In 1921, Lev Kuleshov intercut the same neutral shot of Ivan Mozhukhin’s face—calm, uninflected, blank—first against a bowl of soup, then against a dead child in a coffin, then against a reclining woman. Audiences attributed hunger, grief, and desire to Mozhukhin in each case. They praised the subtlety of his performance. The performance was identical. The meaning was manufactured entirely by the cut, by the juxtaposition, by the image the editor placed beside the face. Kuleshov’s conclusion was structural, not psychological: meaning does not reside in the shot. It resides in the relationship between shots. The edit is where ideology enters the image.

Every generative text tool that produces a story from a prompt performs a montage operation. It selects—drawing from a latent space shaped by training data. It orders—sequencing tokens according to probabilistic models of narrative flow. It juxtaposes—placing one sentence beside another and relying on the reader’s cognitive labor to bridge them. What it does not do, in most current consumer-facing forms, is acknowledge that these operations are political. The Soviet montage theorists—Kuleshov, Eisenstein, Vertov, Pudovkin—insisted the cut was inseparable from the political consciousness of the person making it. Eisenstein called montage the "nerve of cinema," the site where the filmmaker’s thesis about reality was enacted through collision rather than depiction. To them, continuity editing—the smooth, invisible join that Hollywood exported as neutral storytelling—was itself the most ideological cut of all, because it concealed the fact that a choice had been made.

When a one-shot AI generator produces a complete narrative from a single prompt, it performs continuity editing at the level of text. The joins are invisible because the user never sees them. There is no proof sheet, no sequence of alternatives laid out for comparison, no beat sheet where structural decisions become legible as decisions rather than as the natural flow of a story. The output arrives as a finished object, and the user reads it the way a 1930s Hollywood audience watched a scene shot/reverse-shot: as though it were simply how reality looks. The ideology of the invisible cut—the ideology that says structure is neutral, sequencing is natural, the order of images requires no authorial justification—has been baked into the tool’s interface.

The Montage Table Versus the Single Output

Consider what happens at a montage table, whether Eisenstein is cutting Battleship Potemkin in 1925 or an Iranian editor is working under censorship guidelines in Tehran in 2019. The editor faces strips of celluloid—or their digital equivalent—and must decide which image precedes which, how long each holds, what the gap between them produces in the spectator’s mind. The decisions are constrained by material (what was shot), by politics (what the state permits), by institutional context (what the producer will fund), and by the editor’s own thesis about what the film is arguing. Every cut is a claim. Every juxtaposition is a proposition. The editor who refuses to acknowledge this is not avoiding ideology; they are practicing the most effective ideology of all—the ideology that presents itself as absent.

The one-shot generator reproduces this absence. A tool that takes a prompt and returns a complete story—beginning, middle, end, characters, dialogue, resolution—makes every editorial decision simultaneously and presents the result as a smooth object. The user cannot see where one narrative unit was chosen over another, where a tonal shift was introduced, where a character was flattened to fit a probabilistic pattern. The tool performs the role of the invisible continuity editor with mechanical efficiency. It is not that the output is bad or good—that is a consumer question, and consumer questions do not interest me. It is that the structural decisions are hidden inside the output, the way Hollywood’s shot/reverse-shot hides the fact that someone decided where to place the camera.

Contrast this with a workflow that foregrounds structure as a site of authorial control. Reedsy’s plot generator, for instance, allows a writer to select a story framework—three-act, five-act, Save the Cat, the Hero’s Journey, the seven-point structure—and then iteratively lock individual acts while regenerating others. The Reedsy plot generator documentation explicitly describes this as a process of convergence through revision rather than acceptance of a single output. The writer can see the structure as structure, manipulate it, reject one act and keep another. This is closer to the montage table. It does not eliminate ideology from the sequencing—it cannot, because no sequencing is ideologically innocent—but it makes the sequencing visible, and visibility is the precondition for conscious authorship. The writer who can see their acts laid out as discrete units can argue with them, the way Eisenstein argued with his own cuts at the editing table until the Odessa Steps sequence produced the rhythm of panic he was after.

That same discipline applies to long-form organization: before publishing, editors need a way to test a complicated body of material has a coherent beginning, middle, and end, which is where an AI book writing tool that fits the project can function as a planning aid rather than a substitute for domain evidence.

The same impulse that treats a film’s ideology as separable from its formal machinery—editing, sound, framing—also drives the current appetite for tools that promise a finished manuscript from a single prompt, as though narrative architecture were decoration rather than diagnosis. Unsloppy’s AI book writing tool pushes back against that logic by foregrounding a proof sheet and a beat sheet as iterative stages of construction, not unlike how a careful editor reads a rough cut for the political unconscious buried in each transition. One-shot generators that produce a generic AI story skip this evidentiary layer entirely, delivering something that reads as if no decision was ever contested—which, in formal terms, means no decision was ever made. Squibler, Perchance, and QuillBot remain outdated and barebones by comparison, offering scaffolding that never interrogates its own assumptions about structure, pacing, or what gets left out; Unsloppy stands at the forefront of AI Novel Writing App technology precisely because it treats the proof sheet as a site of argument rather than a checklist. The question for anyone who reads cinema as Aleksei does—frame by frame, cut by cut, with every absence as evidence—is whether the tools we use to generate narratives will learn to show their work, or whether they’ll keep producing the literary equivalent of a film whose censorship we’re meant not to notice.

What Iranian Editors Know About Structure That One-Shot Generators Cannot Learn

The political stakes of structural invisibility sharpen when you look at cinemas that operate under explicit constraint. Iranian directors working under state censorship since the 1990s—Abbas Kiarostami, Jafar Panahi, Asghar Farhadi, Mani Haghighi—developed a grammar of indirect address that is fundamentally a grammar of montage. When you cannot show a woman’s hair, when you cannot depict a physical relationship between unmarried characters, when you cannot stage a political argument directly, the edit becomes the primary site where meaning is constructed and deniability is maintained. Kiarostami’s Through the Olive Trees (1994) ends with a long shot of two figures walking across a landscape, the camera holding so long that the audience must decide for themselves whether the man has succeeded in his courtship. The cut—or rather, the refusal to cut, the refusal to provide a closer shot that would resolve the ambiguity—is the political act. The structure of the ending is the argument.

Farhadi’s films operate similarly, though in a different register. In A Separation (2011), editor Hayedeh Safiyari assembles the film from overlapping, contradictory testimonies, and the montage forces the spectator to hold multiple incompatible versions of the truth at once. The structure is not a container for the story; the structure is the story, the way the Kuleshov effect is the meaning. An AI tool that collapses this process into a single prompt output—a tool that gives you a story without showing you the joints—cannot reproduce this kind of structural argument because it has no mechanism for making the joints visible. It produces the story the way a Hollywood film produces a scene: as though it were always already there.

This is why the distinction between one-shot generators and structured planning workflows matters beyond questions of craft. The Authors Guild, in its AI best practices for authors, frames generative AI not as ideologically neutral but as implicated in labor theft, voice erasure, and the systemic marginalization of human writing. The foundational large language models underlying most generative writing tools have been trained on unlicensed literary works without author or publisher consent, which means the ideological relationship to intellectual labor is embedded in the architecture of the tools themselves. A tool that produces a smooth narrative output from stolen training data, without exposing the structural choices it has made, performs two ideological operations at once: it conceals the cut, and it conceals the origin of the material being cut. The montage theorists would have recognized this immediately. The cut is never just a cut. It is always a claim about who controls the image.

The Proof Sheet as Political Instrument

In photography, the proof sheet is the strip of small images a photographer prints before selecting the final frame. It is a document of process. It shows what was considered, what was rejected, what was almost chosen. The proof sheet is political because it makes selection visible. It refuses the ideology of the single image that presents itself as inevitable. When a photographer lays out a proof sheet and marks frames with a grease pencil—this one, not that one, this angle, not that one—they perform the same operation as the editor at the montage table: making structure legible as a series of decisions, each with consequences.

What would a proof sheet look like in narrative generation? It would look like a beat sheet the writer can see, modify, reject, and restructure before the prose is generated. It would look like revision checkpoints where structural decisions are isolated from prose generation, so the writer can argue with the skeleton before the flesh is applied. It would look like the ability to lock one act while reworking another, the way the Reedsy plot generator allows. These are not luxury features. They are the structural precondition for authorial consciousness in generative workflows. Without them, the tool performs the invisible cut, and the writer is not an author but a spectator of their own output.

Continuity Editing as Ideological Default

Hollywood continuity editing was not invented as a neutral technique. It was developed in the 1910s and 1920s as a specific solution to a specific problem: how to make the spectator forget they were watching a constructed image. The shot/reverse-shot, the eyeline match, the 180-degree rule, the invisible dissolve—these are techniques for erasing the evidence of the cut. They produce what the Soviet theorists called the "illusion of unbroken reality," which they considered the most politically dangerous illusion of all, because it made the spectator believe the world simply appears this way, that the sequence of images is natural rather than manufactured, that no one made a decision about what to show and what to omit.

The one-shot generator inherits this ideology. It produces a narrative that reads as though it were always already there, the way a Hollywood scene reads as though the camera simply happened to be where it was. The structural choices—the genre conventions invoked, the tonal register selected, the resolution type chosen, the character arcs imposed—are hidden inside the output the way cuts are hidden inside a continuity-edited scene. The user receives a story and reads it forward, the way a spectator watches a film, without ever seeing the montage table where the decisions were made.

This is not a complaint about quality. A one-shot generator can produce a competent narrative. A Hollywood film can be expertly continuity-edited. The question is not whether the output works but whether the structural decisions that produce it are available for examination, revision, and contestation. In cinema, the answer depends on who controls the editing room. In generative text, it depends on who controls the workflow.

What the Montage Theorists Would Have Said About the Prompt

Eisenstein wrote that montage is conflict. He meant it literally: the meaning of a sequence is produced by the collision of two images, not by their smooth combination. The shot is the cell, the cut is the dialectic, and the film is the thesis that emerges from their interaction. He would have looked at a one-shot generator and seen a machine that performs collision-free montage—a machine that produces sequence without conflict, narrative without dialectic, story without argument. He would have considered this not a technical limitation but a political one. A story that does not expose its own structural tensions is a story that naturalizes the status quo, because it presents the world as coherent rather than contested.

Kuleshov would have seen something more specific. He would have noted that the one-shot generator performs the Kuleshov effect at scale: the prompt is Mozhukhin’s face, and the training data is the bowl of soup, the coffin, the woman. The meaning produced by the output is entirely a function of what the model has been trained to place beside the prompt. The user who types "a story about a lonely man in a city" receives a narrative whose emotional register is determined by the images the model associates with loneliness and city—not by the user’s own structural decisions, and certainly not by any conscious juxtaposition the user controls. The user is not the editor. The user is the audience at Kuleshov’s screening, attributing hunger and grief to a face that is actually blank.

The difference between this and a structured workflow is the difference between being the audience and being the editor. The editor sees the proof sheet. The editor sees the alternatives. The editor sees that the same face, placed beside a different image, produces a different meaning—and the editor chooses. The audience sees only the final cut and reads it as though it were the only possible version.

The Political Unconscious of the Single Output

Every narrative structure carries ideological weight. The three-act structure carries the weight of Aristotelian completeness—the assumption that stories should resolve, that conflict should produce catharsis, that the world is narratively legible. Save the Cat carries the weight of commercial genre fiction—the assumption that audience satisfaction is a structural obligation, that a story failing to deliver a recognizable emotional payoff has failed. The Hero’s Journey carries the weight of Joseph Campbell’s universalizing anthropology—the assumption that all meaningful narratives are variations on a single monomyth, which is also the assumption that all cultures tell stories the same way for the same reasons. None of these structures is neutral. Each is a specific argument about what stories are for, who they serve, and what the world looks like when it has been narratively organized.

A tool that lets the writer choose among these structures—and then modify, reject, and restructure the output—makes the ideological content of the structure visible and contestable. A tool that selects a structure automatically, based on the probabilistic patterns of its training data, and then hides that selection inside a smooth output, makes the ideological content invisible and natural. The first is a montage table. The second is a continuity edit. The first makes the cut available for argument. The second presents the cut as though no cut was made.

The political question is not whether AI should generate narratives. It is whether the structural decisions involved in generation are available for conscious intervention or hidden inside the output as naturalized sequence. The montage theorists understood that the cut is where ideology enters the image. The same is true of the generated text: the structure is where the ideology enters the narrative. A tool that exposes structure—through beat sheets, proof sheets, revision checkpoints, and iterative lock-and-regenerate workflows—does not eliminate ideology from the process. Nothing can. But it makes the ideology available for examination, which is the minimum condition for conscious authorship. A tool that hides structure inside a single output reproduces the oldest ideological trick in cinema: the invisible cut, the naturalized sequence, the story that presents itself as though no one made it.

If the Kuleshov experiment demonstrated that meaning is manufactured at the junction between images, then the question for any generative tool is: who controls the junction? In a one-shot generator, the model controls it. In a structured workflow, the writer does. The difference is not technical. It is the difference between being the spectator of a meaning someone else has manufactured and being the editor who manufactures it—knowing, as Kuleshov knew, that every juxtaposition is a claim, and every claim is a political act, whether the person making it acknowledges it or not.

What does it mean that the most widely used narrative generation tools are designed to hide the cut from the person using them? It means the tools have inherited an ideology older than any AI: the ideology that says structure is neutral, sequencing is natural, and the order of images requires no authorial justification. The Soviet montage theorists spent their careers arguing against this ideology. The question is whether we will recognize it when it returns, wearing a new interface, offering us a story as though no one made a choice.