We don't need better films.
We need to stop breathing the videosphere blindly and start building the conditions for conscious dwelling.
That was the whole talk. What follows is the argument for why dwelling matters, and what it demands of us in an expanded essay.
Let me state the thesis, before anything else:
We are not watching the videosphere. We are dwelling inside it, blindly, by default. The only question is whether we build the conditions for conscious dwelling, or keep breathing until there is nothing left to breathe.
Twenty years of INC's online video network VideoVortex. Twenty years of this conversation about alternatives. Most of our critical art still lives on Instagram. I want that sentence to sit uncomfortably for a moment, because everything that follows is written in its shadow and not as a manifesto for something new, but as an account of why the old conversation kept failing, and what a different one would have to look like.
The Videosphere
The television stood in the corner of the room. There was a frame around it, literal and figurative, and the frame preserved a critical distance: you watched, and watching implied an outside from which to watch. That frame is not weakened. It is gone.
A family watches as US astronaut Neil Armstrong sets his foot on the moon July 20th, 1969. Photograph: STR / AFP/Getty Images
The shift is often narrated as a matter of degree. More video, more platforms, more hours of the day spent scrolling, but the shift is better understood as a change of kind. In the Web 2.0 moment of the 2000s, video was one content type among others, alongside text, images, and links. In the YouTube decade of the 2010s, video became the dominant content type. By the 2020s, video is no longer a content type at all. It is the interface through which everything else happens: education delivered as lecture clips and TikTok explainers, sociality conducted through Stories and Reels and video calls, warfare witnessed through drone footage and helmet cams and livestreamed violence, shopping performed as unboxing and live commerce, memory captured and stored as video by default, identity itself performed on camera as an ambient, unexceptional condition rather than a special occasion. We are not watching the videosphere. We are dwelling inside it, whether we know it or not. There is no longer a position outside from which critique could be positioned. The hybrid condition of being permanently on camera is not an event. It is the weather.
“Me at the Zoo” – the first uploaded video on YouTube in 2005 by Jawed Karim
This atmospheric claim is not a metaphor dressed up as theory; it names a specific, recent shift in what the videosphere is made of. Writing from his sickbed during the May 2026 re:publica conference, Marcus Bösch documented a qualitative change in what actually fills the feed: his For You Page, he wrote, had become "infested with generative AI content not a few glitches at the margins but hundreds of near-identical videos running the same templates. This is not contamination arriving from outside the videosphere. It has become part of the videosphere itself. The videosphere has become a space in which synthetic types are continuously deposited into the collective imagination at scale and largely invisibly. The term for this genre of disposable, algorithmically optimized content is "slop." Bösch notes that slop strategies have been evolving since the earliest viral AI imagery. These strategies are now sophisticated enough to manufacture effective hooks that did not exist before.
If Bösch documents the videosphere becoming infested, the Palestinian artist Dana Dawud, writing in the zine accompanying her Open Secret program, describes the endpoint of that infestation: total naturalization. "Being online is unremarkable,"[1] Dawud writes; audiences are, in her phrase, already post-post the gesture of pointing this out. We are no longer adapting to being online. Online is the default condition, and everything else is now the deviation from it.
Hadi Afif draws out the implication in his account of internet cinema for Do Not Research: Chuck Tryon's "digital everyday," coined in 2007 to name internet life as a new object of study, has simply become the everyday.[2] The prefix has dissolved. This is the final stage of atmospherization. It is not immersion in something new, but the disappearance of the boundary between the inside and the outside of the medium altogether. If the videosphere has been fully naturalized, the demand for art and infrastructure alike is not for content that dramatizes a break with the medium; that gesture is precisely what audiences are past. The demand is for infrastructure that produces different conditions of inhabitation within a medium that cannot be exited.
That last clause needs its own grounding, because it does more philosophical work than it appears to. Dylan Smith, writing in "On Internet and Internet Cinema," argues that contemporary criticism makes a basic ontological error: contemporary criticism treats the internet as something novel and exterior to everyday life, something that must be signaled through visual reference, the painted meme, the digitally-themed sculpture, the sculpture of a smartphone. This is not merely bad art. It rests on a false premise about what the internet is. The internet, Smith argues, is not an object or a container; it is a fundamental extension of human consciousness, in the same sense that television, radio, and print were before it.[3] Every work made today exists simultaneously in the devirtualized and virtualized worlds, which means there is no outside position from which to represent the internet. The artist working now is already an internet artist, whether or not the work knows it. The videosphere is not a thing you move through. It is a structure of consciousness you dwell inside.
Fakewhale's essay "Post-Media Logics and Conditions" provides the dimension that the ontological and naturalization arguments leave implicit: the videosphere not only surrounds us but also operates on us neurologically. Entertainment, Fakewhale writes, is "no longer content to be consumed but an environment to be inhabited, and one that inhabits us in return."[4] The spectator becomes an active node in what Fakewhale calls a bio-computational network, in which mediation adheres directly to biological rhythms through tactile interfaces, variable rewards, and dopamine-optimizing loops that segment time into micro-impulses. Television acted upon the mass audience; the social network acts on the individual's nerve fiber. This is the transition from centralized narrative control to the molecular management of desire, and the neuroscience behind it is not rhetorical: recent studies link frequent engagement with recommendation algorithms to altered dopamine pathways and a dependency built on variable reward schedules structurally similar to those of a slot machine. Researchers studying heavy short-form video use document measurable time distortion and difficulty sustaining attention on unrelated tasks.[5] Boredom, the fertile void in which unscheduled thought forms, was not an accident these systems failed to prevent. Its elimination was the engineering goal.
This is where the distinction between breathing and dwelling becomes necessary, and where Peter Sloterdijk's reading of Rilke sharpens an argument that Heidegger alone cannot finish. We are already breathing the videosphere: involuntarily, continuously, neurologically shaped by it. You do not choose the dopamine loop any more than you choose to inhale. But breathing is also, in every tradition of bodily practice, as by the monk, the martial artist, or the yogi, the first site of discipline. Before any higher practice becomes possible, one learns to be conscious of the breath: not to stop breathing, but to stop breathing blindly. In You Must Change Your Life, Sloterdijk takes the closing line of Rilke's "Archaic Torso of Apollo" — there is no place that does not see you; you must change your life[6] — as what he calls the absolute imperative, the command that overshoots every other. The torso sees you. You cannot step outside its gaze. The only response available is not retreat but transformation: askesis, self-cultivation, the practiced life. The videosphere is our archaic torso. We cannot step outside its gaze. There is no place it does not see us, no critical distance from the atmosphere we breathe. But that is not the end of the argument. It is the beginning of the practice. Dwelling, in this sense, is conscious breathing: the discipline of inhabiting the videosphere with intention, and the willingness to build its conditions differently.
Archaic Torso of Apollo by Rainer Maria Rilke.
If video has become the videosphere, and the videosphere is a structure of consciousness rather than a container, then building alternative infrastructure cannot mean building a different box. It means building different conditions of consciousness, different structures within which attention, thought, and affect operate. The question this raises, and that the rest of this essay tries to answer, is whether pockets of a different videosphere are possible, or whether any alternative infrastructure we build is simply the same atmosphere redecorated in different filters.
The Content Laborer
The capture has accelerated faster than the resistance.
The vocabulary we use to describe participation in the videosphere has not kept pace with what that participation actually entails. "Prosumer" — the term that defined the Web 2.0 moment — still carries the optimism of its coinage: the consumer who produces, the amateur who publishes, the user who is also author. That optimism was not entirely false. Something real shifted in the 2000s: the barrier to publishing moving images collapsed, along with a certain monopoly on who got to speak in the medium. What the term concealed was the infrastructure that opening depended on, and what that infrastructure was doing in return. You could publish to the world, but only through a platform that owned the distribution channel, set the terms of visibility, and extracted value from every view. "Prosumer" is the name of the production side of the transaction. It did not name the extraction side, or the degree to which the platform's architecture was not neutral infrastructure but an active shaper of what got made.
Sony HDR-FX7 HDV Camcorder: Prosumer HD Video in 2006 - Reviewed - YouTube
By the 2020s, the extraction side is no longer concealable. The correct term is content laborer. You are working, for free, under algorithmic management. This is not an expression; it is piecework. The Stack governs not just what gets distributed but what gets made in the first place: TikTok's verticality means your body must fit 9:16; YouTube's recommendation engine means you must optimize for watch time; Instagram's engagement metrics mean you must post daily or disappear; LinkedIn's demand for "authenticity" means you must perform vulnerability to sell. Platform aesthetics are not a byproduct of how users choose to present themselves. They are production constraints imposed from above, as surely as the specifications a factory floor worker receives for the objects they are expected to produce. The creator economy, in this light, is a burnout economy: constant presence required, no sick days, no unions, no healthcare, payment in exposure. The language of "following your passion" performs exactly the ideological function that "prosumer" performed a decade earlier, obscuring a labor relation behind the vocabulary of individual self-expression. Speed is a platform metric. Durability is a human metric. The content laborer produces for the platform's temporality: disposable, optimized for immediate engagement, replaceable the moment attention moves on.
Fakewhale names a further dimension of this labor that the prosumer/laborer framing typically omits. While we entertain ourselves, we are training the machines that will govern us, providing them with the emotional maps needed to perfect persuasion and behavioral marketing.[7]
Content labor is not only about extracting value. It is the data substrate from which platform AI learns to predict and precondition desire. Creators working under algorithmic management are not simply exploited as laborers; without consent or compensation, they supply the training signal that lets engagement-optimization systems improve.
Every scroll, pause, share, and completion rate is processed back into the recommendation engine, so that the creator's labor and the viewer's attention are simultaneously content production and machine-training data. As Fakewhale puts it, the extractive logic of attention capital transforms every moment of leisure into an act of data production, nullifying the classic distinction between free time and labor time.[8] There is no off-clock, and no activity inside the platform environment that is not also work, not for the user's benefit, but for the machine's improvement.
The implication for any alternative infrastructure is that it cannot simply offer different content conditions. It must break the training loop, refusing to process user behavior as an optimization signal, refusing to feed interaction data back into a recommendation engine, refusing to transform creative practice into machine-training labor. This is not a feature request. It is the minimum condition for infrastructure to be genuinely non-extractive, rather than extractive, with better aesthetics.
Algorithmic Clinicality and the Image Problem
Platform aesthetics are not neutral. They are clinical: smooth, optimized, frictionless, A/B tested, and retention-maximized, and the result is a pervasive feeling of being managed, of moving through an environment designed by engineers who know exactly how long you will look before you scroll. Call this algorithmic clinicality: the condition in which aesthetic form is not the expression of human intention but the output of optimization processes running continuously against behavioral data at scale. Its symptoms are everywhere once named. Ring light uniformity: every talking-head video is lit from the same angle and color temperature, not because creators independently converged on a preference but because the platform's training data rewards it. Thumbnail face: the exaggerated shock or delight, mouth open, eyes wide, calibrated for click-through rather than emotional authenticity. The 15-second hook, the vertical crop, the background blur that eliminates context and forces the eye onto the face, each a constraint imposed by algorithmic feedback, worn so thoroughly it now feels like a creative decision.
The deeper problem is what happens when resistance to the clinical gets absorbed into it. TikTok's VHS filter is clinically designed. Instagram's film grain is algorithmically applied. Lo-fi is a genre tag, not a material condition. Even authenticity is optimized, and the result is a state of total aesthetic optimization in which no choice exists that the algorithm has not already anticipated, cataloged, and converted into a preset. Rebellion is pre-packaged. Imperfection is a slider. The clinical has absorbed its opposite, which demonstrates that the problem was never stylistic in the first place. The time from the emergence of an aesthetic to a platform preset is now measured in months. Cottagecore, Y2K, indie sleaze, and corecore are all harvested and recycled before the subcultures that generated them have finished forming. And AI training accelerates this further, learning not just to reward emergence but to simulate it, producing the appearance of organic novelty from within the optimization loop itself.
Cottagecore
Bösch's analysis shows where this logic terminates. He references Roland Meyer, who argues that AI-generated images are now optimized not against the real world but against other images already in circulation: trained on billions of images from platforms and rewarded for matching what viewers already expect to see. Meyer, building on a term from Jacob Birken, calls this platform realism,[9] a realism that has nothing to do with reality and everything to do with image-familiarity. This is algorithmic clinicality at its logical conclusion: the feedback loop no longer references anything outside the platform. What is being copied is not rural life, but how rural life is already represented on platforms. The image operates on familiarity rather than evidence, sidestepping the entire logic of truth and falsehood because it makes no claim that could be checked against anything. Meyer adds that such images are disposable by design: producers generate dozens to find one that engages, and most are never seen. The figure in a given clip is not a stable character whose existence could be verified. The figure is a test image, kept because she happened to perform.
Yanai Toister and Joanna Zylinska's concept of cognitive hacking [10] identifies a harm that runs deeper than aesthetic appropriation. Every time you see an AI-generated image, even one you know is fake, even one you mock or scroll past quickly, the type it represents becomes a little more familiar to your imagination. Across thousands of variants seen by millions of viewers, this slowly reshapes what feels normal, what feels imaginable, what counts as a recognizable shape in the world. Sincere belief, ironic mockery, erotic interest, critical analysis, and scrolling indifference all feed the same engagement metrics and deposit the same type of engagement.
The artifact does not need a particular response. It needs only exposure. This is why a post-aesthetic argument cannot rest on unharvestable style alone. It also needs infrastructure that resists the volume logic on which cognitive hacking depends.
Dylan Smith supplies the missing ethical register for this diagnosis. Platform realism describes a structural condition. Images are trained against other images rather than against reality. Smith's contribution is to ask why this matters ethically, for the practice of art, and his answer is precise: producing self-referential internet imagery, content that signals its own internet-ness without referring to anything beyond itself, is not merely aesthetically weak. It is an act of violence against the image.
Smith treats the image as a unit of language, drawing partly on René Guénon's distinction between sign and symbol: a sign operates within a closed circuit, taken at face value, its meaning imposed by external force and replicated through media networks; a symbol acts within a limitless space bound to something beyond the purely representational, and has affect. Modernity, on this account, has progressively collapsed symbols into signs, drained images of their affect-bearing capacity, and reduced them to their platform-assigned function. When an image refers only to itself, it fails as language: it becomes, in Smith's phrase, a reproduction of a reproduction, a commodity perfectly distilled to its essence, a pure nothingness. Smith's term for the failure to correct this is psychic terrorism: leaving an image trapped in its closed circuit and used as engagement bait without correction or liberation is a form of violence against it.
Which is to say when you do not actively work to liberate the image from itself, you participate in a form of psychic terrorism.[11]
And this diagnosis applies to bad human-made internet art exactly as much as to AI slop. The painted meme and the iPhone sculpture are no less sins than slop. They are the same sin: art that farms the internet for a revelation, then presents it to people who already live there, keeping the image tied to its country of origin rather than letting it discover what it might mean beyond the conditions of its production.
The alternative Smith proposes is alchemical recomposition: treating images as if they were raising the dead, giving them affect, and orienting them outward toward something they do not already contain. Smith's historical model is abstract expressionism. Harold Rosenberg's observation that the act of painting is of the same metaphysical substance as the artist's existence[12] describes painting as an event rather than a static representation, breaking the frame that separates art from life. The challenge for image-based practice under platform conditions is the equivalent liberation: the image must be given a way to move beyond its sign-value, to enter the register of affect, to gain a voice. This is not a stylistic instruction. It is a structural one, not an aesthetic preference. It is an ethical obligation that every image-maker meets, whether they name it or not. Most do not name it. That is the problem this essay is written against.
That the counter-practice already exists, and does not need inventing, is the argument of the next section.
Internet Cinema
The good news — and there is good news — is that a counter-practice already exists and does not need to be invented. Hadi Afif traces its lineage from the Cuban filmmaker Juan García Espinosa's 1969 essay "For an Imperfect Cinema" — it is no longer a matter of replacing one school with another, one "ism" with another; let a thousand different flowers bloom, the future lies with folk art — through Hamid Naficy's account of internet cinema during the 2009 Iranian Green Movement protests, through the CoreCore wave on TikTok, to what Afif calls internet cinema proper: decentralized, collectively authored, oriented toward eliciting affect rather than conveying explicit meaning. The crucial distinction, drawn via the Open Secret program and the theorist Onty, is between art about the internet and art that is the internet. Post-internet art still imagines a position outside the medium from which to comment on it. Internet cinema abandons that position. It does not depict collapse. It enacts collapse through its form, drawing on the same surplus of platform imagery as slop, the same feeds, the same scroll logic, and doing the opposite with it.
That opposite operation is worth stating as sharply as Afif states it: a machine can automate logic. It cannot affect the curate.[13] Slop automates affect through statistical aggregation, longing, acceptance, nationalist sentiment, and erotic interest, produced by optimizing against engagement metrics and cycling through affect-bait registers until something converts, with the model able to synthesize whatever body the affect-bait requires. Internet cinema curates affect through human intentionality that is historically situated, politically deliberate, and formally precise.
The works Hadi Afif traces, Dana Dawud's Palcorecore, Redacted Cut's RNG series, Onty's UltraCore Anxiety, draw on the same flood of platform signifiers but reframe it through curatorial intelligence.
Palestinecore's power lies in its archival impulse: drawing on visual motifs of Palestinian resistance and stitching them into imagery that refuses to fully articulate its meaning while making its intention crystal clear. That is not a style. It is a structural choice about what images are for. The distinction is not between human and machine labor. Both slop and internet cinema involve human hands at some point in the pipeline. The distinction is between a practice that uses human judgment to set affect in motion and a system that uses engagement metrics to harvest and optimize affect until it deposits types into the collective imagination. One is curatorial. The other is extractive.
PalCoreCore (2023) by Dana Dawud
Smith's sign/symbol framework gives this dialectic its sharpest formulation. Slop performs what Smith calls psychic terrorism: it leaves the image trapped in its closed circuit of sign-value, used as engagement bait without correction or liberation. It cannot speak for itself, and because it cannot speak, it cannot claim autonomy. It degrades into a pure commodity. Internet cinema, by contrast, performs alchemical recomposition: it uses images, in Smith's phrase, like raising the dead.
Smith's central example is Dawud's Monad, in which blur, projection mapping, and light work together to dissolve the sign-value of found images, rendering them imperceptible and thereby freeing them to function at the level of affect, becoming invisible, imperceptible, intangible in order to displace the image from its imposed position and let it begin to function autonomously.[14] The sign is returned to the symbol. Any curatorial infrastructure built on this distinction has to ask not whether content is human-made or machine-made, high-resolution or low-resolution, but whether the images within it are being liberated or enslaved, oriented outward toward something beyond their sign-value, or trapped in the closed circuit of platform optimization. Infrastructure that enables enslavement, however "alternative" it claims to be, is just another variety of slop.
The lineage matters for another reason: it breaks the sense, common at gatherings like this one, that we have been talking about alternatives for 20 years without result. Internet cinema did not fail. It did not get captured. It proliferated — #corecore alone reached billions of TikTok views — without becoming a platform product, because it has no manifesto, no center, no canon, no ownership to acquire.
As Afif observes, any claim to ownership over internet cinema is futile, because the internet itself refuses canon. That anti-canonical structure is an infrastructural property, not merely an aesthetic choice, and it is what makes Dana Dawud's Open Secret program a working model rather than a hypothesis, thanks to what Afif describes as an autonomously organized, ever-growing network of artists and internet friends[15].
Open Secret is manifesto-less. It has no rulebook, no center, no ownership structure to acquire or capture. Each screening, from Tehran to Dubai to Jeddah to Beirut to Milan to Montreal to New York to Perth, brings a different set of films or new iterations of earlier ones; as context and conversation shift, the program shifts with them. This is not a design choice. It is the structural consequence of refusing canon. Dogme 95 defined itself by a manifesto and a Vow of Chastity. Internet cinema, contrastingly, has no vow, no rulebook, no center to pin down. It just is, and it has sustained a practice across years and continents without requiring either institutional capture or canonical consolidation. That any such thing could be sustained without either is the proof this essay needs before proposing anything further.
Human Temporality against Platform Temporality
Speed might be the wrong metric entirely. It is a platform metric. Novelty, virality, disposability. Fakewhale's account of what it calls "time without waiting" provides the most precise description available of what platform temporality has actually done, not as metaphor but as a documented neurological and phenomenological condition.
Television operated on a chronological scan, with pauses, commercial breaks, and weekly appointments. Social networks, particularly short-form video platforms, introduced a regime of absolute instantaneity, eliminating temporal friction and offering an uninterrupted flow of stimuli that regenerates infinitely at the touch of a finger.
What has been lost is not just time in general. It is waiting specifically, a space of latency necessary for the formation of thought[16], and boredom, the fertile void in which the mind wanders without a goal predetermined by the logic of profit. Boredom was not a system defect. It was a protective mechanism of the human that entertainment engineering decided to eradicate in order to maximize biological availability to the flow. The destruction of boredom was not a side effect. It was the design goal.
The engineering of digital time fragments attention into increasingly short micro-units, short video, flash post, push notification, preventing the construction of complex narratives and articulated arguments, and producing an eternal "now" devoid of historical depth, a plane of total immanence in which the past is archived quickly and the future is reduced to the next imminent interaction. This temporal contraction produces a form of collective amnesia in which information overlaps without settling into knowledge; heavy short-form-video users exhibit measurable time distortion and significantly greater difficulty sustaining attention. Digital immediacy satisfies the need before it becomes desire, condemning the user to a condition of satiety, fullness without pleasure.[17] The erotic dimension of time, the tension toward what is not yet present, is engineered out of the experience.
What would it mean to build for human metrics instead of platform metrics?
Four dimensions matter, and none of them are engineering specifications so much as structural conditions for a practicing life. Durability: not what's new, but what persists.Infrastructure that lasts ten years rather than ten months, aesthetics that age and accumulate history rather than a logic of disposability. Inhabitability: Can we live inside this? Is it sustainable for the people who make it? Does it support a practice rather than merely a product? Maintainability: who does the work of keeping it running, can that work be collective, visible, and valued, and what happens after the grant runs out? And a fourth dimension that the account of platform temporality makes necessary: restorability.
Infrastructure that does not merely persist, support, and be maintained, but is actively designed to recover the capacities that platform temporality has eroded: sustained attention, tolerance of latency, the capacity to dwell on a work without the next stimulus automatically arriving, the boredom that is the precondition for genuine desire. Sloterdijk's framework names what this restoration actually is: not a technical fix but an askesis. Restorability, in particular, maps onto pranayama, breath practice, the deliberate recovery of a capacity that was always there but now requires conscious cultivation because it has been systematically suppressed. Infrastructure designed for restorability treats its users as practitioners in training, not as consumers to be optimized: friction by design, where latency is an invitation rather than an error; no autoplay, so the next thing does not arrive unless summoned; no engagement metrics surfaced to users, no like counts or view counts or completion rates; temporal depth, so that work accumulates across sessions rather than resetting with each visit; the restoration of waiting as an experience rather than the elimination of it as an inconvenience.
The question any alternative infrastructure has to answer, then, is not only what supports human temporality rather than algorithmic temporality, but how to build for users whose capacity for sustained attention has already been altered by years of dopamine-loop design. Infrastructure that ignores this will simply be slower and less convenient than the platforms, without offering anything that compensates for the friction. Infrastructure that takes it seriously has to be designed explicitly to support the recovery of attention it asks its users to practice.
Making Infrastructure Visible
Most people do not know where "the cloud" actually is. TikTok users do not know their videos are hosted by Oracle. Platforms that call themselves decentralized still often rely on AWS. Infrastructure is invisible by design. It is supposed to just work. But invisibility is unaccountability. You cannot critique what you cannot see, cannot exist what you do not understand, and cannot build alternatives without knowing how the current system actually operates. The media theorist Jussi Parikka calls the layer beneath visible images in-visuality[18]: the image-work that happens before images become visible to humans at all, including moderation pipelines, recommendation algorithms, AI detection systems, labeling infrastructures, and decisions about what gets shown to whom and under what conditions.
Harun Farocki, Serious Games III, 2009. Two channel video, 20 min. Courtesy of the Harun Farocki estate. Operational Images: Between Light and Data – e-Flux Journal #133 February 2023
Markus Bösch documents a specific, concrete instance of in-visuality in action. The metadata for AI-generated TikTok videos includes a field called aigc_label_type, which is set to zero.[19] That field is TikTok's marker for whether content is AI-generated. Zero means no label appears to the viewer. Whether the system checked the video and decided no, or whether zero is simply the default written when nothing has been checked, cannot be determined from outside the platform; either way, the user sees no label. TikTok's policy nominally requires creators to self-label realistic AI content. The accounts producing slop do not. The platform's auto-labeling, designed for content carrying embedded AI metadata, also fails to fire. Neither the policy nor the infrastructure produces the label, and a six-second piece of synthetic political iconography circulates alongside vacation footage and dance clips with no infrastructural marker to distinguish them. This is invisible infrastructure in action, demonstrating that making infrastructure visible is not merely an artistic gesture. It is the minimum condition for accountability.
There are two registers in which this visibility can be built. The first is speculative: imagining what could be, even where it cannot yet be built — solar-powered servers that only stream when the sun is out, mesh networks that only work in physical proximity, collective archives with democratic governance over what persists, AI-labeling systems that are transparent, mandatory, and technically auditable rather than an optional field defaulting to zero. Speculation here is a form of infrastructural critique and a refusal of the demand for an immediate solution. The second register is working infrastructure: what actually exists, even if limited in reach — federated video hosting like PeerTube, whose failure to scale is itself worth studying; self-hosted livestreaming tools like Owncast; the historical precedent of pirate radio; the model of community-controlled public access television. Infrastructure, treated this way, is both medium and message: breakdowns become aesthetic events rather than embarrassments. When the server goes down, you see why, not just that an error occurred.
Maintenance, governance, and energy constraints become visible rather than hidden, because invisible infrastructure is unaccountable infrastructure, and we make ours visible because accountability is the point.
The Machine in Training
The sections above map the videosphere as it has been: captured by platforms, harvested by algorithms, labored by human creators, and made invisible by infrastructure. But Bösch's analysis of a further phenomenon, clusters of AI-generated political content on TikTok, names something qualitatively new, operating at the level of the system rather than the artifact.
Hito Steyerl's concept of the mean image is a place to start. The figure in an AI-generated video is not a person. The figure is a statistical type, a convergence of beauty-filter parameters, ethnic styling cues, platform color grading, possibly a composite or fully synthetic. Asking whether the figure is real is the wrong question, because she was never trying to be anyone in particular.
"Mean" accumulates several senses at once: shabby, cheap content aimed at a demographic the producers code as low-prestige; nasty, content aimed at lonely or isolated viewers; mean as in means, the off-screen workers who script the broken dialogue, run the text-to-speech, source or commission the footage; and mean as in statistical average, a figure assembled from the scraped visibility of real people whose labor is used as set dressing without consent.
Steyerl places these images in a long history. The production of types through statistical aggregation dates back to nineteenth-century photographic typologies, such as Galton's composite portraits, which were used to manufacture supposedly objective images of "criminal" or "racial" types. The technologies have changed. The operation has not.
Hito Steyerl, Mean Images, NLR 140/141, March-June 2023 - newleftreview.org
TikTok's interface displays one video at a time, so each clip looks like a curiosity, an outlier, a one-off. Stepping outside that logic reveals what the interface obscures: clusters of low-effort accounts with random-string usernames running similar AI generations through similar templates, converging on the same affective registers and the same political iconography, elderly figures in decent clothing, pouring rain, Gothic cathedrals, families with prams, marching crowds, lone figures in the foreground. The visual vocabulary of right-wing populism, rendered synthetic, is amplified by the same recommendation engine that fails to label it. The glitches in these videos are not merely AI tells. They are the visible residue of testing different prompts, parameters, audio overlays, and operators, producing similar variants of the same emotional package, with engagement metrics selecting which variants survive and propagate. What you are watching, scrolling through the cluster, is a system iterating. A machine in training.
Bösch calls this dynamic postdigital propaganda: not a campaign run by an identifiable actor with an identifiable message, but an alignment produced within the system itself. Platform infrastructure surfaces what fits the feed; platformized affect rewards what hits the feeling; participatory recursion across accounts that are not coordinating with each other but produce the same thing anyway; synthetic plausibility accumulating into a shared imaginary that does not require any individual viewer's belief to operate. The damage is not only to viewers who believe the content. Every exposure, belief, mockery, indifference, and critical analysis deposits the type. The category becomes more imaginable, not because anyone was persuaded, but because they saw it.
This is the hardest implication for any alternative infrastructure project. It is not simply competing with commercial platforms for attention. It is contending with a systemic condition in which the infrastructure itself, recommendation algorithms, engagement metrics, and invisible labeling actively shape the collective imaginary. The atmospheric claim of Section I of this essay is not hyperbole: we are breathing content that reshapes what feels normal, what counts as ordinary, and what types feel recognizable, while we, in turn, train the machines that will govern us.
Postdigital propaganda does not require intent. It requires only that platforms optimize for engagement without accounting for what engagement is selecting. Any serious response to this has to extend the demand for visible infrastructure and unharvestable material conditions to include a further refusal: infrastructure that does not amplify content purely on the basis of engagement signals, without any account of the cognitive and political consequences of what those signals select for.
That is a radical demand! It challenges the core business logic of every major platform. But it is the demand that Bösch's analysis makes unavoidable. The machine is in training. What it is learning to produce is already visible in the feed. The question is whether there is infrastructure that refuses to be its training data.
Six Dimensions of Infrastructure
To think infrastructurally means refusing the fantasy of a single fix and instead analyzing across several registers at once, each of which any alternative video infrastructure has to answer for.
The technical register asks which protocol carries the work (HTTP, IPFS, BitTorrent, ActivityPub, or something custom) and who controls it. Is it an open standard or a proprietary one? It asks whether the server architecture is centralized, federated, or peer-to-peer, and which codecs and bitrates are chosen, and by whom.
The economic register asks who pays for hosting: individual artists, a collective fund, a grant, advertising, and whether the answer can survive five years, let alone ten. It asks who maintains, moderates, and updates the code, because most alternatives collapse not from a lack of vision but from the maintenance trap. The work of upkeep is invisible, unpaid, and exhausting, and eventually, nobody is left to do it.
The legal register asks who owns the infrastructure, the content, and the data generated about that content; where the servers sit and which jurisdiction's laws apply; whether the infrastructure can be seized, and by whom; what licensing regime governs it, and whether it can be used to train someone else's model.
The energy register asks what actually powers the system: grid, solar, wind, and whether its footprint is genuinely sustainable or only aesthetically green. It raises energy constraint as a design principle in its own right. Infrastructure that only operates when renewable energy is available creates a condition that platforms cannot simulate, because they cannot fake "only works six hours when the sun is out."
The social register asks who governs. Is it a benevolent dictator, consensus, formal democracy, or no formal structure at all? The social register asks how trust is established without a central authority, noting that small communities can often self-govern in ways that break down as scale forces them into rules, moderation, and formal conflict resolution.
And the geographical register asks where the servers physically sit, with consequences for data sovereignty, latency, and jurisdiction; where the users are, given the gap between high-bandwidth regions and the low-bandwidth Global South; and confronts an uncomfortable fact underneath all the others, that no infrastructure can be fully autonomous while the Tier 1 internet service providers beneath it remain corporate.
None of these six registers can be resolved independently of the others. To get a different infrastructure requires different economic relations, like commons-based, cooperative, and collectively governed ones, because there is no purely technical solution to the social, economic, and legal problems that lie beneath the engineering.
The Stream Art Network
None of this is a solution. It is a working hypothesis, a set of principles to test, and a provocation to iterate on. The Stream Art Network is offered here as both a bet and an invitation.
The bet rests on one claim that aesthetics are symptoms of infrastructure. You cannot produce unharvestable aesthetics through design choices alone. You can only produce them through unharvestable material conditions.
Platforms absorbed lo-fi within months; they commodified VHS grain, film scratches, vertical-video glitch, as noted above. What they cannot absorb is degradation that arises from actual bandwidth limits, actual energy constraints, and actual peer-to-peer network conditions, because you cannot harvest an aesthetic without replicating the infrastructure that produces it. You cannot replicate that infrastructure without adopting economic and technical principles that, by definition, a platform cannot adopt without ceasing to be a platform. Real compression artifacts from low-bandwidth protocols, not Instagram's grain filter; peer-to-peer streams that degrade according to actual network conditions, not a preset.
Shared ownership and transparent, consensus-based governance, so that no single entity can sell out or shut the network down.
Energy constraints are treated as a political reality rather than a marginal virtue; permacomputing principles are applied to video, protocols that are visibly slowed to conserve power, solar-powered servers whose latency is visible rather than hidden, and not dark mode's marginal savings dressed as an aesthetic.
Open protocols and interoperability rather than enclosure: ActivityPub streams, Owncast, PeerTube, rather than another walled garden.
And, following from the argument of Section VI, ephemeral transmission that resists archiving and algorithmic parsing,unreplayable, untrainable, true peer-to-peer broadcast without central storage, because, as Dylan Smith's reading of Monad makes clear, an archive is a fixation, a collapse of the event back into a fixed instance, and ephemeral transmission refuses that collapse so that the work cannot be scraped, trained on, or reduced to its engagement metrics, because it does not persist in a form that would admit those operations.
Three demands follow from this:
- restore waiting, as argued in Section V of this essay;
- refuse the training loop, as argued in Section II;
- make infrastructure visible, as argued in Section VI.
That such a network is not entirely hypothetical is why Open Secret matters more than an inspiring example: it demonstrates several of these principles not as proposals but as existing conditions of practice, sustained across continents and years without either institutional capture or canonical consolidation.
What the Stream Art Network adds to that model is an explicit demand for visibility, showing the servers, the energy use, the costs, the labor, making breakdown informative rather than embarrassing, treating maintenance itself as a public practice, and, given the argument of Section VII, a seventh and non-negotiable principle: refusal of engagement-only amplification. No algorithmic amplification based purely on engagement signals; content governance that accounts for the cognitive and political effects of distribution, not just individual clicks; mandatory, auditable, technically enforced labeling of synthetic content, not an optional field defaulting to zero; recommendation logic that is visible and contestable rather than proprietary and opaque.
Andreas Treske speaking at INC Exit Fest | Day 2 - Session 2 on THE VOID
What remains genuinely open cannot be resolved by argument alone, and it would be dishonest to pretend otherwise. Whether staying small is the point or whether broader adoption is the goal changes what success looks like. Whether maintenance work can be sustained without institutional support, or whether it inevitably needs one, shapes what kind of organization this has to become — a cooperative, grant-funded commons, municipal infrastructure, or something without precedent. Whether any infrastructure can remain an alternative indefinitely, or whether all of them eventually centralize as Web 2.0's early alternatives did, is a question this essay cannot answer in advance; it can only be tested by building and watching what happens. Whether the audience is artists performing for each other or a broader public, the design changes at every level. And whether Stream Art Network practice can coexist with a continued presence on commercial platforms, or whether coexistence itself is a form of capture, is perhaps the sharpest open question of all. It is one that Dana Dawud's practice, moving between Open Secret and more visible platforms without apparent contradiction, suggests may not require the purist answer it seems to demand.
Against the Final Envelope
One position in this discourse deserves to be named directly and answered rather than sidestepped, because it is not a weak position. Fakewhale's diagnosis is the most thorough of any account gathered here — the videosphere, the dopamine loop, the molecular management of desire, the destruction of waiting, all of it stands. But Fakewhale's conclusion does not. Resistance, in that account, is configured as the ascetic exercise of refusal and silence; post-mediality is described as the condition in which we are immersed, a climactic mutation of the human intellect, from which the way out does not lie in an impossible return to pre-digital purity but in understanding the rules of the game in which we are prisoners. "A subject constantly entertained," Fakewhale writes, "is a subject who has lost the ability to imagine an alternative to what exists."[20]
This essay refuses that endpoint. It refuses not because the diagnosis is wrong, but because it mistakes understanding the rules for accepting them. Also, not through ascetic refusal of the medium, which would be irrelevance, but through the construction of alternative conditions within the medium. It refers not to silence, but to a different sound, and not to exit, but to a different infrastructure. It asks not for purism, but for practice. Rilke's imperative, read through Sloterdijk, is more precise here than any infrastructural argument alone can be. The videosphere sees you. There is no place it does not see you. Fakewhale's response to this is to understand the prison. This essay's response is the one the torso demands: you must change your life. Not escape it. Not describe it. Change it through practice, through askesis, through the discipline of conscious breathing in a medium designed to make you breathe without thinking.
Fakewhale describes the capture with real precision, then concludes that there are no keys. This essay insists that building the keys is the work. The videosphere is not total. It has been built, technically and economically, which means it can be built differently. The somatic condition of users, shaped by years of dopamine-loop design, is real and must be accounted for, but it is not irreversible. Attention can be rehabilitated. Waiting can be restored. The fertile void can be recovered. The Stream Art Network is a proposal for the infrastructure that makes that recovery possible, one screening, one protocol, and one maintained server at a time.
Coda: Maintenance Is the Practice
Twenty years of the streaming video conversation. The platforms got bigger. The capture deepened. The dopamine loops got tighter. The AI started generating the content. And yet: Open Secret is running. Internet cinema is proliferating. The lineage from Espinosa to Naficy to Dawud is unbroken. The machine is in training, but so are we.
The question was never whether alternative infrastructure is possible. It is whether we will build it. Whether we will maintain it. Whether we will treat the boring, unglamorous, years-long work of keeping different conditions of dwelling alive as the actual practice, not the manifesto, not the theory, not the talk. Maintenance is not the prerequisite to the practice. Maintenance is the practice. That is the askesis.
We are not watching the videosphere. We are dwelling inside it — blindly, by default. The only question is whether we build the conditions for conscious dwelling, or keep breathing until there is nothing left to breathe.
Who's in?
(reworked essay, first presented from at the INC EXIT Conference, Amsterdam, June 24, 2026)
References
Afif, Hadi. "Cross-Sectioning the Slop: Notes on Internet Cinema." Do Not Research (Substack), 10 May 2026. Originally published in Dazed Middle East. https://donotresearch.substack.com/p/cross-sectioning-the-slop
Bösch, Marcus. "Variations of Slop: How AI Content Is Reshaping the Feed." Understanding TikTok (Substack), 20 May 2026. https://tiktoktiktoktiktok.substack.com/p/variations-of-slop
Bösch, Marcus, and Tom Divon. Dancing in uniform: Thirst trap propaganda in military image wars on TikTok.February 2026. Platforms & Society 3. DOI:10.1177/29768624251403864. https://journals.sagepub.com/doi/full/10.1177/29768624251403864
Dawud, Dana. Palcorecore (2023), https://donotresearch.substack.com/p/dana-dawud-palcorecore; Monad+ (2024–ongoing). Open Secret program.
De D, El Jamal M, Aydemir E, Khera A. Social Media Algorithms and Teen Addiction: Neurophysiological Impact and Ethical Considerations. Cureus. 2025 Jan 8;17(1):e77145. doi: 10.7759/cureus. 77145. PMID: 39925596; PMCID: PMC11804976. https://pmc.ncbi.nlm.nih.gov/articles/PMC11804976/
Espinosa, Juan García. "For an Imperfect Cinema." 1969. Jump Cut, no. 20, 1979, pp. 24-26. https://www.ejumpcut.org/archive/onlinessays/JC20folder/ImperfectCinema.html
Fakewhale. "Post-Media Logics and Conditions, from TV to Social Networks: The Evolution and Engineering of Entertainment." Fakewhale Core, 3 May 2026. https://log.fakewhale.xyz/post-media-logics-and-conditions-from-tv-to-social-networks-the-evolution-and-engineering-of-entertainment/
Guénon, René. Introduction to the Study of the Hindu Doctrines. 1921.
Haidt, Jonathan, and Zach Rausch. The Anxious Generation. Penguin, 2024, with updated analysis in World Happiness Report 2026.
Jiang, Y., et al. "Losing Track of Time on TikTok? An Experimental Study of Short Video Users' Time Distortion." Behavioral Sciences 15.7 (2025): 930.
Manovich, Lev. "Database as a Symbolic Form." 1999.
Naficy, Hamid. "Internet Cinema: A Cinema of Embodied Protest." 2021. https://cinema.iranicaonline.org/article/internet-cinema-a-cinema-of-embodied-protest/
Onty. "The Fourth Secret of Internet Cinema." Ultra / Open Secret Zine. https://www.ultra.art/p/the-fourth-secret-of-internet-cinema
Parikka, Jussi.Operational Images: From the Visual to the Invisual. University of Minnesota Press, 2023. http://www.jstor.org/stable/10.5749/j.ctv2z86232.
PeerTube, Owncast, ActivityPub documentation. https://docs.joinpeertube.org/api/activitypub
Permacomputing principles and low-tech design methodologies. https://permacomputing.net/principles/
Rosenberg, Harold. "The American Action Painters." 1952.
Sharpe BT, Spooner RA. Dopamine-scrolling: a modern public health challenge requiring urgent attention. Perspect Public Health. 2025 Jul;145(4):190-191. doi: 10.1177/17579139251331914. Epub 2025 Apr 12. PMID: 40219910; PMCID: PMC12322333. https://pmc.ncbi.nlm.nih.gov/articles/PMC12322333/
Sloterdijk, Peter. You Must Change Your Life. Trans. Wieland Hoban. Cambridge: Polity, 2013.
Smith, Dylan. "On Internet and Internet Cinema." Heaven Interface (Substack), 23 May 2026. https://heaveninterface.substack.com/p/on-internet-and-internet-cinema
Steyerl, Hito. On mean images and statistical typology, 2023. https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images
Toister, Y., & Zylinska, J. (2025). Image thinking after artificial intelligence.Journal of Visual Culture,24(3), 431-451. https://journals.sagepub.com/doi/10.1177/14704129251405715
Tryon, Chuck. "New Media Studies and the New Internet Cinema." 2007. https://quod.lib.umich.edu/p/postid/pid9999.0005.102/--new-media-studies-and-the-new-internet-cinema?rgn=main;view=fulltext
Zuboff, Shoshana. The Age of Surveillance Capitalism. Harvard Business School / PublicAffairs, 2019, with updated analyses 2025–2026.
--
Notes
[1]Dana Dawud, in the zine accompanying the Open Secret program, 2026.
[2]Chuck Tryon, 'New Media Studies and the New Internet Cinema', 2007, as discussed in Hadi Afif, 'Cross-Sectioning the Slop: Notes on Internet Cinema', Do Not Research (Substack), 10 May 2026, originally published in Dazed Middle East.
[3]Dylan Smith, 'On Internet and Internet Cinema', Heaven Interface (Substack), 23 May 2026.
[4]Fakewhale, 'Post-Media Logics and Conditions, from TV to Social Networks: The Evolution and Engineering of Entertainment', Fakewhale Core, 3 May 2026.
[5]De, D., 'Social Media Algorithms and Teen Addiction: Neurophysiological Impact', Cureus/PMC, 2025; B. T. Sharpe et al., 'Dopamine-Scrolling: A Modern Public Health Challenge', Perspectives in Public Health, 2025; Y. Jiang et al., 'Losing Track of Time on TikTok? An Experimental Study of Short Video Users' Time Distortion', Behavioral Sciences 15.7 (2025): 930.
[6]Rainer Maria Rilke, 'Archaic Torso of Apollo', as read in Peter Sloterdijk, You Must Change Your Life, trans. Wieland Hoban (Cambridge: Polity, 2013).
[7]Fakewhale, 'Post-Media Logics and Conditions'.
[8]Fakewhale, 'Post-Media Logics and Conditions'.
[9]Roland Meyer, building on a term from Jacob Birken, cited in Marcus Bösch, 'Variations of Slop: How AI Content Is Reshaping the Feed', Understanding TikTok (Substack), 20 May 2026.
[10]Yanai Toister and Joanna Zylinska, on cognitive hacking, 2025.
[11]Dylan Smith, 'On Internet and Internet Cinema'.
[12]Harold Rosenberg, 'The American Action Painters', 1952.
[13]Hadi Afif, 'Cross-Sectioning the Slop'.
[14]Dylan Smith, 'On Internet and Internet Cinema', on Dana Dawud's Monad.
[15]Hadi Afif, 'Cross-Sectioning the Slop'.
[16]Fakewhale, 'Post-Media Logics and Conditions'. https://log.fakewhale.xyz/post-media-logics-and-conditions-from-tv-to-social-networks-the-evolution-and-engineering-of-entertainment/
[17]Fakewhale, 'Post-Media Logics and Conditions'.
[18]Jussi Parikka, on in-visuality and image infrastructure.
[19]Marcus Bösch, 'Variations of Slop'.
[20]Fakewhale, 'Post-Media Logics and Conditions'.