1. Prologue: on watching al-Aqsa burn before the fire has been lit…

In September 2024, during my routine scroll through social media – an unsettling mix of global catastrophe updates, friends’ pictures and weekend plans – I stumbled upon a video that instantly caught my eye. It showed the al-Aqsa Mosque compound in Jerusalem engulfed in flames (Middle East Monitor 2024). Given the volatile context following the October 7, 2023, Hamas’ “al-Aqsa Flood” terror attack on IsraelFootnote ¹ and the consequent genocidal war on Gaza initiated by Netanyahu’s government, the idea of an extremist-led retaliation on Islam’s third holiest site didn’t seem far-fetched. For a few long minutes, I did believe it. I played the video several times before realizing what should have been obvious since the get go: the short clip was AI-generated. The synthetic video had been circulated by extremist channels and was quickly debunked by journalists, activists and news outlets. And yet, the imagery lingered, unsettling, plausible and haunting. What if? That single question pulled me deeper into a world of synthetic horror.

I began encountering more of these videos: some showed al-Aqsa burning; others went further, depicting the construction of the Third Temple on the mosque’s historical site – referred to as al-Haram al-Sharif (The Noble Sanctuary) or Temple Mount. The Third Temple refers to a holy temple whose hypothetical construction should succeed the First Temple and Second Temple, both destroyed in ancient times. The idea of building the Temple holds deep theological and eschatological significance in Judaism, especially within Orthodox traditions, where it is regarded as the holiest place of worship and a key event in the advent of the Messianic Age.Footnote ²

The destruction of al-Aqsa and its replacement with the Jewish holy temple may appear as fantasies to those unfamiliar with the long-contested history of Jerusalem – a city that stands both as symbolic and material flashpoint in the Israeli–Palestinian conflict. Yet these visions did not emerge from nowhere, nor are they inventions of the generative AI age.Footnote ³ They echo a much older struggle, one that has unfolded since Israel’s occupation of East Jerusalem in 1967. At that time, the Israeli authorities agreed to preserve the Ottoman-era status quo – an arrangement granting Muslims exclusive prayer rights within the holy compound. But from that moment onward, numerous religious and political actors have sought to subvert or overturn this agreement: some by attempting to pray or perform Talmudic rituals within the restricted area or carrying out symbolic animal sacrifices; and others by plotting far more violent interventions to demolish al-Aqsa and prepare the ground for the construction of the Third Temple.Footnote ⁴

In the aftermath of October 7, these radical ideologies have moved from the underground and the periphery to the center of political power in Israel. With the ascent of the far-right parties Jewish Power and Religious Zionism, represented in Benjamin Netanyahu’s government by National Security Minister Itamar Ben-Gvir and Finance Minister Bezalel Smotrich, ideas once confined to the fringe have become institutionalized. Praying on the Temple Mount – once a taboo even among Jewish Orthodox authorities as not officially endorsed by the Chief Rabbinate – is now publicly not only defended but actually sought by those who occupy institutional roles in the government, like Ben-Gvir himself, who has repeatedly performed Talmudic rituals on the holy site and defiantly called for the reconstruction of the Temple atop the al-Aqsa compound.Footnote ⁵

Under the influence of such institutional figures, what once seemed taboo extremist theology now openly circulates as policy discourse. As Yishai Sarid, author of The Third Temple – a speculative novel set in a future where Israel has become an isolated kingdom devoted to temple worshipping and sacrifices – observes in the afterword dated August 2024: “In 2015, when the book was published in Israel, some readers and critics saw it as science fiction, an imaginary dystopia.” Yet this is no longer the case, he notices, as “hardline fundamentalist groups, which have accumulated much political power, believe that Israel should be governed by the laws of the Torah (…).” “The focal point of this vision is the rebuilding of the temple and the renewal of sacrificial rites,” Sarid makes clear (Sarid, Reference Sarid and Greenspan2024, 305–6).

Not only in Israel, but across the globe, the claims of the messianic right have gained unprecedented visibility – and legitimacy – amplified by Donald Trump’s return to power for a second term. His administration, firmly backed by Christian nationalists and staffed with hawkish figures from the reformed evangelical sphere, has brought these once-marginal beliefs into the corridors of state. Among them is Pete Hegseth (now US Secretary of War, the newly renamed Department of Defense) who, as early as 2018 while still a Fox News host, openly endorsed Jewish control over the Temple Mount, proclaiming: “There’s no reason why the miracle of the re-establishment of the Temple is not possible” (Hegseth, quoted in Israel National News 2024).

Though the Trump administration has not officially embraced these positions, the presence of such sympathies at the highest levels of government signals a profound ideological shift. Even within the conservative mainstream, this turn has sparked unease. In an October 2025 episode of his show, political commentator Tucker Carlson engaged in a discussion with economist Jeffrey Sachs, voicing concern that Israel’s current political fervor might culminate in the destruction of al-Aqsa to make way for the Third Temple.Footnote ⁶

Within this fevered political climate – both at a local and global level – watching AI-generated videos of al-Aqsa burning, and the Temple rising in its place, becomes an experience at once hallucinatory and disturbingly plausible. The synthetic images flooding my feed no longer feel detached from reality; they seem to prefigure it. Since the violence of October 7, both material and symbolic destruction have accelerated, and these generated images – accompanied by captions, hashtags and endless comment streams – have collided with real-time footage, such as Ben-Gvir’s repeated incursions into the holy compound, the performance of Talmudic rituals and the settlers’ offerings upon the sacred ground.Footnote ⁷ In this convergence, the synthetic and the documented have fused into a single continuum of representation – a feedback loop in which anticipation and action constantly reinforce one another. The result is a visual ecosystem where computation and documentation have melted together so seamlessly that one sustains the other.

It is within this ecosystem that I, too, was deceived by the spectacle of al-Aqsa in flames. It seemed entirely plausible: that a fire had been lit, that someone had filmed it and that the clip had widely spread across networks. Generative AI had merely given final form – visual, tangible, seductive – to ideas that had long smoldered in the political underground. Now, clad in the photorealistic polish of the machine, those ideas had entered the mainstream, quietly redrawing the boundaries of what the eye, and the conscience, could endure.

2. A new politics of vision

This essay traces the emergence of a new politics of vision born from synthetic imagery – images like the burning of al-Aqsa or the serene rising of the Third Temple. Generative technologies are giving shape to a novel “regime of the sensible” (Rancière Reference Rancière2013), one that can enable subtle yet devastating forms of violence: violence that operates through conjecture and projection, through the symbolic manipulation of the future by means of visual anticipation. I refer to these emerging visual – and political – arrangements as instances of “speculative violence” (Della Ratta Reference Della Ratta2025), a mode of harm that neither bleeds nor screams, that offers no explicit horror, yet embeds destruction within the smooth surfaces of the machine’s imagination. In the luminous calm of these synthetic visions, such as the temple ascending in divine symmetry, rising as the counter shot to al-Aqsa’s wrapped in flames, violence persists in disguise. Beneath the radiance of perfection lies the shadow of erasure, the quiet certainty of annihilation. It is a violence of the aftermath, enacted through the glossy, frictionless aesthetic of synthetic imagery, performed in the seemingly innocent guise of speculative imagination.

I argue that by giving these possible futures a synthetic form before they occur, generative AI invites their normalization. The acts of displacement, destruction and disappearance that underwrite these projections become aesthetically naturalized in the dominant language of today’s media ecosystem, absorbed into the collective imagination before they even unfold. Through repetition, plausibility, and a photorealistic aesthetics, AI strips provocative and inflammatory ideas – such as the replacement of al-Aqsa with the Third Temple – of their political shock by ways of visualization. Speculative visuals may not depict violence outright, yet the worlds they conjure quietly imply the disappearance of others, embedding erasure into the very fabric of what they make visible.

A remarkable example of this logic in action is the AI-generated video known as Gaza Riviera, produced by two Israeli creators based in Los Angeles and catapulted to global fame after US President Donald Trump shared it across his social media on February 26, 2025 (Hall Reference Hall2025). The synthetic imagery envisions Gaza as a luxury coastal resort: the US President and Israeli Prime Minister Benjamin Netanyahu recline on chaise longues, cocktails in hand, surveying a gleaming skyline of high-rises and yachts. In this fantasy, peace becomes synonymous with property, commerce replaces coexistence, and Palestinian life is nowhere to be seen. Turning Gaza into the “Riviera of the Middle East” would, in fact, demand the total erasure or displacement of its people: a vision of prosperity built on deportation and ethnic cleansing.

The video – described by commentators as grotesque, kitsch and surrealFootnote ⁸ – nonetheless went viral. While being labeled an AI “fantasy” (Suciu Reference Suciu2025), in fact, it translated into an image what Trump had said only weeks earlier. On February 4, 2025, standing beside Netanyahu at a joint press conference, the US President had promised to “take over” and “own” Gaza, pledging to transform the war-torn territory into “something phenomenal,” a Mediterranean Riviera of luxury, jobs and stability for all, save for the Palestinians themselves, who, he suggested, could be “relocated to a beautiful area where they can be happy” (Al Jazeera 2025).

Trump’s remarks sparked immediate outrage (Lanard Reference Lanard2025; NBC News 2025; The New York Times 2025). Across the world, critics denounced the proposal as little more than a rebranded blueprint for ethnic cleansing, an annihilation plan disguised as an urban development scheme. Yet only weeks after condemnation, the Gaza Riviera made its global appearance: an AI-generated visualization of Trump’s words aestheticized and made to circulate as if the dream was already under construction. What had sounded, when merely spoken, like provocation or trolling in Trump’s signature style, was now given visual form, rendered plausible, inevitable, maybe even desirable – already on its way to becoming real (Zupančič Reference Zupančič2025).

To close the circle, the AI visuals became the synthetic anticipation of what would be concrete policy plans being proposed regarding Gaza’s “future.” Among them was the G.R.E.A.T. initiative (Gaza Reconstruction and Economic Acceleration Trust), a title that folded neatly into the US president’s obsession for “greatness,” treating it as both a political slogan and a universal brand. The plan envisioned a radical reconfiguration of Gaza in its “spatial design, economy, and governance” (The Washington Post 2025), positioning the Strip as a new node within the US-brokered Abraham Accords – a bridge of capital and commerce linking Israel to its Arab partners, finally engaging them in peaceful neighborly relations. Its rhetoric of renewal and rebirth echoed perfectly the language of Netanyahu’s Gaza 2035 vision, released by the Israeli Prime Minister’s Office in May 2024.Footnote ⁹ In that project, Gaza appeared entirely purged of its people and culture, which was rendered aesthetically by sanitized figures in the gleaming white thobes of the Gulf region, strolling across immaculate promenades. It was hardly a coincidence that the very same renderings released under the Gaza 2035 label resurfaced to visually illustrate the G.R.E.A.T. initiative. The images migrated seamlessly from one plan to another, from one blueprint to the next, changing names and labels yet remaining identical in substance, with each iteration reaffirming the exclusion of any Palestinian agency in envisioning the future of their own land.

When, in October 2025, Trump unveiled his 20-Point Peace Plan,Footnote ¹⁰ a rebranded, apparently more sober – at least in the name – version of G.R.E.A.T., the aesthetic and ideological continuities were unmistakable. Despite the new name, the plan recycled the same underlying logic: a complete erasure of any Palestinian sovereignty, or even mere contribution, over the country’s future, and a rebranding of “peace” as a business paradigm of economic growth and stability stripped of any social or political justice. What Gaza Riviera achieved visually, the 20-Point Plan proposed institutionally. The video aestheticized the future; the policy sought to implement it. With the plan’s enthusiastic endorsement by a wide coalition of nations – including several Islamic statesFootnote ¹¹ – the circuit initiated by the AI-generated imagery reached closure. What had begun as speculative fantasy had completed its transformation into an administrative blueprint: the imagination of erasure had become its instrument. A continuum fusing the synthetic imaginary and the policy blueprint had been established.

Speculative violence unfolds precisely in this interval between image and implementation, between synthetic frames and policy. It is the instrumentalization of imagination itself, the conversion of aesthetic projection into political groundwork. Here, harm operates not through direct assault but through normalization: erasure naturalized, annihilation aestheticized, displacement rendered desirable. Speculative violence is the quiet harm that prepares the ground, which makes the future governable by first making it visible. Generative AI offers a seamless and widely accessible means of rendering such desires and beliefs into legible images, coded in familiar aesthetic registers. It does not itself generate these genocidal futures, yet it removes the final barrier to imagining them: almost anything can now be visualized, frictionlessly and on demand.

Focusing on synthetic imagery produced in the context of Israel’s genocidal war on Gaza, I aim to illuminate and further reflect on the dynamic enacted by speculative violence. However, this is by no means something confined to war zones: the new politics of vision that generative AI enables extends far beyond the borders of the Middle East. Amid the resurgence of the far right across Europe and the consolidation of a second Trump administration in the United States, generative AI has become an engine for imagining, visualizing and operationalizing violent, exclusionary and supremacist futures within Western political and cultural spheres (Della Ratta, Reference Della RattaForthcoming).

My argument is not that generative AI, as a technology, is inherently destined to aestheticize domination or reproduce racist, supremacist and genocidal logics. Rather, it is that certain political actors and movements possess a powerful imagination of the future, one that envisions renewal only through annihilation, redemption through destruction and purification through erasure. Generative AI provides the perfect visual apparatus for this ideological horizon, translating latent fantasies of domination into glossy simulations of inevitability. It gives form to what already exists as an ideological project, rendering it visible, legible and thus ready to be operationalized in the present.

I am interested in tracing how the visualization of imagined futures exerts real effects in the now: how the framing of speculative events through generative imagery, such as The Riviera of the Middle East or the rising of the Third Temple, shapes political discourse and reconfigures collective perception in the present. The synthetic Gaza Riviera did not, in itself, cause the implementation of Trump’s 20-Point Plan. Yet it provided the visual scaffolding – the aesthetic plausibility – that helped render the plan imaginable, acceptable and ultimately normalized within the visual economy of our times.

In examining how the visual framing of future events shapes the present, this essay foregrounds the archival dimension of generative AI, reflecting on the ways synthetic images come to constitute new forms of archival knowledge (Hoskins Reference Hoskins2025). Generative AI operates as a kind of prophetic visual apparatus, reversing the traditional role once played by visual media in Western tradition. Whereas a central – though not exclusive – dimension of photography and visual media in Western cultures has been their evidentiary function, their claim to index and attest to reality, generative technologies invert this logic: they give photorealistic form to what might occur, visually scripting engineered futures into the texture of the present.

Synthetic visuals such as Gaza Riviera or the burning of al-Aqsa are not mere digital curiosities or isolated aesthetic experiments. They accumulate into a speculative archiveFootnote ¹² – a corpus oriented toward documenting possible futures rather than recording past realities. Even unrealized, they exert force in the present, shaping collective perception, structuring affect and normalizing certain horizons of violence and belief before they ever come to pass.

Circulating through networked environments where algorithmic moderation often overlooks or misclassifies synthetic content that does not depict explicit harm, these images acquire durability, visibility and momentum. Their persistence raises urgent questions about the kinds of archival knowledge they generate and the ways they displace traditional evidentiary practices grounded in the so-called “evidence-image” (Didi-Huberman Reference Didi-Huberman2008, 89). By stabilizing certain visions of the future through aesthetic allure, algorithmic visibility and narrative coherence, these images, in fact, instantiate a new archival logic. This logic, which privileges anticipation over memory, operating as a tool for normalizing projected futures, defines what I call the future archive: an aesthetic and political formation that preserves not what has happened, but what is projected to happen; what is imagined as likely, desirable or inevitable.

Unlike traditional archives, which secure authority by fixing records of the past, the future archive consolidates power by storing, selecting and scripting the future. It collapses anticipation and recollection, projection and preservation, affirmation and erasure into a single visual and epistemic formation. Synthetic images do not record reality; they install it. They normalize speculative futures within the collective imagination, embedding them as the evidentiary records of tomorrow.

The future archive operates at the intersection of memory and simulation, storage and generation, access and recombination. It is neither a passive repository of traces nor merely a generative engine, but a dynamic formation that merges and reconfigures both. It is at once mnemonic and performative – an archive that remembers by inventing and invents by remembering and building on past knowledge. Recursive, self-updating and cumulative, it absorbs the past, anticipates the future and scripts both into the present.

These operations mirror the logic of generative AI. On the one hand, AI bears deep continuities with traditional archival technologies: it classifies, stores and organizes knowledge, extending the epistemic order that defined earlier regimes of visual and informational control (Crawford Reference Crawford2021). On the other hand, though, as the essay will discuss, it introduces a radical departure. The archive, once static and finite, becomes generative, capable of producing a seemingly infinite stream of artifacts derived from its own corpus. Each prompt activates this mechanism of recombination: a process that simultaneously accesses the archival body of data on which the machine was trained and reconfigures it into something new (Meyer Reference Meyer2023). Generation thus emerges as a hybrid act of retrieval and invention, a continuous negotiation between memory and imagination.

This marks a profound transformation in the archival paradigm. The very act of prompting exposes AI as an archival technology of speculation, one that not only indexes the past but fabricates the future from it, merging storage with creation and recollection with projection. In doing so, it converts the archive into an engine of perpetual renewal, where knowledge is seemingly endlessly reanimated in synthetic form.

It is within this generative logic that speculative violence takes shape. By extending the archival function into the domain of prediction and fabrication, generative AI does not merely record or recall the world, but it prefigures it. The future archive, powered by these speculative operations, becomes a mechanism for scripting and legitimizing futures before they unfold, rendering them visible, familiar and thus acceptable.

This essay examines how such imagery, produced through the intertwined logics of memory and projection, feeds into the corpus of the future archive, and how this archive is operationalized in the present: at once continuous with older regimes of knowledge and radically disruptive of them. Ultimately, it reflects on the visual, technological and ideological entanglements through which generative AI transforms politics into an aestheticized form of world-making.

3. The logic of speculative violence

Speculative violence names a mode of harm enacted not through direct physical force but via the visual manipulation of futurity enabled by generative AI, the moment when its world-making potential becomes an engine of annihilation, erasure, dispossession and destruction. It unfolds across multiple registers. On one level, it manifests as material violence, as in the Gaza Riviera project, which envisions peace through real-estate development and prosperity implemented via ethnic cleansing – the dream of stability built on erasure or deportation of an entire people. Yet it also operates as a form of symbolic violence, in Pierre Bourdieu’s sense (Bourdieu and Wacquant Reference Bourdieu and Wacquant1992): a mechanism of domination exercised through symbols, ideas and beliefs, shaping perception and consent without the need for overt coercion.

Like symbolic violence, speculative violence legitimizes harm by naturalizing unequal power relations and embedding them in visual narratives that appear not only plausible but inevitable. Both forms act on the individual and collective imagination, securing complicity through affect and belief. But while symbolic violence reproduces existing hierarchies through the internalization of dominant norms, speculative violence reaches forward in time: it conditions the emotional and cognitive ground upon which future violence will appear acceptable. It is the pre-emptive aesthetic of domination, the work of power performed not on what is but on what will be.

Speculative violence is anticipatory: it projects future harm into the present through visual simulation, fostering early acceptance of potential acts of destruction. It can thus be understood as an extension of symbolic violence into the domain of futurity – its pre-enacted form – visually rendered and operationalized through generative AI. It is the techno-social apparatus shaped by the latter that determines the mechanisms through which speculative violence unfolds. Whereas symbolic violence is mediated through institutions, discourse and social norms, speculative violence circulates through algorithmically governed platforms that present themselves as neutral, decentralized and nonhuman. The pervasive myth of technological neutrality renders its operations even more insidious, cloaking ideological function in the language of innovation, efficiency and inevitability.

In this sense, speculative violence emerges as a technologically powered form of symbolic violence, where domination is aestheticized and normalized through the photorealistic realism of generative AI.Footnote ¹³ The smooth surfaces of synthetic imagery conceal the roughness of power, where annihilation appears as design.

To grasp its full operation, it becomes essential to recognize speculative violence as a byproduct of generative mediation itself. While generative AI does not inherently produce speculative violence, the latter is inseparable from the media ecosystem that generative technologies define, one organized around automation, acceleration and the endless recombination of data. At the core of this system lies the prompt, the generative gesture through which memory and projection, archive and anticipation, are bound together into the image of a possible world (Meyer Reference Meyer2023, Reference Meyer2025; Somaini Reference Somaini2023).

Prompting is not simply the AI-powered equivalent of browsing an archive or querying a web search engine. It is not a matter of locating and retrieving preexisting objects within a repository, nor should the training datasets through which AI learns to “see” the world be imagined as static vaults from which ready-made answers are extracted. Prompting is a generative act, calling something into being. It transforms the archive from a site of “excavation” into one of “construction” (Foster Reference Foster2004, 22), a space where the possible takes shape.

As art historian Roland Meyer observes, prompts are “meant to produce what they describe” (Meyer Reference Meyer2023, 102), drawing from the dataset not as a fixed reserve but as a field of latent potential. When a prompt is issued, it activates this latent field, setting the machine to actualize possibilities that are present only in virtual form – producing a new image, a video, a text that did not previously exist, even if assembled from fragments of what once did. In this sense, prompting becomes a kind of conjuring: a technological performative through which description becomes creation, and memory becomes invention.

Artist and critic Eryk Salvaggio likens this process to a séance, as if “a sort of digital archive is reanimated” through prompting (Salvaggio Reference Salvaggio2023). The information needed to fulfill the prompt is already present in the archive-like structure of training datasets – most likely scraped from the web – which deep-learning algorithms translate into the so-called latent spaces (Somaini Reference Somaini2023, Reference Somaini, Ackerman, Gefen and Somaini2025). These are abstract environments where cultural data are rendered into simplified and computationally tractable forms, awaiting activation. A prompt thus “recalls,” as if conjuring the dead, the mathematical representations that are lying dormant in these spaces. Yet, more than resurrection or reanimation, the act of prompting rather evokes recombination: a weaving of fragments into new, synthetic wholes. Unlike traditional archival devices such as libraries, museum collections and repositories, latent spaces are not meant for preservation and access but rather “for transformation” (Somaini Reference Somaini, Ackerman, Gefen and Somaini2025, 37), enabling the generation of new cultural objects that result from endlessly combining existing ones.

Different generative methods exist – Salvaggio discusses diffusion models in particular – but all operate according to the same principle: that recombination happens not so much on the premises of cultural or semantic logics but, rather, on statistical ones. In latent spaces, every data point is treated as a kind of “synonym” for the thing it is meant to represent. These synonyms are clustered, and the machine samples from them to generate outputs. As an example, an image of a cat within a dataset is abstracted into a set of numerical features that, in the latent space, contribute to shaping a generalized representation of “catness.” When prompted with references to this category, the system does not retrieve the original image – or any other image – but instead samples from the statistical distribution encoded in the latent space, recombining these features to generate a new instance of a “cat” in the specific form prescribed. What the human eye perceives as an output of the prompting process – an image, for example – is, in fact, the product of statistical operations: a rendering produced by mathematical calculations of probability and pattern extraction related to the abstract idea of “catness” that the system has assembled through the training process.

The pictures thus generated are no longer truly pictures at all, but “statistical renderings” (Steyerl Reference Steyerl2023, 82), predictive reconstructions assembled through pattern analysis, what Salvaggio calls “hypothetical images” (Salvaggio Reference Salvaggio2023). Rather than capturing reality or preserving memory, these visuals simulate what is statistically plausible, drawing upon prompts that are extracted from vast training corpora now reduced to raw operational material, stripped of history, affect and situated experience.

According to Salvaggio, diffusion models disintegrate the image: cultural memory is atomized into aesthetic fragments and algorithmically reassembled without reverence, care or historical weight. “The images of the Wehrmacht on holiday live side by side with images of their victims,” he writes. “In that way, Diffusion is disintegration” (Salvaggio Reference Salvaggio2023). Such juxtapositions are not accidental glitches but symptoms of a system that renders visual culture into a field of undifferentiated patterns, where the moral hierarchies that once structured meaning dissolve into statistical equivalence.

The term diffusion betrays this logic: it describes a process of dispersal, a soft, continuous scattering of forms. “These images are diffused,” Salvaggio observes. “None of the training data is being destroyed,” yet it is nonetheless “being desecrated” (Salvaggio Reference Salvaggio2023). The violence here is subtle but profound: not the erasure of images, but their profane survival, their transformation into data ghosts, endlessly recombined, emptied of memory yet haunted by it.

Within the generative visual regime, what once counted as visual testimony risks being displaced by algorithmic simulation. The past is mined for tropes, flattened into aesthetic residue and repurposed for frictionless circulation. Hypothetical images no longer tell us what was: they speculate on what might plausibly have been. Their logic is not that of remembrance or documentation but of probability and entropy.

Plausibility is a key concept in defining the generative visual regime. It lies at the center of what Meyer – drawing on Jacob Birken’s intuition – calls “platform realism” (Meyer Reference Meyer2025), an aesthetic formation born from the conditions of AI-based image synthesis. In an environment shaped by the generative possibilities, AI models are engineered to produce ever more “‘realistic’ renderings” (Meyer Reference Meyer2025, 3), a goal dictated as much by corporate design and market optimization as by the anticipations of users’ expectations and tastes.

The pursuit of this new form of realism, shaped jointly by platforms and users, is centered around plausibility, understood as the degree to which synthetic images conform to the normative assumptions of coherence, physical integrity and spatial logic that structure collective perception. Realism, here, is no longer about aesthetic endeavors or artistic intentionality but, rather, about the statistical optimization of outputs to match user demands and collective visual habits. What was once the work of intention, composition and witness becomes the outcome of calibration: a realism generated not to reveal truth but to satisfy the infrastructural needs of the platform.

And yet this process should not be understood as a mere, blind operation of statistical rendering. As Meyer notes, the generative process must satisfy prompts in ways that resonate broadly, thereby maximizing the likelihood of audience approval. In this sense, plausibility is not just a technical metric but also a social one, defined by what is most likely to win popular consent and approval.

Artist and media theorist Hito Steyerl has coined the term “mean” (Steyerl Reference Steyerl2023) to describe this statistical dynamic: a convergence toward averages and medians, where likeness gives way to likeliness. In such a process, meaning is not preserved but violated, expropriated and made cruel: turned into something malevolent, nasty, malicious – mean in every sense of the word. This moral and aesthetic flattening becomes starkly visible in Salvaggio’s prompting experiment, where snapshots of the German Wehrmacht coexist with portraits of Holocaust victims. Here, the machine’s indifference produces a terrible intimacy: perpetrators and victims rendered equivalent, folded into the same average for some statistical reason that only the algorithm knows.

This meanness of AI – the convergence toward a statistically “neutral” middle that can also become morally vicious – is crucial to understanding plausibility not merely as a technical affordance but as the aesthetically flattened result of an “incessant digital plebiscite” (Meyer Reference Meyer2025, 7). It is a realism as consensus, attuned to the crowd, engineered for affirmation and optimized for approval, fine-tuned to please.

In this sense, plausibility and popularity converge. The believable becomes what is most liked, and the credible what is most circulated. This pursuit of affirmation – “applause” and “approval” (Meyer Reference Meyer2025, 7) – is structurally tied to another defining feature of platform realism: its conservative relation to the past. Generative AI models perform a “backward prediction” (Meyer Reference Meyer2025, 11), as their images of the future are built through recombinations of data drawn from the past.

Yet this past is not historically grounded or contextually alive. It appears stripped of specificity and reduced to a generalized aesthetic repertoire, resulting in what Meyer – citing Fredric Jameson – calls “pastness” (Meyer Reference Meyer2025, 11–14): a decontextualized evocation of earlier epochs, recognizable only through stylistic clichés loosely associated with a given era. It is nostalgia without history, memory without meaning, a flattened temporal field for the sake of aesthetics.

In this sense, platform realism enacts a kind of structural nostalgia. It gives shape to futures that are visually haunted by stylized versions of the past, images in the style of whatever epoch is summoned through the prompt. The resulting temporality is neither wholly retro nor futurist but a recursive loop of imagined pastness. Slogans like Make America Great Again are deeply rooted into this logic, as they call into being a stylized fantasy of how the past once should have appeared (Waite Reference Waite2025; Watkins Reference Watkins2025).

Perfectly aligned with the structural nostalgia that underpins platform realism, both Netanyahu’s and Trump’s AI-generated visions of Gaza as the Riviera of the Middle East draw upon pastness to imagine a sanitized, depoliticized future for the Strip. These synthetic landscapes depict a Gaza finally “liberated” from Hamas and its subterranean life in tunnels, only to replace it with a fantasy lifted from colonial repertoires long dissected by Edward Said in Orientalism (Reference Said1978). Beneath the glossy surfaces of these renderings lingers a visual grammar built on tropes such as the harem and the souk: a stereotyped repertoire from the past recast as an emblem of renewal.

An earlier example, the AI-generated advertisement Come Visit Gaza! – released on Hulu in January 2024, predating both Netanyahu’s Gaza 2035 and Trump’s Gaza Riviera – follows the same logic (Kuntsman and Stein Reference Kuntsman and Stein2026). Visually recalling the genre of tourist advertisement, it reimagines Gaza as a luxury destination of immaculate beaches and upscale restaurants, where veiled women smile over a plate of “Arabic food” eaten with chopsticks. The irritating fusion collapses cultural specificity into a consumable sameness: the “Orient” flattened into an aesthetic commodity for global audiences. It is the same structurally conservative nostalgia that animates Trump and Netanyahu’s visuals on Gaza. For all their futurist rhetoric, these synthetic images recycle, both aesthetically and ideologically, a reimagined and depoliticized past.

This vision of progress is tied not to history or place; it bears no trace of lived Palestinian reality. Instead, it mirrors the architectural vocabulary of Dubai and other Gulf metropolises, a globalized fantasy of consumption and urban spectacle. Gaza’s supposed future becomes an apolitical, consumerist mirage, reconnecting to a sanitized exoticism reminiscent of One Thousand and One Nights. It is an image untethered from context, perfectly attuned to the logic of speculative violence: a projection that erases the present in order to implant, within the future, a reengineered and structurally nostalgic pastness.

4. Speculative violence as “hyperstition”

Speculative violence operates across two planes that appear contradictory yet are, in truth, interdependent: a future-oriented imagination, and pastness as its very fuel. Its vision of what is to come is built from what has already been: from recycled aesthetics to inherited hierarchies of representation. Likewise, it unfolds through two entwined temporalities: anticipatory and immediate. While projecting a vision of the future, it simultaneously acts upon the present, rendering the not-yet as both visually and affectively real.

This is not merely the articulation of what might happen, nor a neutral hypothesis about possible outcomes. It is a speculative projection that manufactures feasibility: one that, through repetition and circulation, helps construct the very conditions for its own realization. In this feedback loop between vision and event, imagination becomes infrastructure, and the projected future begins to materialize within the now.

In this sense, speculative violence resonates profoundly with the concept of hyperstition, first formulated by the Cybernetic Culture Research Unit (CCRU) at the University of Warwick in the 1990s (CCRU 1999; Carstens Reference Carstens2002). Hyperstition names the process by which fictions or marginal narratives, through circulation, repetition and belief, acquire social and material traction – ultimately influencing and reshaping reality itself. Unlike superstition, which implies irrational belief against evidence, hyperstition operates through recursive feedback loops between the virtual and the real. As philosopher Nick Land famously put it, hyperstitions are “fictions that make themselves real” (Land Reference Land, Mackay and Brassier2012, 554).

They act as future-oriented agents, retroactively conditioning the present, operating through a non-linear temporality in which narrative precedes, prefigures and catalyzes its own realization. The phenomenon is legible in movements such as QAnon and other conspiracy-driven networks that, regardless of factual grounding, generate concrete political and social effects. This is not accidental, as Land’s theorization of hyperstition underpins the intellectual genealogy of far-right imaginaries – and of the neoreactionary movement known as the Dark Enlightenment, with echoes among influential figures both in Silicon Valley and with the second Trump administration (Della Ratta, Reference Della RattaForthcoming) – clarifying how speculative narratives are mobilized as a political strategy rather than remaining mere belief. The latter is turned into an instrument of world-building: a form of action that precedes verification. When collectively sustained and algorithmically amplified, it no longer simply mirrors reality; it begins to manufacture it.

Speculative violence functions as hyperstition as it does not merely prefigure the material realization of events – such as the construction of the Gaza Riviera or the Third Temple – somewhere in a distant future, but rather imagines them into being, giving them visual form through which acceptance and normalization can take root in the present. It operates through visual recursion: each repetition, each share, each algorithmically generated frame compounds plausibility, training the collective imagination to regard annihilation, erasure and dispossession as destiny.

Ultimately, whether the Gaza Riviera will ever be built exactly as envisioned in these dystopian AI renderings becomes secondary. What matters is that the images – and the discursive ecosystems surrounding them – have already ceased to function as mere speculations about what might come. They now operate as operational myths, shaping affect, belief and perception in real time. As the Trump 20-Point Peace Plan moves toward implementation (at the time of writing), the speculative has folded into the actual: the imagery did not cause policy, yet it prepared its terrain, softening resistance, aestheticizing violence and turning prophecy into governance.

Compared to the 1990s, when the term was first coined by the CCRU, the visual immediacy and circulatory potential of today’s AI-driven media ecosystem have dramatically amplified the conditions under which hyperstitions come into being. What once required years of subcultural dissemination or speculative writing now unfolds through automated synthesis and instantaneous virality. Generative technologies have transformed hyperstition from a textual and conceptual experiment into a visual infrastructure, a system capable of materializing belief as image – photorealistic, clickable and infinitely reproducible. These tools radically accelerate the diffusion and affective uptake of hyperstitional content, particularly through algorithmic feeds that privilege emotional intensity and engagement over verification.

Even the notorious 2004 remark – attributed to a senior advisor in George W. Bush’s Republican administration – that “we’re an empire now, and when we act, we create our own reality” (Suskind Reference Suskind2004), was uttered in a pre-platform media landscape. At that time, social media, algorithmic content curation and generative image technologies had not yet arisen to provide the powers that be with a fully fledged, visually persuasive ideological apparatus. The statement, made by an administration deeply invested in prosecuting the post-9/11 “War on Terror” – a conflict already fought as much through images as through weapons – was both provocative and prophetic. Yet the technological infrastructure capable of materializing such a worldview into a pervasive, self-reinforcing visual system had not yet matured. What existed then as political bravado has, two decades later, found its machinery: a platformed empire that literally manufactures its own reality, one prompt and one image at a time.

This capacity to project and materialize hyperstitions, or, more broadly, political ideologies, into a shared visual reality is now made possible through generative AI. Within the emergent media ecology of platform realism, speculative violence translates the hyperstitional logic into synthetic imagery, helping to construct visions of the future that become operational in the present. What hyperstition lacked at the time of its conceptualization – a technological infrastructure capable of turning narrative into compelling visual storytelling – is now provided by generative AI.

In this sense, AI functions as a paradigmatic medium of our era, crystallizing the shift underway in the post-9/11 world, and emphasizing the need to reconsider the foundations of politics in an age where the Enlightenment’s optimism has been shaken by traumatic events. As remarked in “The Straussian Moment” (Reference Thiel2007), a provocative essay by prominent tech entrepreneur Peter Thiel, the future, in the aftermath of the Twin Towers attacks, must contend with violence not as an aberration to be eliminated, but as a structural element to be reintegrated into political and historical consciousness.Footnote ¹⁴

In today’s political context, marked by the consolidation of the Israeli far right and the resurgence of a Judeo-Christian alliance under a second Trump administration, violent events such as the reconstruction of the Third Temple replacing al-Aqsa, or the making of the Gaza Strip into a luxury resort devoid of Palestinian life, no longer appear unlikely. Once again, the notion of plausibility proves central to understanding how speculative violence operates through generative imagery. As previously discussed, plausibility must not be reduced to mere statistical likelihood but rather referred to as an “expectable expectation” (Meyer Reference Meyer2025, 7–11), a set of aesthetic and narrative conventions calibrated to align with dominant norms and codes, thereby securing public recognition and approval.

This is aesthetic plausibility that caters to the average, the familiar and the algorithmically favored. However, what is at stake with the Riviera of the Middle East or the building of the Third Temple also concerns geopolitical plausibility, a condition grounded in the long and troubled genealogy of the Israeli–Palestinian conflict. The AI-generated vision of a beach resort emerging from the ruins of Gaza – initially met with global backlash due to its glaring erasure of Palestinian presence and implicit endorsement of ethnic cleansing – is not merely a visual provocation. It becomes conceivable precisely because it resonates with a historical continuum of dispossession and dehumanization of Palestinians, who have long been rendered invisible within dominant discourses (Baroud and Rubeo Reference Baroud and Rubeo2021). The synthetic future depicted is, thus, not only statistically plausible or aesthetically palatable; it is historically and politically legible and legitimized.

Its plausibility is reinforced by a long record of historical precedents. As an example, the al-Aqsa Mosque was set on fire in 1969 by an Australian Christian fundamentalist who claimed to be fulfilling a biblical prophecy tied to the rebuilding of the Third Temple. It has also been repeatedly targeted by Jewish extremist groups – most famously in 1984, when members of the Jewish Underground plotted to blow up the site. Yehuda Etzion, one of the group’s founders, later recalled: “In my mind I already saw the Dome collapsing on itself with a huge cloud of rising dust” (Shavit Reference Shavit2013, 218).

Today, when prominent Israeli figures such as National Security Minister Itamar Ben-Gvir openly and bluntly violate the fragile status quo on the Temple Mount, especially amid the escalation of violence following October 7, 2023, these are not to be understood as mere symbolic provocations but are indeed concrete political trajectories. Strikingly, Jerusalem appears in CCRU’s collective volume Digital Hyperstitions (CCRU 1999) discussing the incendiary millennial visions of Judeo-Christian fundamentalists seeking to hasten the end times by destroying the al-Aqsa mosque and rebuilding the Third Temple on its ruins.

The generative images of the Third Temple or the Gaza Riviera, thus, do not emerge in a vacuum. They are not detached from reality but densely entangled with it. They echo and amplify futures already latent within the political imagination of powerful actors, translating ideological ambitions into visual probability. In this way, speculative violence operates precisely at the juncture between technological simulation and historical continuity, rendering the once-unthinkable both seeable and sayable.

What this dynamic produces in the collective imagination is a process of normalization and acceptance by means of visualization; an effect that, I would argue, connects directly with what Walter Benjamin called the “optical unconscious.” Benjamin discusses this idea, fascinating and mysterious as in his characteristic style, across several of his writings. In The Work of Art in the Age of Its Technological Reproducibility, he engages with the concept in the context of his reflections on photography and visual technologies. Here, the optical unconscious refers to dimensions of reality that escape unaided human perception but are made visible through technological mediation. Just as psychoanalysis reveals hidden drives and latent content within the human psyche, photographic techniques - according to Benjamin - expose concealed structures and patterns in the visual field, details that are often overlooked in everyday experience yet registered mechanically and unconsciously (Benjamin Reference Benjamin and Jennings2008, 37–38).

However, it is in an earlier and less frequently cited text, News about Flowers (1928), that Benjamin offers an articulation of the “experimental metaphor” (Conceição Reference Conceição2016, 6) of the optical unconscious, in striking resonance with the mechanisms of generative AI. In this short essay, the photographic camera is not merely a visual apparatus that registers details overlooked by the naked eye; it reveals an entirely new mode of seeing that had no perceptual equivalent before technological mediation. In this formulation, it is not so much the psychoanalytic framework that is evoked, but rather the concept of mimesis, which Benjamin reimagines in far more expansive terms than its classical use as a theory of representation. Mimesis is not simply the imitation of the real; it is “the organon of experience” (Benjamin Reference Benjamin1999, 868), a primal and almost magical faculty through which hidden correspondences within the cosmos are perceived. It is a sensory-affective capacity to register likeness not in appearance but in resonance, rhythm and affinity.

The optical unconscious is where visual technologies activate and extend this mimetic faculty. The camera becomes not a neutral recorder but a tool for making perceptible the invisible threads that bind the subject and the world in ways that conventional seeing cannot access. With the optical unconscious, the mimetic power transforms into an aesthetic possibility. Therefore, photography and visual media in general possess the inner quality of stimulating “the mimetic faculty in human beings” (Conceição Reference Conceição2016, 7), allowing for a deeper, almost animistic relationship with the material world.

When placed alongside generative AI, this understanding of the optical unconscious takes on a new relevance, as this technology operates in a similar mimetic register: not by directly imitating the real but by extracting patterns and latent affinities from data, and recomposing them into plausible futures. Like Benjamin’s photographic camera, generative technologies do not just show what is, but suggest what could be (or should be), by activating the imagination’s latent potential to recognize and accept unfamiliar combinations as meaningful.

Thus, speculative violence emerges through the subtle stimulation of this mimetic sensibility via the use of generative AI, turning future-oriented visualizations into emotionally and cognitively acceptable realities. Yet, it does not simply offer a visualization of the future through synthetic images; it actively constructs and stages prospective visions as if already folded into the texture of the present. This is where speculative violence departs from a strictly Benjaminian notion of the optical unconscious as revelatory of the mimetic bonds within the cosmos. Speculative violence, in fact, actively produces visual coherence where none exists, imposing a future that has not happened – and may never happen – as if it was already latent, unfolding, and sort of corresponding to the “natural” order of things.

In this way, generative AI becomes a political tool of prefiguration. The optical unconscious is no longer simply a register of unseen relations; it becomes a mechanism for engineering visibility itself, framing collective affect, habituating public perception and normalizing potential harm through plausibility. What Benjamin intuited in the infancy of mechanical reproduction now reaches a new intensity in synthetic foresight: speculative violence embeds itself in the seamless, the coherent, the visually credible – images that do not provoke resistance precisely because they appear innocuous enough to pass without question.

Of course, not all generative AI engages in speculative violence. However, my argument is that when generative AI is mobilized for its forward-oriented capacities, particularly in politically charged contexts such as the Israeli–Palestinian conflict, or current discourses surrounding remigration in Europe and the United States, it tends to channel its visual outputs toward latent forms of violence. These are not overtly aggressive representations, but rather synthetic projections that implicitly normalize erasure, displacement or destruction. Such images do not necessarily shock or provoke per se. Rather, they bypass critical judgment by embedding contentious political futures within a grammar of visual plausibility, one that feels coherent, aesthetically appealing and intuitively acceptable. In doing so, such synthetic images enact a form of visual engineering where ideology is smuggled in under the seemingly neutral guise of speculative design.

Although not all generative AI serves this function, its entanglement with violence echoes long-standing critiques of visual technologies as inseparable from military and strategic logics (Virilio Reference Virilio1989). Speculative violence does not signal a break with past optical regimes but, rather, their algorithmic intensification – a quieter, less openly visible and more refined phase in the militarization of visual culture, where the future itself can be operationalized as a site of ideological battlefield.

5. The future archive: is it really about the future?

The circulation and accumulation of synthetic images that project speculative futures acquire a hyperstitional force that does not merely represent a possible tomorrow but actively summons it into being: a dynamic that becomes legible when generative AI is understood as an archival technology.

On the one hand, generative AI stands, in fact, in continuity with older regimes of knowledge production and storage; on the other, it radically disrupts them, introducing unprecedented forms of preservation, access, recombination and creation.

Far from emerging ex nihilo, generative AI inherits and intensifies the archival logic of earlier visual media – particularly photography – which historically functioned as an apparatus of classification, surveillance and social normalization.Footnote ¹⁵ These visual regimes transformed ideology into visual evidence, embedding hierarchy, suspicion and control directly into the domain of the image. Generative AI stands in continuity with this lineage, reactivating and amplifying its classificatory logics through automated synthesis, scale and opacity (Crawford Reference Crawford2021). As an archival technology, AI does not merely store or retrieve the past but reorganizes it into latent structures that shape what can be rendered visible, plausible and expectable. In doing so, it reproduces the epistemic and colonial violence of earlier visual archives, now operating through algorithmic systems that silently govern perception, classification, and foresight under the guise of technological progress.

Yet this continuity is disrupted by the generative dimension of AI, which marks a radical break with past archival logics. When activated through prompting, in fact, latent spaces do not simply retrieve stored content; they synthesize it into new objects. Though the output may appear as a discrete image or video, its ontological status differs profoundly from a traditional archival artifact. As Meyer notes, it “is not just an element of an archive but its product, a contingent outcome” (Meyer Reference Meyer2023, 105) that interpolates existing materials into something that never previously existed. Rather than storing stable objects, latent spaces continually remix them, producing endless variations that are neither purely past nor entirely new, but suspended within a speculative register of possibility. This constant dynamic transformation marks the shift from an archive that preserves to an archive that generates – a shift foundational to what I refer to as the future archive.

The recombinatory and generative potential of latent spaces shifts the archive from a backward-facing apparatus of documentation to a forward-oriented engine of cultural production. In this new regime, value is no longer anchored in what has been preserved, but in what can be plausibly rendered as possible. The archive becomes a site where fragments of the past are reassembled into visions that gesture toward histories that never fully were, and futures that are not yet: an apparatus for conjuring worlds, rather than merely recording them.

This generative, forward-leaning capacity of latent spaces evokes Foucault’s conception of the archive. In The Archeology of Knowledge (Reference Foucault1972), Foucault develops the idea of the archive as an ordering principle, a system that governs which statements can be made and which remain unsayable. If discourse names the production of statements, the archive names their selection – the structuring logic that orders visibility, intelligibility and legitimacy. The archive, in this sense, is not a mere collection of documents or a sum of institutions and subjects, but the very system that determines the conditions of their possibility. It selects, orders and legitimizes, delimiting the boundaries of what a society takes to be knowledge. As Eliassen (Reference Eliassen and Røssaak2010) underscores in his reading of Foucault, this makes the archive inseparable from power: it shapes the field of visibility and intelligibility through which subjects come to recognize themselves and others, embedding power’s logic into the grain of collective memory.

Latent spaces, though computational rather than institutional, reproduce this same structure. They operate at the threshold of visibility, functioning as selective mechanisms that govern what can be generated, and thus what – out of a sanitized, abstracted reservoir of past knowledge, of pastness – will become material for future artifacts. What they exclude is as significant as what they allow to appear. In this sense, latent spaces enact the archival function at scale: filtering, ranking and conditioning the emergence of visual knowledge in ways that remain opaque, yet authoritative.

The temporalities activated by latent spaces mobilize the idea of the archive becoming generative and producing new meanings, new records and new futures. Every archival act of selection, classification or documentation is projective, as it conditions the field of future legibility. Yet, in latent spaces this projective quality becomes literalized. The archive no longer merely enables future meaning: it directly fabricates future artifacts. From this perspective, generative AI represents not a departure from archival logic but its intensification. Whereas earlier archives conditioned the future by shaping what was knowable, latent spaces produce the future directly through algorithmic synthesis. They collapse the distance between archival memory and futurity, generating speculative artifacts that prefigure the very same worlds they help bring into being.

In conclusion, generative AI does not simply inherit the archival logic of the past: it transforms it into a fully anticipatory apparatus. If the traditional archive governed futurity indirectly, by shaping what could be remembered, said or known, the generative archive governs it directly, by producing the very forms and formats through which futures become imaginable, plausible and politically actionable. In this shift, the archive’s futurial drive is no longer a latent orientation but the primary operational mode of today’s knowledge creation ecosystem.

The Gaza Riviera is emblematic of this new archival condition. Its glossy simulations are not simply fantasies projected onto an uncertain horizon; they are archival operations in themselves, recombining the aesthetic residues of Orientalized fantasies of the past into a politically instrumental future. The towers, marinas and palm-lined promenades it envisions do not break with history so much as remix it, flattening Palestinian life into a consumable pastness, while reactivating colonial imaginaries under the guise of futurity, fragments of visual culture reassembled into a forward-oriented script. Gaza Riviera stands as a paradigmatic object of the future archive: an anticipatory artifact whose power lies in naturalizing a speculatively violent horizon, making dispossession appear not only plausible but prefigured, as though the future were already inscribed in the residue of the past.

6. Conclusion: the archive as a battlefield of the future

This essay has traced the visual, technological and ideological entanglements between generative AI and the construction of politically charged futures rendered in visual forms. By framing generative AI as an archival technology – a future archive – we are better equipped to understand the latent power of synthetic imagery as a mode of world-making, in historical continuity with other visual apparatuses of colonial control. Synthetic images draw their authority from a deep visual and historical reservoir, recombining past representations into aesthetically plausible visions of what is to come. Their danger lies precisely in their plausibility: the way they align with ideological, cultural and historical trajectories already at work in the present.

Through the lens of speculative violence, I have shown how synthetic images prefigure and normalize harm before it occurs. Whether imagining Gaza as a luxury resort, or visually enacting the destruction of al-Aqsa to make way for the Third Temple, these visuals participate in the symbolic conditioning of the future, engineering emotional and political receptivity in the present. They shape collective affect not through coercion but through coherence: a sort of visual familiarity that disarms critique and paves the way for acceptance and normalization.

In this light, generative AI does not herald a break from the past but rather intensifies longstanding visual regimes of surveillance, control and erasure, repackaged through a seductive grammar of innovation. What is at stake is not only how we imagine the future, but how we remember the past; and who gets to author both. Understanding the archive not as a static repository but as a dynamic, generative apparatus is essential for grasping the stakes of this moment. The future archive is not a promise of progress but a contested site where old forms of domination are rehearsed in generative, photorealistic languages. To intervene critically, we must learn to read these images not only for what they show, but for what they obscure, anticipate, and seek to make inevitable.