eyes slowly twitch their tails as humans in hazmat suits shuffle all around, monsters are fished out of the dark blue sea and flopped onto fishing boats, their limbs rubbery with half-life, shimmering with the telltale quiver of a figure generated by a slush of data and software.
For the 2024 holidays, Coca-Cola released an AI-generated holiday ad with inconsistent focal blurs and liquefied gazes. Online commentators expressed dismay at the video’s “creepy expressions,” which evoked nothing but “death and loneliness.” AI-generated video stokes Gothic horror’s foundational fear of unstable boundaries, which congeals into a monstrous body that is always encroaching. Its metastasizing outlines and liquefied skins trigger what psychoanalyst Julia Kristeva calls “abjection,” meeting the Gothic monster with defensive disgust. “At the border of my condition as a living being” this disgust re-spawns the self with a reinforced outline for the ego’s own protection: “I expel myself, I spit myself out, I abject myself with the same motion through which ‘I’ claim to establish myself,” Kristeva writes.1 The undead eyes that sparkle throughout the Coca-Cola ad recall the kind of bodily breakdown that leads to Kristeva’s ultimate example of boundary-crossing, “the corpse, the most sickening of wastes,” itself “a border that has encroached upon everything.”2
On November 15, 2024 critic and curator Hilton Als posted what might very well be the first bit of AI-generated art to appear on his legendary Instagram feed: A gory video montage of elderly people melting into puddles of their own blood on sweltering city sidewalks. Flesh melts like flavored ice and blood pools into a sticky mess in this swirl of mismatched texture and movement. Nothing acts as it should and no line remains fixed in a world generated by programs like the oxymoronically named Stable Diffusion, OpenAI’s deep learning text-to-image model, and its liminally named competitor Midjourney.
II. Spaghetti
In 2023, a video of Will Smith eating spaghetti made using Stable Diffusion went viral as an example of the technology’s abject failures. Smith is seen in a handful of scenes with a disfigured face, shoving handfuls of pasta in the general direction of his mouth. Sometimes the noodles disappear into his cheek or chin, or they appear after his hand pulls away. Chewing sounds play asynchronously over the image alongside a voice calling out to “Uncle Phil.” Since then, “Will Smith eating spaghetti” has become an informal benchmark for text-to-video generators.
The first example of abjection Julia Kristeva describes in Powers of Horror is one of food3: “Food becomes abject only if it is a border between two distinct entities and territories.” In the 2023 video, we either count a dozen Will Smiths or lack the continuity necessary to perceive even a single, coherent image of the actor. If we attempt to assemble a singular figure or carve out a fixed shape from this mass of figurations, we identify both one and many monstrous images of Will Smith: some with bulging eyes, dislocatable jaws, disappearing teeth, sunken foreheads, or expanding skulls encased in a slick skin of greasy motion.There is simultaneously too much and not enough Will Smith to assemble a non-monstrous depiction of the whole.
Traditionally, a moving image is praised when it is smooth and continuous. What we often perceive as a singular character on screen is materially either a series of frames on a celluloid reel or a set of pixels pointillistically conjoining an image, a separate audio track, multiple takes, and sometimes more than one actor. But classical cinema prizes a film’s ability to bring these disparate elements together into a clear and legible narrative with minimal continuity errors, no plot holes, and consistent characterization. For Will Smith to look human and for the spaghetti to look worthy of consumption, their representations have to be consistently and unambiguously distinct and separate throughout the video. As newer text-to-video models hit the market, amateur users pursued these criteria. A 2025 attempt using KLING AI and ComfyUI shows greater visual distinctions between the actor and his meal: the spaghetti is solid and his skin looks impermeable, he moves one way and the food another; he raises his fork and the spaghetti discretely disappears into his mouth. It is clear that the two are separate until one neatly ingests the other.
Still, it lacks weight, friction and specificity. The technology is preoccupied with exploiting genericisms and generating a diamond-hard surface that shows no signs of where it is or where it comes from. The result is a blur of averages: Smith’s mouth opens too wide, and the pasta jiggles like jelly. It’s more sexual than spiritual. The meal — chasing a mouthful of spaghetti with a gulp of orange juice, a hamburger held as if it was a hash brown with the bun squished down like foam when bitten — is unhealthily rendered, like old frying oil. This over-coherence is Gothic anxiety at its worst, overly-invested in normalcy and disambiguation.
In Skin Shows: Gothic Horror and the Technology of Monsters, Jack Halberstam uses the term Gothic to name the moment when “interpretation becomes monstrous, spawns monsters and fixes otherness in highly specific sites.”4 If a Gothic novel’s task is to “unwind the messy skein of identities” to separate the “good from the ugly, the bad from the pure, the perverted from the kind, the sexual from the spiritual, the beautiful from the unhealthy,” then “the process of narration itself is Gothic.”5 When we watch these videos, assimilate their images, read the comments and, in my case, write about it in an essay, our interpretations spawn monsters as we read— in reality, there is no Will Smith, and there is no spaghetti, it is the narrative that produces them. This habit to consume and delineate, to outline figures and give order to ambiguity, are constitutive parts of this mode of monstrous production. These bodies are only partially machine-made, and human-made at a remove. AI doesn’t perceive images, it only knows its data analogues. It’s not creative, it’s generative. We’ve been spawning the monsters given form through AI for as long as there has been a dataset to train their models and for as long as we’ve been part of that data.
While the concept of uncanniness in technology first arose from a 1970 essay written by Masahiro Mori—a robotics professor at the Tokyo Institute of Technology— to assess product design, it can also be used to describe AI’s aesthetic outputs. Mori’s “uncanny valley” diagrams the dip in positive affect produced by a nearly-human-looking robot, puppet, or prosthetic. He had theorized that the more human something looks, the more we’d like it, until it reaches a point where it looks so human that minor differences are amplified into eeriness and abjection. In these cases, we experience the “eerie sensation” that constitutes the “uncanny valley” of this otherwise ascendant line. Central to this (un)naturalism is speed. Mori writes that “a smile is a dynamic sequence of facial deformations, and the speed of the deformations is crucial.” The difference between a good video of Will Smith eating spaghetti and a monstrous one, lies in the speed and sequence of the figure’s deformations — its ability to integrate just enough weight and friction.
III. Slime
Low quality AI-generated content struggles with borders and outlines and it should come as no surprise that popular critiques link AI slop’s brain-rotting potential to other kinds of abject substances: “the amalgamated gross style all of these videos have is like nauseating, i don’t know how to describe it, but it looks like the film is festering in real time,” one user shared on X. Another asked: “Can someone do a scientific breakdown on what it is about AI images that make them look so like,,,slimy? Glazed? I don’t know how to explain it but why do they all look GREASY.”
Citing Remu Bora, Eve Kosofsky Sedgwick notes that “to perceive texture is always, immediately, and de facto to be immersed in a field of active narrative hypothesizing, testing, and re-understanding how physical properties act and are acted upon over time.”6 By leading with texture — be it by accident or by design — the AI video undulates with questions about its material narrative: What direction is it moving in? Is it floating on air, water or zero gravity? Is it melting or are the lines rendered poorly? Is the shape-shifting diegetic? Would all this ambiguity be edited out if given the chance?
Texture, especially the gelatinous, squishy, trembling, slippery kind of the AI-generated figure, gives these videos their affective and aesthetic power. “A particular intimacy seems to subsist between textures and emotions,” Sedgwick writes, quoting Bora again to explain that texture “tends to be liminally registered ‘on the border of properties of touch and vision’.”7 Because it can be seen as well as touched, texture works from both far away and up close. It enacts the Gothic drama of emergence, pushing through the quiver of an outline or a porous malformation to break through the surfaces of commodification.
The 2007 introduction of the iPhone set off a trend in consumer electronics in favor of the smooth and shiny surfaces of polished aluminum and hard plastic. As long as one avoids the Uncanny Valley, a robot that executes frictionless motion is more attractive than one that moves to the breaks and clicks of its mechanics. And the same is now being said of us too. In the realms of beauty, skincare, and for-Instagram photography, “glass skin” is the reigning ideal: a texture that “signif[ies] the willed erasure of history,” a glossy mirror to 21st century consumer culture.8
On Instagram, Jess MacCormack shares AI-generated video portraits of doll-like figures with oversized eyes and faces made pearlescent by tears and makeup. The nasolabial folds are blurred by powdery light, browbones glisten with grease-paint, and the cheeks run with the high-gloss of thick tears. As the video cuts through various figures made-up in the same style, red lips part to reveal the mouth’s hyper-realistic, wet insides. Artists like @kentskooking use Midjourney and ComfyUI to render images and videos that feature organs, skins, nails and gums, people and animals swishing through and around each other. In one notable portrait, a man holds his shirt open to reveal his organs shifting around in what looks like a sous-vide bag filling his abdominal cavity. In another, texture fills the frame as Lisa Frank-colored fur shifts into hands while swatches of amorphous pink skin shine as if covered in oil or drool.
In the AI-generated video, gloss becomes slick, and smooth gives into squish as the technologies’ inability to offer coherent figuration is exploited into an over-coherence that approaches the grotesque side of cuteness.9 Movement turns AI-generated shine into a signifier for sweat, saliva, or disambiguated wetness. Shifts in light illuminate the quivering outlines— those gaps and lapses— that make way for the Gothic. The shiny, squishy-smooth surface of swollen over-coherence tempts with its wetness and excess. The pleasures of the Gothic lie in their titillating emergence from these exploitable holes.
A “perfected” AI-generated video — with subtly smooth motion and tasteful shine — might alienate these pleasures, or worse, create monsters that “stabilize bias into bodily form and pass monstrosity off as the obverse of the natural and the human.”10 Without its shaky outlines or slippery surfaces, the Gothic is just cruel; a stunted reading of the same rotted discourse about otherness. It’s worth rejecting the commercial sheen of the AI-generated video through Gothic readings that insist that “there is no one generic form that resembles ‘life’ and another debased form that deviates from the natural order of things.” 11
IV. Slick
In Postmodernism’s chapter about video, Fredric Jameson unearths how “the deep underlying materiality of all things has finally risen dripping and convulsive into the light of day; and it is clear that culture is one of those things whose fundamental materiality is now for us not merely evident but quite inescapable.”12 Behind the shine of the glass smartphone screen and the neat interfaces of text-to-video software, stir the hungry beasts of wealth consolidation and resource exploitation. Behold, the material behind the software and the exploitation in excavation!
The push for accelerated AGI development is itself a push to colonize energy production and natural resources stretching centuries into the past and the future. In Geology of Media, new media theorist Jussi Parikka explains how “fossil fuel use offers access to carbon stored from millions of years of photosynthesis: a massive energy subsidy from the deep part of modern society, upon which a great deal of our modern wealth depends.”13 AGI acceleration of the past few years has ushered in a fossil fuel resurgence. The energetic cost of maintaining the data centers used to train generative AI is expected to place AI’s energy consumption between that of Japan and Russia by 2026. Never mind its water usage. Never mind how the demand for rare earth metals needed to produce AI’s powerful semiconductors contributes to the violent conflicts in Sudan and Congo, and the exploitation of child laborers worldwide.
Is it surprising, then, to learn that “the emergence of industrialization since the 19th century and the molding of the environment with mines, smelting facilities, and sulfur dioxide from coal energy”14 also gave birth to the Gothic monsters we know today? Halberstam begins his history of these monsters in the nineteenth century, which “metaphorized modern subjectivity as a balancing act between inside/outside, female/male, body/mind, native/foreign, proletarian/aristocrat.” Through their category defiance and boundary perversion, monsters “reveal certain material conditions of the production of horror.” They press a sheet over the face of normalcy to suffocate its pretenses.
“We inhabit a world in which we are confronted constantly, if intermittently, with spectacular displays of aesthetic power, often in close coordination with displays of financial, political, and military might,” Sianne Ngai writes.15 I want to reintroduce AI-generated horror as a specific breed of the Gothic monster that the consumer co-creates through her subordinate alignment with the financial, political and military might that AGI evangelists are successfully capturing. Videos of cats and families shape-shifting into pinkish internal organs with raspberry intermissions, a time traveling ring of criminals in various shades of necrotic flesh and metallic blue sweat, a frantic game show of dolphin-human hybrids competing to eat towers of fleshy wedding cake — as wide-ranging as they can be, these made-for-the-feed horror videos all bear the signature slick and shine of their AI origination, as thin and flavorless as spit to make the underlying violence go down easy.
These spectacular displays of aesthetic power work through our online subjectivities. In Ngai’s book on 21st-century aesthetics, she writes that “the forms that our aesthetic experiences of the cute, the interesting and the zany revolve around—the squishy or extrasoft blob, the open-ended series, the incessant flow—are thus relatively shapeless or unstructured.”16 Yet it would be mistaken to think that today’s squishy, disjointed and liquefied Gothic monsters are the end of the line. As generative AI moves towards conquering the whole of perceptible reality, it will pursue tighter figurative economies and more cohesive subjectivities in an attempt to suffocate the Gothic or worse, force it to “stabilize bias into bodily form” by giving it a diamond-hard finish.
V. Surprise
Halberstam defines Gothic as “the rhetorical style and narrative structure designed to produce fear and desire within the reader.” I like to think of the Gothic as a texture and a motion, always pushing through and bleeding in. Theorists—including Kristeva, Jameson, Parikka and Halberstam—regularly indulge in the Gothic drama of emergence. They dig, uncover, reveal, demonstrate, and illuminate their arguments, as if pulling them out of earth and shadow. They follow an argumentative hunch to unveil a hive of interlocking ideas. Sedgwick even complains that “anything but a paranoid critical stance has come to seem naïve, pious or complaisant” in our fucked-up world of violence and abuse.17 It is an all-too common affective and aesthetic routine, “exposing and problematizing hidden violences in the genealogy of the modern liberal subject” through “infinitely doable and teachable protocols of unveiling.”18 Behold, things have histories! Behold, we live in a society! Behold, I read Sedgwick as saying, we’re all paranoid!
Writing about video — the post-modern medium par excellence — Jameson describes “helpless spectators” that are as “immobilized and mechanically integrated and neutralized as the older photographic subjects, who became, for a time, part of the technology of the medium.” In the case of the AI-generated horror video this effect is threefold: We’re all accounted for in its datasets, we might be involved in co-creating it through prompts, and we’ve probably come across them in our feeds.
An inversion is in order. In a 2021 talk for the Unsound Festival, Benjamin Bratton takes aim at the unquestioned ideologies that guide AI acceleration, “which are in turn hobbled by very clumsy misconceptions of what is and what is not artificial and thus what is or is not intelligent.” He gestures towards an “Inverse Uncanny Valley,” wherein we see ourselves “through the eyes of the machine" and are disturbed by our inability to recognize ourselves. I read this as an invitation to stay with the monsters, conduct the Gothic drama of emergence in slow-motion, giving ourselves time to make out the fears and anxieties that brought it to life, taking the chance to reincorporate them before we run. If what we see through the gaps of the AI-generated image is ourselves, then the Gothic has done its job well.
We will surely be threatened by these “confrontations with what we are but don’t imagine ourselves to be.”19 We find ourselves paradoxically detached but conjoined to the current of AI-generated video. A series of quivering outlines, discontinuous borders. We’d see ourselves implicated and complicit, in need of a stabilizing narrative to pin the monster down and motivate our protests, boycotts, and divestments from generative AI as it currently exists. But we also have to see ourselves as perpetually misshapen, lava-lamp people with fluid centers that live for the Gothic drama of emergence by moving through cycles of disgust, pleasure, shame, recognition, alienation, laughter, and exaltation. ◼
.jpg)
