Making Images Between History and Algorithm

I spend much of my time surrounded by machines. A small server built from recycled GPUs hums in my studio, a laptop runs multiple AI models, and external drives store datasets I have collected. These machines are integral to my process and act as recorders of its outputs. I fine-tune models to generate images, experiment with automated composition, and observe how outputs circulate when removed from corporate platforms. Each act is simultaneously creative, investigative, and infrastructural - producing, observing, and negotiating with the systems that shape our visual world.

The initial promise of digital tools felt liberating. Compressed videos, low-resolution memes, and AI-generated artworks all suggested a democratisation of visibility. Yet circulation alone is not liberation. The platforms that host our images extract value at every stage through engagement metrics and algorithmic curation. Even the ‘poor image’, celebrated for its rapid, viral mobility, participates in these economies. Visibility itself has become a resource, controlled by infrastructures far beyond individual creators.

In this context, it has become easier to imagine the end of the world than the end of capitalism itself - what Mark Fisher termed ‘capitalist realism’. We are told repeatedly that there is no alternative, that the present arrangement of economic and social relations is the only viable reality. This ideological closure extends into the visual realm. The images that circulate most widely, train our algorithms, and shape our collective imagination are those produced within and for capitalist social relations. Minor, local, and unwritten histories become harder to visualise when the infrastructures of seeing are optimised for other purposes.

This raises a fundamental question: what forms of practice remain possible within systems designed for extraction? My work attempts to answer this not through withdrawal, but through deliberate engagement with constraint - economic, technical, and infrastructural. I recognise from the outset, however, that my ability to frame limitation as creative opportunity is itself a position of privilege. Access to salvaged hardware, technical knowledge, and the time for experimentation are not universally available. The same constraints that enable my practice could represent insurmountable barriers for others.

Working with AI image generation presents a specific problem for someone concerned with local Danish histories: dominant models either fail or produce distorted outputs when prompted for recognisably Danish scenes. Trained primarily on American and Chinese internet imagery, they default to generic results. The particular light of Nordic winters, the architectural vernacular of social housing, the visual culture of local political organising - the specific textures of Danish social life - are lost.

This is more than an aesthetic problem; it is a political one. It raises the question of whose visual cultures get encoded into the infrastructures that increasingly shape how we see and remember. When systems trained on billions of images from dominant cultures become our primary tools for generating new ones, they reproduce a particular visual hegemony. The minor, local, and unwritten histories become progressively harder to visualise as these systems are optimised to reproduce patterns prevalent in their training data.

The choice to write this text in English exemplifies the tensions inherent in working within dominant infrastructures. English functions as the lingua franca of global discourse, much as corporate platforms function as the default spaces for digital circulation. For a Danish artist concerned with local specificity, writing in English means participating in a linguistic infrastructure that both enables international communication and risks flattening the very particularities the work seeks to preserve. This mirrors the position of delivery drivers who must use platform-controlled smartphones to work, or activists who use corporate social media to organise: we work with tools we do not necessarily endorse because they are the dominant material conditions of our time. The alternative - linguistic or technological isolation - would replicate the individualist logic that neoliberalism promotes, severing us from the collective and interdependent nature of infrastructure itself. Acknowledging these dependencies is not capitulation but a precondition for solidarity: shared resistance begins not with impossible demands for purity but with clear-eyed recognition of our entanglements.

My work Frihed, lighed og Hip-hop (Freedom, Equality and Hip-hop) emerged from turning this technical limitation into a generative constraint. The project asks: what if Danish labour movements and 1990s hip-hop culture had intersected in ways they never quite did? What if local politicians had genuinely engaged with youth culture rather than merely dismissing it? What if collective political action had taken forms outside the narrow channels documented in official archives?

These are not idle counterfactuals. They are engagements with minor histories - the oral accounts of what almost happened, what some people remember happening, what felt possible at the time even if it never materialised. These are what I term ‘real lies’, narratives that are part fact, part speculation, but always grounded in actual persons, places, events, or possibilities that never came about. Each image is based on meticulous research - oral histories, material archives, photographs, personal accounts - but visualises something not captured in the official record. This is the true lie: all the images contain factual elements, they are just not the facts that official history chose to preserve.

In this sense, these speculative images do not function as traditional agitprop - which typically delivers clear political messages through unambiguous imagery - but rather adapt agitprop strategies to contemporary conditions. I am attempting something closer to system-critical Chinese memes: content that flies under the censorship radar without losing alterity and critique, that maintains political edge whilst remaining visually plausible enough to circulate. The images must be ambiguous enough to avoid immediate dismissal yet pointed enough to provoke political questions about what was possible and what might yet be possible.

Generating these images required months of work that remains invisible in the final outputs. I spent countless hours in local archives, photographing newspaper clippings, municipal documents, and street photography from the 1990s. I conducted interviews with former activists, hip-hop artists, and community organisers, collecting their stories about what happened and what did not but might have. My method draws from anthropology from below, working primarily with oral and material history rather than official documentation.

This material becomes part of the training data, though not straightforwardly. Standard AI models do not understand ‘Danish 1990s social housing aesthetic’ or ‘the specific way light looks in Aarhus in November.’ The models require teaching through careful curation, iterative training, and constant adjustment. Each generated image requires dozens of attempts, parameter adjustments, and prompt refinements to coax the algorithm towards something that feels true to the place and time I am exploring.

I run this work on local infrastructure - a server in my studio whose heat warms the space during winter months, allowing me to wear fewer clothes and making the energy expenditure feel visceral and immediate rather than abstracted into distant data centres. The studio floods with rainbow-coloured LEDs from the gaming graphics cards, pulsing in sync with image generation, a visual rhythm of computational labour. I power the system using wind energy purchased from a supplier, though I remain dependent on the electrical grid and its fossil fuel components. The server’s physical presence - its noise, its heat, its light show, its occasional failures - keeps the material conditions of image production visible rather than hidden behind seamless cloud interfaces.

Running models locally also allows me to alter datasets and bypass the guardrails set up according to Chinese or American cultural sensibilities. Corporate AI systems encode the moral and aesthetic standards of their origin cultures, restricting what can be generated in ways that often have nothing to do with actual harm prevention. For instance, many models prohibit the word ‘dirty’ because of American ideas about sexual morality, making it nearly impossible to generate realistic images of working-class life, manual labour, or urban grit - anything that is not pristine and sanitised. Historical labour imagery often features sweat, soil, physical exertion - the material reality of work - yet these elements trigger content filters designed for very different concerns. Local control means I can work with material that corporate systems would flag or sanitise, allowing me to generate images that reflect the actual texture of Danish working-class history rather than an algorithmically cleaned version acceptable to American content moderators.

The resulting imperfections - the uncanny qualities and artefacts of algorithmic processing - are part of the work’s meaning. When a generated hand melts slightly into a turntable, or when architectural details do not quite cohere, these glitches signal that we are looking at something constructed. They are visual markers that say: this is not documentary evidence but speculative reconstruction, a minor history made visible through negotiation with algorithmic systems.

This is especially apparent when generating human figures. Outputs that deviate from the normative proportions encoded in training data - often reflecting narrow Western beauty standards - are typically dismissed as glitches or failures. These ‘other bodies’, with extra fingers or shifting features, can be read as inadvertent critiques of the homogeneous data on which the system was trained. However, I must acknowledge that whilst these algorithmic failures reveal the narrowness of training data, they operate differently from the actual exclusion and misrepresentation that disabled people and others with non-normative bodies experience in AI systems. The former are speculative glitches in imagined histories; the latter are failures to represent real, existing people with dignity and accuracy. Both reveal problems with training data, but confusing them risks treating actual exclusion as merely an aesthetic concern.

The term ‘AI slop’ is often used to describe a vast, differentiated mass of content, from aesthetically anaesthetising videos to potent political memes. Whilst much of this material is designed for frictionless consumption, I embrace the ‘sloppy’ or imperfect aspects of my own outputs for different reasons: they make the constructedness of the image visible whilst still producing something with emotional and historical resonance. The goal is not photorealistic deception but plausible speculation - images that feel like they could document real events whilst remaining visibly mediated. The vagueness of these images, clearly marked as AI-generated in accompanying text, positions them as small fictions based on meticulous research. This clarity about their status as constructions is essential: they are provocations, not forgeries.

Aarhus’er #1 extends this practice into collaborative territory. Working with local hip-hop artists and community members, the project creates space for collective memory-work around neighbourhoods, youth culture, and social change. Participants contribute stories, photographs from personal archives, and memories of spaces and moments that official history ignores or simplifies.

A legal graffiti wall functions as both material site and living archive. Residents inscribe their own narratives - tags, pieces, political slogans, memorial writing. The wall becomes a record of minor histories, constantly changing as new layers cover old ones. Graffiti is inherently temporary, an appropriation of inhabited space that asserts presence and claims visibility. Yet the local styles that once defined Danish graffiti culture have been largely washed away by the global homogeneity of US-centric hip-hop aesthetics - another instance of how dominant visual cultures colonise local expression. The wall embodies the logic of remix over the myth of originality, a constantly evolving canvas of layered voices where each mark is not a final creation but a version, a reinterpretation that insists on visibility whilst refusing permanence.

The images I produce enter a contested visual terrain where AI-generated content is already being weaponised for various ends. In Denmark, we have seen conspiracy theorists and COVID sceptics use AI imagery to spread misinformation. Crypto scams deploy deepfakes of politicians and famous actors to lend credibility to fraudulent schemes. These uses demonstrate that AI image generation is not a neutral technology but a site of active political struggle over collective memory and visual truth.

More insidiously, dominant political narratives use selective historical memory to serve present agendas. The Social Democratic Party, for instance, misrepresents the story of the early Danish labour movement. At the Workers Museum in Copenhagen hangs a famous banner bearing the slogan ‘8 hours of work, 8 hours of freedom and 8 hours of rest.’ Yet contemporary social democratic politics seems to have forgotten the freedom part - the historical demand for time outside both labour and biological necessity, time for culture, politics, and collective life.

Traditionally, the labour movement featured dancing, sports, collective reading, theatre, and communal celebration. These were not diversions from political struggle but integral to it - expressions of what freedom might look like, practices that built solidarity through joy and shared culture. Now the focus is solely on productivity over pleasure, on work conditions rather than life beyond work. Here there is something to be learned from hip-hop’s joyous dancing criticality: the insistence that resistance includes celebration, that politics can be embodied in movement and rhythm, that collective struggle need not be grim. Just as the collective struggle of labour movements might well inspire hip-hop - and already has at the margins - so too might hip-hop’s cultural practice revitalise our understanding of what labour struggle was and could be again.

The labour movement’s vision included not just fair wages and rest but genuine liberation from the tyranny of work. This historical amnesia serves a political purpose: by reducing labour struggle to demands for decent working conditions alone, it renders invisible the more radical imagination of what work and life might be. My images attempt to visualise this forgotten freedom - what it looked like when political organising included dance halls and reading circles, when collective life was about building alternative worlds, not just negotiating better terms within existing ones.

In this context, ceding the territory - refusing to engage with AI tools on political or aesthetic grounds - means allowing others to dominate the visual battlefield. Right-wing movements have proven adept at using these technologies for propaganda. Commercial interests flood platforms with AI content optimised for engagement and profit. I know of no one else doing precisely this kind of work, though it seems an obvious move, so there are probably others I am unaware of - or at least I hope so. I choose to work with technologies I do not particularly like because the alternative is to surrender the capacity to visualise collective political possibilities, local histories, and minor narratives to those who would use the same tools for very different ends.

My work attempts to reclaim AI image generation for different purposes: not to deceive or profit, but to visualise what official histories exclude, to make visible the radical possibilities that dominant narratives have forgotten, and to insist that local specificities matter against algorithmic homogenisation. I hope to create resources for new movements and maintain practices until they again become relevant, keeping alive ways of seeing and remembering that capitalist realism would prefer we forget.

This practice does not exist outside platform capitalism. I use commercial hardware and rely on corporate electricity grids even when purchasing renewable energy from suppliers. The server runs on components manufactured by the same companies dominating AI development. Complete autonomy is neither possible nor desirable - it would mean isolation from broader struggles and communities, replicating the individualist logic that neoliberal ideology promotes. Infrastructure is inherently collective and interdependent - we cannot build alternatives alone, only together and in negotiation with existing systems.

What is achievable is partial autonomy: specific spaces where different logics can operate, where decisions reflect local needs and political commitments rather than corporate optimisation, and where the material conditions of production remain visible and accountable. Running models locally rather than through corporate APIs means I control the data, bypass culturally inappropriate guardrails, and keep the computational labour physically present in my studio. This partial autonomy is meaningful even when incomplete.

Ultimately, this work argues that images are not just representations but interventions in how we remember, imagine, and understand historical possibility. Against capitalist realism’s insistence that no alternatives exist, these images visualise what collective political action might have looked like, demonstrating that local histories matter against globalised homogeneity and that minor narratives deserve visibility even when dominant systems make them difficult to produce.

The images from Frihed, lighed og Hip-hop do not document what happened. They visualise what could have happened, what feels true to the spirit of a time and place, what some people remember almost happening. They are real lies that carry emotional and historical truth whilst remaining visibly constructed, marked by the algorithmic struggle to make them appear at all. Each is grounded in research - an actual person, place, event, or possibility - but imagines how things might have been if the labour movement’s vision of freedom had not been forgotten, if hip-hop’s joyous criticality had met socialist organising, if collective life had centred pleasure and culture alongside struggle.

This labour of making local images visible becomes part of the work’s meaning, demonstrating that minor histories require active reconstruction rather than simple retrieval. They exist in tension with systems optimised for other purposes, and their visualisation demands meticulous research, technical negotiation, and patient adjustment of algorithms that would rather show us something else - or would rather show us nothing at all if it cannot be clean, sanitised, and acceptable to content moderators who have never heard of Danish labour history.

By running this work on local infrastructure - where the server’s heat warms my studio, where rainbow LEDs pulse with computational rhythm, where energy consumption remains visceral - I attempt to keep the material conditions of image production visible rather than abstracted. By working with oral and material history from below, by basing each image on actual persons and events whilst imagining their unrealised possibilities, by embracing imperfection as a marker of constructedness, by clearly marking images as AI-generated fictions based on research, and by refusing to cede visual territory to commercial or reactionary interests, I maintain that other practices remain possible within and against platform capitalism.

The results are modest, imperfect, and rooted in specific localities. They do not challenge platform capitalism’s scale or overturn AI’s corporate ownership. But they demonstrate that local and minor histories can be visualised even when algorithms resist them, that forgotten radical visions like the labour movement’s demand for freedom and joy can be remembered differently, and that spaces - however small and temporary - can be created where we might see beyond what dominant systems show us.

In an age where capitalist realism forecloses imagination of alternatives, where it is easier to envision apocalypse than systemic change, this work insists on the value of other possibilities. The images require a suspension of disbelief, not to deceive but to open space for alternative imaginaries. They are a refusal of the idea that what exists is all that is possible, a small attempt to visualise worlds where collective political action shaped history differently, where the labour movement’s vision of freedom and pleasure was not forgotten, where hip-hop’s joyous criticality met socialist organising, where local cultures resisted homogenisation, where minor narratives retained visibility.

Perhaps that is enough: working inside systems we cannot fully escape, making images that matter to specific people and places, creating resources for movements that may emerge, maintaining practices until they again become relevant, keeping alive ways of seeing and remembering that challenge what dominant systems would have us accept as the only reality. The work is provisional, partial, and situated - but it demonstrates that even within capitalist realism’s foreclosure, spaces remain where we might imagine and visualise differently, where real lies based on meticulous research can make visible what official history excludes, where the past’s unrealised possibilities can provoke questions about the present’s seemingly inevitable arrangements.

#stuffiwonderabout #tingjegspørgermigselvom

Kristoffer ørum @Oerum