The Next Dinosaur: CGI, AGI and Beyond

 

“We’re out of a job.”

“Don’t you mean extinct?”

Alan Grant and Ian Malcolm, Jurassic Park (1993).

PROLOGUE: A SUMMER NIGHT, 1993

The exchange between Alan Grant and Ian Malcolm occurs after the paleontologists have seen what John Hammond has accomplished. In the film, creatures that once could be reconstructed only through fragments of bone have returned to flesh and movement. Grant's profession, at least as he momentarily imagines it, has been overtaken by the thing he studies. Malcolm's reply carries an extra sting because something comparable was happening within the production itself. Phil Tippett had been enlisted for the movement of dinosaurs using go-motion, a painstaking descendant of stop-motion animation; computer-generated movement tests had convinced Spielberg to revise the plan. Tippett's reported reaction, a remark about becoming extinct, became part of the movie's mythology and was echoed in the dialogue. A conversation ostensibly about prehistoric animals was also a joke made by artists about the possible disappearance of their own methods. Few films have ever contained a more exact description of the industrial change they were helping to bring about. [17][41][42]

This is the double meaning of The Next Dinosaur, now widened by its subtitle CGI, AGI and Beyond. The phrase names the technological marvel still waiting for its audience, the new brachiosaurus that makes a whole cinema catch its breath. It also names the old technology about to become a fossil. In 1993, digital imagery seemed to be the future and model animation seemed vulnerable; in 2026, the established apparatus of sculpting, rigging, animating, simulating, lighting and rendering digital objects faces systems that can synthesise apparently finished moving pictures. The method that brought an extinct animal back to the screen may one day find itself described as an extinct method. Both interpretations belong in the title. The attraction is also a warning: the next dinosaur can be the newcomer and the creature it displaces, depending on which side of a technological threshold one happens to stand. No artistic medium changes as neatly as a family tree in a school textbook, but the irony is too exact to ignore.

It is Friday, 11 June 1993, and somewhere in America the house lights are fading in a cinema that still smells of warm projection equipment and buttered popcorn. On the screen there will be an island, a helicopter, a gate and a peculiar promise: that a creature extinct for tens of millions of years can be photographed as though it has never left the Earth. The spectators do not need to understand rendering, polygonal models, texture maps, articulated skeletons or the extraordinary arithmetic hidden in racks of Silicon Graphics workstations. They need only watch the first brachiosaurus lift its neck into the sunlight and see the astonishment pass across the faces of Alan Grant and Ellie Sattler. There is a second, less visible audience for the same spectacle: the makers of movies themselves, who have spent a century discovering what could be built in front of a lens and are now watching the boundaries of that physical world loosen. A new relationship between filmmaking and reality is being announced without a manifesto. The announcement takes the form of an animal looking alive. Industrial Light & Magic would later record that barely six minutes of the dinosaurs in Jurassic Park were computer-generated, alongside approximately nine minutes of practical puppetry. Six minutes were sufficient to rearrange the industry. [1][2]

The familiar legend makes this sound inevitable. Viewed from three decades later, Jurassic Park seems less like a gamble than the arrival of an era already waiting to begin. Yet nobody in that auditorium knows that feature films will eventually be filled with digitally manufactured cities, skies, creatures, armies, automobiles, oceans and people. Nobody knows that an artist who could once paint a landscape on glass will one day navigate a virtual landscape, and that a second generation of software will eventually offer to invent the landscape from learned representations rather than construct each surface through a traditional three-dimensional scene. In 1993, digital dinosaurs can still be described as an extraordinary trick. The turning point is partly technical, partly economic, and partly psychological. The screen has crossed a threshold at which the computer ceases to announce itself as a computer. What audiences can believe has altered. What directors can afford to imagine will alter next, though affordability itself remains a stubborn problem.

Now move the calendar forward thirty-three years, to October 2026. A short video generated by an artificial-intelligence model can present a convincingly photographed human figure, fluid camera movement, persuasive weather and lighting, an event that appears physically elaborate and an atmosphere that an effects department might once have spent weeks developing. The result can still break down spectacularly when asked to repeat a performance, preserve exact spatial relationships, meet a director's revision or sustain coherence across a sequence. Nevertheless, the old border is shifting again. Generative systems such as ByteDance's Seedance 2.0 have produced images capable of startling professional filmmakers while simultaneously provoking claims of copyright infringement and unauthorised exploitation of performers. In February 2026, film studios and the Motion Picture Association publicly challenged ByteDance over material generated with Seedance. Their objections are part of the history, not an incidental impediment to progress. The new technology has arrived with arguments about ownership built into its machinery. [3][4]

Somewhere in that argument sits James Cameron, the filmmaker who had already been central to the industry's first revolution. He helped turn a computer-generated water tentacle into a credible screen presence in The Abyss, made liquid metal terrifyingly expressive in Terminator 2, and later built Pandora into an environment whose blue-skinned inhabitants could sustain epic drama. In 2025 he discussed the need to accelerate effects work and reduce the costs of expensive computer-generated films; in 2026 he said his ambition for Avatar 4 and Avatar 5 was to accomplish them in half the time at two-thirds of the cost. His own words supply the second act of a long historical argument. His declared rejection of synthetic replacements for actors supplies its difficulty. The provocative prediction explored here is that Cameron, precisely because he understands the compromises between technology and performance, may be the filmmaker who gives generative visual effects their equivalent of a Jurassic Park moment. It is a prediction, not a claim that he has approved a generative-AI pipeline for the sequels. At the time of writing, the evidence does not establish such a decision. [5][6][7]

To understand what might happen on Pandora in the 2030s, it is necessary to remember what happened before that first cinema audience encountered a dinosaur. Every apparently sudden revolution has a prehistory. It begins with craftsmen cutting masks, tracing outlines, developing negatives, exposing one image through another, assembling tiny architectural worlds and asking whether the film camera can be persuaded to show an event that never took place. That is where the new argument properly begins, before the screen had become a computer's canvas and when the word digital still described a specialised department rather than a characteristic of almost every frame.

1933 TO 1993: SIXTY YEARS OF MOVING MONSTERS

Go back another sixty years from Spielberg's opening weekend, to 1933, and the effect has an ancestor standing on the Empire State Building. In Merian C. Cooper and Ernest B. Schoedsack's King Kong, Willis H. O'Brien and a small group of extraordinarily patient artists animated a metal-jointed ape, miniature prehistoric animals and miniature environments one exposed frame at a time. Photographic effects joined living actors to manufactured creatures through rear projection, glass paintings and other compositing methods. The scale of the labour is astonishing: PBS's history of visual effects reports that the stop-motion prehistoric creatures alone required approximately fifty-five weeks of animation. There were earlier pioneers, including O'Brien's dinosaurs in The Lost World (1925), so 1933 was a cultural summit, not the invention of stop-motion. Nevertheless, King Kong established a durable public image of how cinema could create life out of objects that had never lived. [44][45][46]

The tradition lasted through changing fashions, budgets and competing technologies. Ray Harryhausen, who had seen King Kong as a boy, became its most famous heir. He animated fantastic creatures in The 7th Voyage of Sinbad (1958), the fighting skeletons of Jason and the Argonauts (1963), and Medusa in Clash of the Titans (1981). Japanese monster cinema developed a parallel craft of performers in suits, miniatures and photographic effects; Hollywood likewise relied on animatronics, optical composites, mechanised creatures and makeup. It would be false to describe 1933 to 1993 as an uninterrupted monopoly by stop-motion. Yet for an audience encountering a giant imaginary animal, the route through a physical model and photographic illusion remained foundational across generations. Harryhausen's distinctive creatures were tactile constructions occupying space before a camera. The animators did not merely invent an image; they handled a body into a performance. [45][47][48]

This is the elegant symmetry concealed in the dates. King Kong helped establish the photographed model monster in 1933; Jurassic Park helped establish the physically convincing digital monster in 1993, exactly sixty years later. The thing on the screen was again a creature, and the revolution was again a method for granting that creature believable weight. Phil Tippett was, in this genealogy, a descendant of O'Brien and Harryhausen, carrying their physical knowledge into the late twentieth century. Spielberg's advance did not make those earlier films retroactively bad. It widened the available grammar of motion, reduced some limits on scale and perspective, and gradually redrew the industrial boundary between what was animated by hand and what was rendered through calculations. Stop-motion remained alive in specialist filmmaking, from the eerie poetry of The Nightmare Before Christmas to the modern work of Aardman and Laika. Technologies may lose a central commercial position without vanishing from art. [47][48]

Place 1933, 1993 and a possible 2033 alongside one another and a suggestive pattern appears: roughly every sixty years, or in this forecast forty years after CGI's blockbuster emergence, audiences may encounter a new instrument for imagining credible impossible life. The dates are an author's framework, not a technological law. There is no mechanism demanding a new form on a year ending in three. Their value is historical perspective. O'Brien worked with articulated objects and individual frames; Spielberg's effects collaborators used modelled geometry and numerical rendering; a future director may choreograph and constrain generated imagery through combinations of real performances, digital scene structures and learned image synthesis. Each era inherits visual problems that the previous craftsmen could already recognise: movement, presence, continuity, believable material and emotional conviction. What changes is how the workshop solves them.

I. WHEN WORDS AND WORLDS HAD WEIGHT

Before digital title tools became routine, even a deceptively simple opening credit might depend on physical artwork, a rostrum camera, precise registration and successive photographic exposures. Lettering could be drawn, printed, cut, photographed against high-contrast backgrounds, matted into photographed material or passed through an optical printer. Compositors combined photographic elements by protecting some portions of the image while exposing others; titles did not float onto film through a frictionless software interface. Effects teams worked with glass paintings, traveling mattes, miniature sets, rear projection, optical duplication and motion-control rigs. None of this means old cinema was literally more honest. Its tricks were elaborate and frequently invisible; its realism was manufactured by human hands, chemical processes, lenses, mirrors and extraordinary patience. What differs from the present is where the labour was stored. An optically composited shot preserved the traces of a sequence of exposures, while a present-day digital composite may preserve them as editable layers, nodes, tracked planes and procedural instructions.

Consider the titles of Alfred Hitchcock's Vertigo in 1958. Saul Bass's design and John Whitney's hypnotic spirals make a nonsense of the idea that cinema moved directly from handmade lettering to modern computer graphics in the 1990s. Whitney used a repurposed electromechanical analogue computing mechanism derived from military fire-control technology to guide abstract patterns, which were photographed and incorporated into the sequence. The effect was mathematical, mechanical and cinematic simultaneously. It would be inaccurate to treat it as a modern three-dimensional CGI sequence, but equally inaccurate to imagine that the computer entered film history only when a dinosaur ran across a screen. In Vertigo, calculated imagery helps express obsession and psychological instability; the technology earns its place through the emotional work it performs. Film historians can recognise this as an early meeting between graphic design and machine-controlled visual production, one whose lessons survived changes in hardware. [8][9]

The same distinction matters in Star Wars. The original 1977 trilogy used optical effects, miniatures, practical photography, computer-assisted design and meticulously photographed models in different combinations. The famous opening crawl was initially produced as physical artwork photographed with careful camera movement and perspective, rather than composed by an editor dragging a preset effect onto a timeline. Ken Ralston later described the difficulty of aligning a long piece of artwork, keeping the camera close to the surface and avoiding minute bumps in a shot that had to remain perfectly smooth. Its apparent simplicity depended on days of adjustment. The later prequel crawls could be generated digitally. Most moviegoers would not have attended a screening merely to admire the crawl, yet the change gives us a miniature history of a broader transition: once a difficult physical solution becomes a manageable digital instruction, artists begin devoting their attention elsewhere. The old work is not magically erased; the distribution of effort changes. [10]

Every development in effects also introduced an argument about legitimacy. A matte painting was wonderful when the viewer did not know it was there, yet a spectacular miniature could become an attraction in its own right. A practical dinosaur might have weight and surface interaction but be difficult to move at speed; a digitally animated dinosaur might move freely while struggling to reproduce convincing contact with the ground. Contemporary arguments about whether a computer-generated image is 'real' therefore inherit an old ambiguity. A motion-picture image has always been a constructed representation, even when the event was photographed. What audiences usually mean is that the world obeys enough coherent rules for them to stop examining the method. That standard applies with particular force to the coming generative era. A striking still image is an invitation; believable storytelling requires a sustained agreement between camera, movement, performance, environment and viewer. The 1990s supplied the first mainstream demonstration of how comprehensively that agreement could be renegotiated.

II. THE GLASS KNIGHT WHO CAME BEFORE THE DINOSAURS

The creature the industry should remember first is neither a dinosaur nor a terminator. It is a knight formed from the fragments of a stained-glass window, a hallucination in Barry Levinson's Young Sherlock Holmes, released in December 1985. Its sword emerges from a church window with all the impossible conviction of a nightmare. The knight is a landmark because it was the first fully computer-generated character integrated into a feature film, developed by the Lucasfilm Computer Division in collaboration with Industrial Light & Magic. John Lasseter, who would subsequently become central to Pixar's feature-animation history, was among the artists responsible. This is almost certainly the moment recalled by viewers who remember hearing about an early 'glass warrior' in an Indiana Jones-related production: the title was Young Sherlock Holmes, not The Young Indiana Jones Chronicles. The confusion is understandable. The films inhabited adjacent worlds of Spielberg-associated adventure, Lucasfilm effects and unusually ambitious practical-digital mixtures. [11][12]

There is a remarkable human modesty in that achievement. The artists were not primarily attempting to prove the computer's superiority over filmmaking; they were trying to solve a specific scene. Muren wanted the knight to feel dangerous and convincing, and he recognised that a hallucination permitted some abstraction. In the language of present-day AI debates, he had discovered a carefully bounded use case. He was not promising a fully digital cast, an entirely synthetic city or a production method that would make every craftsperson redundant. He selected an effect for which the available technology had an advantageous combination of novelty, controllable duration and narrative tolerance. The eerie translucency of glass, the discontinuity of the pieces and the scene's dreamlike premise helped conceal limitations that would have been more exposed in a human face or a mundane physical interaction. The present-day parallel with the isolated AI-assisted collapse in Netflix's The Eternaut is unusually close. In both cases, a new method enters a professional production through a specific visual problem that would be costly or difficult to solve another way. [11][13]

The knight also demonstrates how misleading 'the first' can be. Computer graphics had appeared in various forms before 1985. The Star Trek II: The Wrath of Khan Genesis demonstration in 1982 is generally identified as a pioneering all-digital cinematic sequence. Tron and The Last Starfighter placed large amounts of computer-generated imagery before movie audiences in the early 1980s, while earlier works explored computer images and simulated points of view. Some were overtly electronic, technological or stylised in their narrative purpose. The knight was a different technical proposition: a mobile animated figure that belonged inside the live-action photographic space. The transition from graphic spectacle to an object that appears to share the camera's world is the core of the analogy. A future generative effects 'first' may likewise be preceded by many films that used AI in credits, background plates, concept art, restoration, clean-up or isolated shots. Its actual claim to historical importance will depend on what becomes newly credible to the audience. [14][15]

The standard popular history then moves through James Cameron. The Abyss in 1989 used a computer-created, water-like pseudopod whose surface appeared to reflect its environment; Terminator 2: Judgment Day in 1991 integrated a liquid-metal antagonist capable of morphing through forms that practical fabrication could not plausibly achieve at the same scale and speed. Cameron's achievement was to insist that those effects behave dramatically. The water tentacle is a visitor exploring human faces and gestures; the T-1000's fluidity conveys an implacable predator that defeats ordinary physical resistance. Both belonged to stories whose premises made their digital qualities meaningful. They were not yet a blueprint for rendering an entire ordinary world, but they helped demonstrate that digital visual effects could become a storytelling language rather than a technical insert. Even the T-1000 was a hybrid: Robert Patrick's actual performance, prosthetics and mechanical effects remained essential. That mixture foreshadowed Jurassic Park more accurately than the idea of an instant replacement of analogue craft by software. [16][17]

III. THE SIX-MINUTE REVOLUTION

The making of Jurassic Park supplies one of the most useful cautionary stories in technological history because the revolution was, at first, an internal competition. Spielberg assembled an extraordinary effects team, including Stan Winston's practical creature builders, Phil Tippett's animation expertise and ILM's digital specialists. The assumed workflow involved full-scale animatronic dinosaurs for close encounters and techniques descended from stop-motion, including go-motion, for creatures that had to run or attack in wide shots. Tippett understood movement in bodies composed of weight, joints, inertia and intention. He was a master of a world in which an animator moved a model and photographed the result. Computer graphics offered a path to unrestricted movement and camera placement, but early digital creatures were not automatically convincing. Their success depended on matching a cinematographer's visual world and the accumulated observations of people who knew how animals moved. [1][2][18]

The change came when ILM artists, including Steve 'Spaz' Williams and Mark Dippé, produced tests that persuaded Spielberg the digital creatures could sustain the film. Anecdotes about Tippett declaring himself extinct have entered Hollywood folklore; the paradox is that the practical-animation specialist's experience remained indispensable. Tippett and his colleagues helped bring performance and naturalistic motion into the new process. ILM developed a Dinosaur Input Device that could transfer movements from an articulated armature into a computer animation system. Here was the transition in miniature: an expert who feared that the computer had removed his medium discovered that his understanding of movement could direct the computer. The image on screen changed its technical origin; the knowledge required to make it persuasive retained deep historical continuity. Any account of AI that ignores this distinction confuses the displacement of a technique with the disappearance of its practitioners' intelligence. [2][18]

The most astonishing number remains six. ILM states that Jurassic Park used approximately six minutes of computer-generated dinosaurs and nine minutes of practical dinosaurs. This was not a digital film disguised as a physical one. Its success depended on placing the right method in the right shot. The immense T. rex animatronic delivered tactile presence and interaction; digital animation allowed a creature to move in ways no practical model could conveniently reproduce. Cinematography, lighting, sound, editing and dramatic delay made individual images feel larger than their screen time. The film's famous brachiosaurus reveal depends on the actors looking and reacting before the creature dominates the frame. The method becomes historically decisive because it is subordinated to an emotional event. Audiences remember wonder, appetite, fear and awe rather than the length of a render queue. [1]

The lesson for generative visual effects is straightforward but easy to miss. The next Jurassic Park need not contain hours of AI-generated material. It may require a sequence of the right length, dramatic clarity and unimpeachable quality, deployed in a film of sufficient cultural reach that viewers and filmmakers can see a new method crossing the old practical boundary. The first genuinely decisive AI moment might be a single complex action sequence that would previously have demanded a long pipeline of modelling, simulation, animation, lighting, rendering and compositing. Its claim to novelty would be strongest if the shot remained individually directable: camera path, environment, light, action, continuity, surface detail, emotional timing and interaction all independently adjustable without having to gamble that a fresh generation would preserve everything the director liked about the previous one. The breakthrough is not that a machine can make a beautiful image; machines have done that for decades. It is that a machine can reliably participate in making the precise sequence a filmmaker wants, at a cost and speed that change what productions attempt.

What matters about Tippett's story is the interval between the fearful phrase and the revised method. Popular retellings can make his remark sound like a terminal verdict, a craft worker seeing a screen test and leaving the building forever. In fact, the Dinosaur Input Device embodied an unusually concrete compromise. Animators manipulated a physical creature armature fitted with encoders, making an expertise developed through direct handling available to the newer digital production. The device did not automatically solve every animation problem, and computer animators still needed to adjust performances, but it gave an evolving industry a grammar in which old knowledge could remain audible. That is a more demanding historical precedent than the comforting cliché that everyone will simply learn new tools. It suggests that established artists need access to the development of the new workflow, credit for the knowledge it incorporates and credible routes into the jobs that emerge. The next generation of Tippett-style specialists may understand physical motion, audience perception and narrative timing better than an image model trained on the visual results of their predecessors. Whether the industry allows that knowledge to survive inside its new machinery will be a matter of institutions and choices, not an automatic gift of technical progress. [17][42]

IV. WHAT THE 1990S ACTUALLY ARGUED ABOUT

Our picture of the early CGI era is distorted by the knowledge that it won. There was enthusiasm and astonishment, but there was also criticism about whether digital spectacle could crowd out drama, whether synthetic imagery would feel insubstantial and whether new production methods would marginalise established crafts. Those anxieties were expressed in studio meetings, specialist magazines, trade journalism, reviews and conversations among effects artists long before their present-day equivalents could spread through instantly searchable online platforms. Yet it would be wrong to write that the 1990s possessed no internet debate. Usenet groups, bulletin boards, early websites, newsletters and fan communities existed; they simply did not operate with the current speed, scale, recommendation algorithms or capacity to turn a single image into a global labour controversy within hours. The channels through which technological criticism becomes socially visible have changed as much as the tools themselves.

One difference is that early CGI could be framed as the arrival of a difficult, specialised trade. Artists at ILM and other facilities were learning complex software, creating assets and developing methods requiring considerable expertise. Their work was laborious, expensive and scarce. It could threaten older specialties while offering an image of technical craft in its own right. A public demonstration of an AI video model in 2026 often arrives in the form of a user entering a brief prompt and receiving a detailed, photorealistic result. Whether that demonstration hides a larger network of artists, engineers, training-data producers and industrial infrastructure is often invisible to viewers. Its rhetoric is radically different. The promise sounds like the removal of work rather than the introduction of new specialised work. That is why the two eras can inspire similar arguments about authenticity while producing different levels of professional fear. The perceived ease of the interface is central to the modern dispute.

The original reporting makes the parallel more concrete. On 6 June 1993, five days before Jurassic Park opened widely in American cinemas, the Los Angeles Times published a long account of Spielberg's dinosaur craftsmen. Its attention was divided between the extraordinary promise of digital creatures and the stubborn physical expertise of the practical team. Stan Winston's animatronics and the new computer-generated animals were described as complementary parts of one illusion, while the artists had studied real animal motion and anatomy to make their extinct subjects persuasive. This is a valuable correction to the notion that contemporary Hollywood greeted CGI with a single organised campaign of hostility. The immediate public story included wonder, technical curiosity and enormous commercial expectation. Yet a subtler threat was already present for specialists whose skills were bound to specific tools. A motion-picture profession could celebrate its newly expanded powers while individual artists worried about what might happen to their working lives. The contradiction is familiar to anyone following generative video in 2026. [36]

A second contemporary voice is especially useful because it came from outside the effects business. In August 1993, the evolutionary biologist and essayist Stephen Jay Gould published 'Dinomania' in The New York Review of Books. Gould approached Jurassic Park with a scientist's interest in reconstructed creatures and a historian's sense that popular entertainment could accelerate tools with wider consequences. He described the enormous advance in digital reconstruction while acknowledging the older techniques that had made the film possible. This is not a record of a unanimous early-1990s rejection of computer imagery; it is evidence that sophisticated observers understood the development as a potentially expansive new kind of representation. Present-day comparisons should therefore avoid manufacturing a lost age of uniform resistance. The distribution of enthusiasm, scepticism and anxiety differed, partly because the communications environment differed and partly because the apparent relationship between technological progress and human craft was different. [37]

Historian Julie A. Turnock offers a useful alternative to the folklore of effortless progress. In The Empire of Effects she traces how ILM helped produce the dominant aesthetic of digital realism through techniques inherited from live-action cinematography: imperfections of lenses, irregular camera behaviour, atmospheric clues and framing that seems contingent rather than mathematically pristine. A digital image looks credible partly because it imitates evidence a real camera would leave behind. That analysis corrects the assumption that increasing geometric detail automatically creates realism. The effect must participate in an entire photographic language. It also illuminates why generative models can seem so startling. By learning statistical regularities across large bodies of moving imagery, they may reproduce not only an object's appearance but also the ancillary signatures of cinematography. Yet imitation of a camera's look is not the same as a robust model of spatial geometry or controllable dramatic action. A plausible single shot can conceal precisely the weaknesses that surface when two shots must join seamlessly. [19]

The argument over 'soulless CGI' became especially complicated as digital imagery ceased to be exceptional. The prequels to Star Wars made digital environments and creatures essential to the franchise's look; superhero films later expanded that practice to enormous scale. Some viewers came to prefer the texture of photographed models, while others embraced worlds that physical production could never have constructed. The change did not yield one settled aesthetic judgement. It created options, incentives and a new baseline. Something similar could happen with AI. A 2027 film may be condemned for a visibly generated creature that would be entirely unremarkable in 2034, just as a digital effect praised for its novelty in one decade can appear crude in the next. Conversely, some generated material may age faster than older practical effects because its patterns become recognisable as the signature of a particular model family. The history of visual effects offers reasons for optimism about innovation and reasons for scepticism about every prediction of inevitable aesthetic improvement.

V. PANDORA AND THE SECOND DIGITAL REVOLUTION

When Avatar reached cinemas in 2009, Cameron made another bet against conventional expectations. The industry had already accepted computer-generated creatures, but his ambition was to make a large proportion of a feature's emotionally important action occur inside a computer-generated world without sacrificing the presence of performers. Pandora was not merely a painted setting behind live actors. Its forest, vegetation, mountains, skies, flying animals and humanoid inhabitants were integrated into a filmmaking system in which actors' performances could be captured and interpreted as digital characters. The director's virtual camera, performance-capture techniques, stereoscopic photography and Weta's evolving effects pipeline helped make the technological programme distinctive. The selling point was not simply that pixels were replacing scenery. It was that a cinematic world could be inhabited, directed and photographed as though the director had acquired access to an impossible location. Cameron's recurring engineering instinct was visible again: he demanded a system that enabled him to make the film he wanted, then organised the production around inventing or adapting that system. [20]

The personal history matters because it gives the generative-AI prediction greater seriousness than the ordinary assertion that a famous director will adopt whatever is fashionable. Cameron repeatedly approaches visual-effects problems through the grammar of photography. In The Abyss, the watery pseudopod needed to occupy existing live-action environments and respond to human faces. In Terminator 2, the T-1000 had to persuade audiences that one actor and an impossible material were part of the same character. In Avatar, a performer's intention needed to survive translation into a creature with different anatomy and a world with different physics. Each project solved a problem of continuity between the photographed, the fabricated and the imagined. The unresolved technical question for generative systems is precisely one of continuity. It is difficult enough to synthesise a convincing moving image; it is harder to ensure that an actor's smallest shift of expression, a necklace, a blade of grass, a shadow and the camera's route through space remain coherent through a hundred revisions. Cameron's expertise could be valuable because he knows which imperfections the viewer forgives and which break a scene's reality.

The 2022 sequel, Avatar: The Way of Water, extended that ambition into underwater performance capture and especially demanding interactions among water, skin, hair, fabric, movement and light. Water has long exposed the weaknesses of visual-effects pipelines. It refracts and reflects, shifts unpredictably and produces fine-scale motion across countless interacting surfaces. The problem is not solved by a still image that looks wet. An underwater dramatic sequence must coordinate the breathing, swimming and movement of characters with camera placement, water volume, particulate matter, lighting, contact and continuity. Cameron's work illustrated the scale of bespoke engineering still required decades after Jurassic Park. The rise of generative AI does not retroactively make those difficulties trivial. Its advantages become truly valuable when coupled with geometric, physical and performance constraints, the disciplines that distinguish production systems from visual demonstrations. [21]

Wētā FX's own production accounting makes the scale of the resulting challenge almost tangible. Of 3,289 shots in the finished Way of Water, the company says it worked on 3,240 visual-effects shots, including 2,225 water shots, while only two shots in the entire film lacked visual effects of any kind. Its team describes a newly developed water-simulation toolset, sophisticated interaction between water and skin or hair, and pipelines devised to accommodate the performances shot above and below the waterline. Those numbers matter when considering Cameron's ambition for the next films. Any new generative technique would have to survive a manufacturing environment containing thousands of mutually related images, decisions and editorial revisions. Saving minutes on a single impressive clip would matter much less than making a repeatable workflow modestly more efficient across thousands of shots. [21]

By the time Avatar: Fire and Ash arrived in December 2025, the franchise had accumulated a formidable technical history and an equally formidable economic burden. Disney reported that the film passed one billion dollars worldwide by early January 2026; subsequent industry reporting placed its eventual global receipts around $1.4 billion to $1.5 billion, an extraordinary sum outside the peculiar financial universe occupied by the biggest blockbusters. Yet that haul was appreciably below the approximately $2.3 billion associated with The Way of Water. In April 2026, producer Rae Sanchini described development of the fourth and fifth instalments as moving ahead, while stressing that the announced release dates were not absolutely fixed. The economic challenge is therefore real even though apocalyptic descriptions of the franchise's commercial situation would be excessive. A series can remain one of the largest box-office properties in history and still confront a model of production that is difficult to sustain at its preferred scale. [22][23]

That helps explain why a proposal to finish future films more quickly and cheaply does not necessarily signal Cameron's loss of interest. Film production is not a contest in which artistic devotion is measured by how much money can be consumed by the end credits. A director can love the work and still regard its resource demands as an obstacle to further work. Indeed, the history of effects is full of breakthroughs motivated by artists refusing to accept that an image must be made at an impossible price. The initial use of go-motion, motion control, digital compositing, crowd simulation and virtual production can all be understood partly as attempts to move the boundary between desired images and available resources. In this sense, Cameron's 2026 position follows his career rather than contradicts it. The risk is that cheaper production can invite larger ambitions until apparent savings disappear. The opportunity is that an artist might spend less time waiting for iterations and more time deciding what the scene should accomplish.

Cameron's relationship to AI is also more discriminating than either celebrants or critics sometimes acknowledge. In September 2024, he joined the board of Stability AI, saying he wanted to understand developers' thinking and how their methods might fit visual-effects workflows. In a 2025 conversation with Meta's Andrew Bosworth, he connected the technology to a desire to reduce the cost of major effects-heavy films. He explained his ambition in terms of improving artists' throughput rather than eliminating half of the workforce. Later that year he stressed that generative AI had not been used on Avatar: Fire and Ash and expressed strong opposition to synthetic actors displacing human performance. These statements can coexist. A director can want generative tools for secondary imagery, iterations, procedural assistance or technical operations while insisting that an actor's performance originates in an actor. Treating every form of machine learning, procedural animation, photogrammetry and prompt-based synthesis as the same thing only obscures the underlying argument. [5][6][7]

The future of Pandora may therefore lie in a layered pipeline: photographed and captured human performance at its centre, established three-dimensional geometry and simulations providing spatial discipline, and generative methods accelerating specific operations around them. Perhaps the largest prospective gains will arise in areas that rarely feature in trailers: consistent environmental extensions, the preparation of lighting variations, secondary animation, detail synthesis, cleanup, rotoscoping, texture development and compositing. The comparison with early CGI is illuminating here. The visible revolution might be announced by a spectacular generated environment, while the actual transformation occurs in thousands of ordinary decisions no spectator can identify. A better, faster process might leave no unmistakable stylistic signature at all. For a franchise whose appeal depends on audiences believing in an entire world, that invisibility could be a defining virtue.

VI. OCTOBER 2026: AN IMAGINED ROOM AT LIGHTSTORM

Let us imagine, carefully and openly, a room in James Cameron's offices in October 2026. This is not a description obtained from a visitor or an anonymous employee, and no dialogue that follows should be attributed to Cameron. Its purpose is to make an industrial decision imaginable. There is a sequence from the fourth Avatar film on a screen, perhaps a riverbank at twilight or the interior of a settlement we have not yet visited. One version contains the performance-capture data that already grounds the scene; another carries temporary animation and lighting; another explores an environmental treatment. A producer wants to know what the sequence would cost under established methods. An artist wants to know whether the proposed new method can be revised after the director changes the rhythm of a character's movement. An engineer wants to know whether the model is secure, licensed, versionable and reliably aligned with the actual scene. The hypothetical argument begins not with an artist asking an AI to make Pandora, but with a filmmaker insisting that a particular shot be finished without sacrificing control.

The numbers on the imagined production wall are grounded in what Cameron publicly said. Speaking on The Empire Film Podcast in 2026, he explained that Avatar 4 and 5 remained among the projects he was considering and that his team would examine technologies capable of making them more efficiently. His formulation was half the time for two-thirds of the cost, with roughly a year needed to determine how to achieve the goal. The quotation matters because the two measures are distinct. A reduction to two-thirds of current costs means saving about a third, not half. Reducing duration by half is a much more radical acceleration, and simply doubling the number of artists would be an expensive and often inefficient way to attempt it. He did not announce that a particular proprietary AI model would make the films, and he did not say that the remaining sequels would be generated by prompts instead of acted. Reading an AI production announcement into that interview would distort the evidence. Reading no technological significance into it would ignore his separately stated interest in generative effects workflows. [5][6]

Our imagined office would have to confront a question familiar to anyone who has supervised complex creative work: what does 'faster' mean after revisions? A model that produces an attractive first pass in seconds could be slower overall if it takes weeks to preserve the details a director approved while changing only a hand gesture or the direction of an animal's gaze. A conventional three-dimensional asset is costly to build but remains available to be placed, lit, rotated, animated and rendered repeatedly. A free-form video generator may have low initial friction but poor editability. To earn its place inside the pipeline, it has to make revisions cheaper rather than simply make first attempts spectacular. It must be possible to request the fifth variation of a shot, with the previous four still reproducible, while an effects supervisor knows exactly what has changed. In the absence of these guarantees, a celebrated speed increase may be a demo-room illusion.

An artist at the table might defend the conventional pipeline by pointing to its accumulated discipline. A production owns digital models of characters, creatures, machinery and environments and can send them through many departments. A scene can be blocked, captured, animated, simulated, lit and composited. Work remains attributable to particular people and steps. Another artist might counter that this infrastructure carries enormous overhead, particularly when rendering fine details in environments that must appear dense with life. Generative assistance could offer a way to keep the 3D foundation while replacing some of its most repetitive labour. These positions need not be enemies. In the most credible version of the future, the effects supervisor's job becomes deciding which constraints must be exact and which visual detail may be proposed by a learned model. The pipeline evolves according to the nature of each shot rather than by decree that every department has been replaced.

There is a moral argument in the room too. If a model can generate the appearance of an actor, who authorises that use? If it learns from images protected by copyright, what licensed material entered its training and who is compensated? If a facility's private assets are used to fine-tune a system, can they leak through later generations? If junior artists once learned through tasks the generator now performs, where do they acquire their judgement? If a production begins to depend on a system controlled by another company, what happens when prices rise, terms change or a model disappears? These questions are not abstract complaints from people afraid of change. They are part of whether the technology can support the trust, repeatability and contractual certainty on which large-scale filmmaking depends. The studios' public dispute with Seedance 2.0 demonstrates that cinematic quality alone will not settle the matter. [3][4]

Perhaps the imagined discussion ends with a deliberately modest instruction. Keep the performance; keep the geometry; keep the camera; test a generative process on a controlled portion of the environment; measure the total human work required; compare the result with a conventional render; document every source; repeat under revision. In that hypothetical sequence of decisions lies a plausible path to a genuine breakthrough. Cameron's historical habit has been to demand that technology serve difficult, concrete filmmaking objectives. If he eventually uses generative methods on a later Avatar feature, he may do so as an engineer of controlled experiments rather than as an evangelist for a single fashionable interface. But there remains another possibility: the proposed savings arrive through improved traditional VFX tools, machine-learning assistance outside the controversial category of generative video, reorganised scheduling or a combination of these changes. His stated target does not by itself establish which path will prevail.

VII. THE ETERNAUT: THE GLASS KNIGHT OF THE NEW ERA?

A useful early scene in this new chronology is not from a blockbuster at all. In the Argentine Netflix science-fiction series The Eternaut, a building collapses in Buenos Aires. The image belongs to a show whose broader dramatic premise concerns an extraordinary catastrophe experienced through particular people and streets. In July 2025, Netflix co-chief executive Ted Sarandos identified that collapse as the company's first generative-AI final footage in an original production. The production worked with Netflix's own specialists, and Sarandos said the scene had been finished approximately ten times faster than with traditional effects and at a reduced cost. Those are company claims rather than an independent published cost audit, but the announcement was historic because it attached a real production, a delivered image and a concrete efficiency comparison to a technology generally discussed through prototypes. A shot had escaped the demonstration reel and entered a finished mainstream programme. [13][24]

The parallel to the 1985 glass knight is useful precisely because neither shot established full technical maturity. The knight did not prove that a computer could render a convincing human performance through an entire feature. The building collapse did not prove that a generative model could direct a cast of persistent digital characters across a continuous ninety-minute film. Each showed that a carefully defined task had crossed the line into professional use. The production economics, however, were reversed. The knight was an expensive experiment almost justified by the knowledge it generated; Netflix promoted the collapse partly because it could make a visually ambitious moment affordable for the show. There is a new industrial story in that reversal. Generative tools may reach the production mainstream through budgets that previously could not accommodate elaborate effects. The techniques may then migrate into spectacular blockbusters after workflows mature, exactly as isolated CGI experiments migrated into the enormous cinema of the 1990s.

Audience reaction offers another lesson. The Eternaut enjoyed strong international viewing and critical appreciation; its AI-assisted effect did not overwhelm discussion of the series. That outcome should not be overstated as a referendum in favour of generative AI, since viewers may not have known how the shot was made or particularly cared once absorbed by the narrative. It is nevertheless evidence that an AI-generated element can pass through a popular work without destroying audience trust. Viewers who express principled objections to AI in entertainment may still regard a particular series as excellent, and viewers who like generative effects may dislike the work that contains them. Reception has several causes. A scene's technique becomes culturally controversial when the manner of its production acquires meaning outside the story, especially when unions, artists, companies and platforms turn it into a public example of a disputed practice.

The opposite case is Marvel's Secret Invasion in 2023. Its opening title sequence employed generative AI and immediately drew criticism from artists and commentators. The controversy concerned the choice to make synthetic imagery an expressive introduction to a major entertainment property at a time when Hollywood creative workers were already preoccupied with digital replacement. Even before an audience reaches the drama itself, title design announces the film's identity, and the credit sequence carried an unintended additional message about how the studio valued certain kinds of creative labour. The series' broader critical reception was poor for reasons extending well beyond that title sequence, so it would be misleading to treat low audience ratings as evidence of an AI boycott. The more precise lesson is that an image can be technically appropriate to a thriller about mistrust and imitation while politically disastrous as a public symbol of automated creative work. [25]

This is why the question 'Who will be first?' demands at least three answers. The first person to make an entirely AI-generated feature may be an independent filmmaker with a tiny budget. The first studio to integrate generative imagery reliably into a mainstream work may be a streamer or a mid-budget producer. The first director to create a globally recognised equivalent of the Jurassic Park moment may be making a large theatrical feature whose success cannot be ignored. History tends to compress these distinct achievements into the name of a single movie. In 2026, the achievements remain separated. That separation is a temporary state, not necessarily an indication that any one company controls the next stage of cinema.

For an editor encountering the new footage inside a working post-production studio, the test is less theatrical than a viral clip. Can the software return to the same collapsed building a week later, after a director changes the framing? Can its debris be held in exactly the same relationship to a photographed actor? Can a supervisor receive a change to only the dust while preserving the lighting on every other element? The questions expose a paradox of abundance: when finished-looking footage becomes easier to generate, the exact image a filmmaker wants may remain extraordinarily difficult to secure. A traditional CGI workflow can be cumbersome precisely because it stores innumerable decisions in editable and inspectable forms. Generative systems will become revolutionary for major features when they preserve those production virtues and remove enough of the mechanical labour to create substantial gains. That possibility is exciting, but it remains a standard to be proven by completed work rather than a promise demonstrated through selected clips.

The calendar supplies a smaller coincidence, one that is appealing precisely because it should not be mistaken for proof. In 1985, Young Sherlock Holmes offered an early entirely digital character inside a conventionally produced film. In 2025, Netflix's The Eternaut demonstrated a limited use of generative AI in its finished effects, with a collapsing building that the company said was produced much more quickly than a traditional alternative. Two years ending in five, forty years apart, each attached to a bounded effect rather than a whole new production grammar. The pairing resembles a pair of laboratory doors opening onto different technical eras. It does not mean that the stained-glass knight and the Buenos Aires collapse were comparable achievements in every respect, nor that the next landmark must wait eight years as the digital dinosaurs did. It gives the essay a memorable hinge: 1985 to 1993, and perhaps 2025 to 2033. The second interval is a forecast awaiting a film, not a date entered in a studio calendar. [11][12][24]

VIII. THE SEEDANCE EVENT, AND THE TERRIBLE POWER OF A DEMO

ByteDance's Seedance 2.0 gave the industry a reminder that the next revolution will not be a purely aesthetic succession. Released in February 2026, the model could generate short cinematic clips from text and other inputs, and examples depicting recognisable performers and studio characters circulated rapidly. Among the much-discussed clips was an AI-generated action scene portraying likenesses of recognisable Hollywood actors. The effect of watching a professional-looking scene apparently conjured from a minimal written instruction was startling enough that industry writers publicly expressed anxiety about their livelihoods. Yet the model's demonstration of audiovisual competence coincided almost immediately with warnings and legal demands from studios whose characters appeared in generated examples without authorisation. Disney, Paramount Skydance and other major companies objected, and the Motion Picture Association pressed ByteDance to curb allegedly infringing activity. ByteDance said it would strengthen safeguards. [3][4][26]

We should distinguish the claims carefully. A generated video resembling a Hollywood character does not, by itself, disclose precisely what copyrighted works were present in training data, nor does a cease-and-desist letter establish a final judicial determination. The studios alleged infringement and unauthorised use; the company described efforts to limit misuse. The underlying ethical questions remain substantial even while legal conclusions depend on jurisdiction, evidence and future decisions. More broadly, a visually successful result created from an unlicensed likeness is not a legitimate production method simply because it looks convincing. A film studio trying to use generative imagery in a released feature must account for whether its models, reference assets and performer rights are usable under binding agreements. The problem is as practical as it is legal. A shot that cannot be distributed without risk is not a cheaper shot, however fast it was produced.

The famous viral AI clip is also an unreliable standard for assessing cinematic production. Short videos are selected for maximum persuasive impact, and failures are rarely promoted with the same enthusiasm as successes. A model may generate dramatic movement and a plausible camera path but struggle with exact continuity, geography or detailed instructions over several takes. A human viewer may accept a one-off encounter in a fifteen-second clip because there is no need to recognise a costume's stitching from the previous scene. A feature film is a sequence of promises. The blade held in a character's right hand must remain the same blade; a wound must appear in the same place; the environment must be navigable; a facial expression must respond to the performance of another actor; the dramatic rhythm must survive the cut. Editing conceals many discontinuities in conventional cinema, but it cannot rescue a production built on uncontrollable variations in every asset.

The enthusiasm provoked by Seedance nevertheless matters, even for artists who recoil from its training and rights controversies. The clips make a capability visible. The same happened when Spielberg saw the first digital dinosaur tests, when Cameron assessed the water pseudopod or when early audiences watched a knight step out of glass. The demonstration lowers psychological resistance to the idea that an effect is possible. It does not remove engineering constraints, but it changes the questions people ask. Instead of whether a model can make a cinematically credible event, the discussion turns to what it would take to make that event controllable, authorised, scalable, accountable and artistically original. For the first time in the current transition, the scale of Hollywood's hostile reaction demonstrates something close to inverse recognition: companies are treating the tool as consequential enough to threaten valuable intellectual property.

A separate commercial force is emerging alongside the conflict. In June 2026, Lionsgate and Runway expanded their AI partnership, with Lionsgate taking an equity stake and the companies announcing a joint development programme. Initial projects were expected to include short-form episodic material. Netflix had already moved to acquire Ben Affleck's InterPositive, a filmmaking technology business whose approach emphasised integration with creative workflows. These strategies are not the same as a published plan to make a full-length franchise movie with generative imagery, but they indicate that established companies are testing the infrastructure through which generative tools might become normal production equipment. The decisive film may arrive as a downstream consequence of corporate investment that scarcely attracts public attention while it is occurring. [27][28]

IX. A STAR WARS GALAXY AT ANOTHER CROSSROADS

There may be no film franchise better equipped than Star Wars to illustrate the continuity between practical illusion, digital transformation and an impending generative phase. The original film in 1977 helped make an effects-intensive cinematic world into a modern blockbuster model, drawing on photographic miniatures, elaborate composites and motion-control technology. The prequel trilogy that began in 1999 made digital creatures, environments, extensive computer animation and eventually digital cinematography conspicuous ingredients of the production. The sequel trilogy from 2015 to 2019 consciously recaptured aspects of the original films' tactile design and frequently publicised practical sets and creatures while remaining, of course, major users of digital effects. Its reception included disagreements over narrative planning, character development, continuity and artistic direction, not merely disputes over which techniques appeared on screen. To claim that the sequels contributed nothing technologically would be too sweeping; to recognise that the prequels are historically associated with a particularly dramatic production transformation is reasonable. [29][30]

Ahmed Best's performance as Jar Jar Binks remains instructive. The character was entirely computer-generated on screen, but his existence depended on performance reference, motion capture, animators and extensive keyframed facial work. In a Lucasfilm retrospective, animation supervisor Rob Coleman emphasised the extent to which the animators built on Best's work and the degree to which George Lucas treated the digital character as a participant in the scene rather than an object requiring continual exhibition. Jar Jar inspired intense audience disagreement, but the character also exemplified a question that would later define Cameron's work: how does a human performance travel into a synthetic body? The notion of a 'digital actor' is often carelessly used to suggest that computers replace actors. In both the prequels and Avatar, a great deal of the emotional content began with an actual person. A future Star Wars production could use generative tools while preserving precisely that performance lineage. [30]

The contemporary slate makes the comparison especially inviting, but it is necessary to distinguish established releases from projects in development. By October 2026, the franchise had multiple announced and reported feature-film plans, including Star Wars: Starfighter for 2027 and a proposed new trilogy being developed by Simon Kinberg with recent reporting about Jon Watts's involvement. Not every reported project had been fully approved for production; other films associated with Rey and various filmmakers remained at different stages of development. No reliable public evidence establishes that the proposed new trilogy is being built around generative VFX. Nevertheless, its timing places it near the likely period when professional AI systems become more directable. A film series capable of committing enormous resources to visual invention could choose to make that transition conspicuous, or it could choose to shelter behind the perceived authenticity of physical effects. Both outcomes would fit the franchise's history. [31]

A compelling new Star Wars film might offer a visual challenge analogous to the brachiosaurus: an immense populated city whose thousands of inhabitants behave consistently, an aerial battle directed with precision across a continuous geography, or a creature performance with subtle emotional range that conventional animation would make prohibitively laborious. Yet there is a peculiar problem with a franchise that has already shown audiences nearly every imaginable kind of spectacle. If a generated spaceship looks like the spaceship a renderer could have produced ten years earlier, viewers may have no reason to feel that a threshold has been crossed. Its industrial novelty may be invisible. A director would have to use the efficiency of the new tool to attempt something distinctive in scale, intimacy or duration. The question becomes artistic: what image has the filmmaker not been able to pursue because the preceding method was too expensive, too slow or too inflexible? A technological transition that merely provides familiar images for less money may change Hollywood's economics without generating a moment audiences remember.

The commercial and moral obstacles are pronounced. The Star Wars brand depends on distinctive designs, characters, performers and decades of fan investment. Disney's objections to unauthorised Seedance-generated depictions of Marvel and Star Wars material show that it cannot simply ignore questions about ownership when considering generative tools for its own productions. A licensed, protected, internally governed system would be more plausible than indiscriminately uploading proprietary characters and production assets into a publicly accessible generator. The franchise could experiment with models trained on authorised materials and integrated with ILM's existing asset management, but that is an engineering possibility, not a disclosed plan. It would also need to explain how such systems respect the rights and work of actors and artists. For the company, the issue is not just whether an image can be generated. It is whether a generative process can be made as reliable, auditable and commercially defensible as the effects department it supplements.

There is an appealing historical symmetry in the possibility that Lucasfilm might lead another revolution. The Lucasfilm Computer Division helped create the stained-glass knight in 1985, ILM gave Spielberg the dinosaurs in 1993, and the Star Wars prequels transformed expectations for digital environments and characters. George Lucas's institutions have repeatedly supplied the techniques that other filmmakers then made famous. Cameron himself profited from their accomplishments and pushed their possibilities further. Yet institutional prestige offers no guarantee of another breakthrough. An incumbent can be slowed by established pipelines, audience expectations and the cost of risking valuable characters. The next leap could be made by a smaller production willing to rebuild its workflow. If the decisive theatrical breakthrough ultimately comes from Star Wars, the strongest historical explanation may not be that its owners guessed correctly about a video model. It may be that a facility such as ILM finally turned generative methods into a disciplined production art.

X. FROM CGI TO AGI: A NAME FOR THE FRONTIER

It is worth pausing over the initials in the title. Computer-generated imagery became CGI, a convenient umbrella for many methods and production eras. For this essay I propose a parallel expression: AGI, Artificially Generated Imagery. It is a deliberate piece of critical wordplay, not a claim that the visual-effects industry has adopted those initials or that AI-generated images must replace every conventional computer graphic. The established professional language in 2026 is generally generative AI, generative video, or generative VFX. The initials AGI already have a different, widely used meaning in computing: artificial general intelligence, the much-debated prospect of machines capable of flexible human-level performance across a broad range of cognitive tasks. That collision of meanings is intentional. The proposed AGI of the image belongs to the present discussion of cinema; the other AGI belongs to a much larger discussion about intelligence, agency and the organisation of human work. The two futures may influence one another, but neither should be casually substituted for the other. [43][49]

The double reading makes the title more expansive without falsifying the terminology. The industry's cameras once recorded miniatures and people, then increasingly recorded nothing at all while digital systems constructed the image. Artificially Generated Imagery might one day describe a recognisable new layer of that evolution, even though CGI remains technically correct for computer-created imagery generally. A production could use hand-animated characters, simulation, machine-learned denoising, generative backgrounds and photographed performances in the same shot. The success of the terminology would be measured by whether it helps an audience imagine such distinctions, not whether it sounds futuristic on a conference stage. One should also resist using artificial general intelligence as shorthand for any attractive video-generation model; attractive pixels are neither evidence of general reasoning nor a substitute for an artist's decisions.

There is a mischievous extra echo from a different landmark in digital cinema. In The Matrix (1999), Agent Smith lectures Morpheus about evolution, invokes the dinosaur and declares, “The future is our world, Morpheus. The future is our time.” The line was intended as an expression of the machines' ruthless certainty about human obsolescence, not a cheerful account of technological progress. It is nonetheless a perfect quotation to hold at arm's length beside Tippett's experience. Artists in 1993 feared that a machine-made monster was pushing their methods into history. A present-day animator may hear Smith's confidence in the sales pitch of an automatic image generator. The film's actual moral complexity warns against treating inevitability as an ethical argument. A machine's ability to make an image gives it no entitlement to the work, image or identity of the people from whom that ability was learned. [50]

XI. THE UNCANNY VALLEY IS NOT ONE VALLEY

There is a familiar hope that generative AI will advance until the uncanny valley disappears, leaving photorealistic images indistinguishable from images made through ordinary cinematography. That hope contains a reasonable technological prediction but also a conceptual confusion. Masahiro Mori's 1970 essay on the uncanny valley was a proposition about the relationship between human likeness and the uncomfortable feeling created by something approaching, but not reaching, human appearance. The concept has subsequently informed discussion of robots, synthetic faces, animation and digital humans. It does not describe every imperfection in computer-generated imagery, nor does it guarantee that increased surface fidelity will remove every kind of unease. The same person may be comfortable with a photorealistic dragon, troubled by a simulated dead actor and entirely untroubled by a digital correction to a photographed background. Technical appearance, perceived agency, provenance, performance and emotional identification are separate variables. [32]

Generative video introduces several distinct uncanny valleys. There is a perceptual valley, where anatomy, skin, motion or gaze appears almost but not quite natural. There is a temporal valley, where an individual frame is convincing but bodies, objects and environments refuse to maintain identity across time. There is a causal valley, where the physics of an action look superficially persuasive yet fail scrutiny: a hand lifts an object without bearing its weight, water splashes against a surface with the wrong momentum, or a shadow moves before the object that casts it. There is a narrative valley, where a generated performance imitates an emotion's outward appearance without fitting the character's accumulated experience. Finally, there is an ethical valley, where an image may look perfect but becomes disturbing when viewers learn that it was produced through unauthorised replication or an artist's uncredited work. Future models may narrow some of these gaps rapidly. They will not all close through the same improvements.

The first CGI revolution overcame many perceptual obstacles by imposing increasingly accurate models of three-dimensional space and light, but its strongest filmmakers supplemented technical precision with expressive judgement. A dinosaur's skin can be imperfect if its arrival is staged with extraordinary timing; a flawless digital face can fail if its eyes do not respond to another performer. Generative methods inherit this asymmetry. A model trained on enormous quantities of footage may reproduce natural-looking details that are expensive for conventional artists to paint or simulate, but a director must still decide which details communicate character, scale and feeling. The surface is a servant of the scene. This principle explains why Cameron's insistence on preserving human performance is not merely a moral declaration. It is a practical position about where the information driving an emotionally complex scene originates. He may be willing to automate some of the machinery of depiction while refusing to automate the intention behind an actor's gesture.

The decisive technical advance may involve pairing generative models with controllable representations rather than perfecting pure text-to-video output. Existing professional workflows often retain three-dimensional geometry, skeletal rigs, depth information, camera tracks, masks, lighting references and movement instructions. A future system could use those elements as binding constraints and generate only the visual attributes that benefit from learned synthesis. The director could preserve an exact take while changing an environment's atmosphere, alter the intricacy of costume materials without destabilising the body beneath them, or generate finer simulation detail while a physically defined object continues along an approved route. Such hybrid methods would weaken the distinction between 'CGI' and 'AI' as rival camps. AI would enter the CGI pipeline as another method of constructing, refining or rendering portions of an image, just as digital compositing entered filmmaking without abolishing photography.

Recent scientific literature places unusually precise boundaries around this possibility. A 2025 survey of generative AI for film creation, presented in the computer-vision research community, reviewed neural radiance fields, diffusion approaches, image-to-video synthesis and three-dimensional generation alongside the filmmaking problems of repeatable characters, coherent style and controllable motion. The authors were describing a rapidly expanding technical repertoire, but the recurring requests from working artists concerned revisions, continuity and fine-grained intervention. Those requirements suggest why a compelling stand-alone demonstration may remain far from the demands of feature-film production. Digital cinema's next revolution is likely to be measured through the number of decisions directors can reliably preserve and change, rather than the number of apparently realistic seconds a system can produce in a single generation. [38]

A study published on 24 September 2026 offers an illuminating experimental example. In Discover Artificial Intelligence, Kai Zhang and colleagues described CineFX-Diff, a diffusion-based framework combining textual instructions, reference images and spatial masks with temporal constraints and a physics-aware finishing stage. Working with a dataset of 6,800 annotated effects clips, the researchers reported improved quantitative results against selected video-generation baselines and faster inference. Their evaluation also asked twenty post-production professionals to judge whether outputs could serve as useful reference material or drafts. Those results indicate a plausible direction for specialist effects tools, but should not be confused with independent proof that a feature-film facility can now finish thousands of blockbuster shots with the same quality or lower total cost. The dataset was assembled by the researchers, its licensing imposes reproducibility constraints, and the authors themselves acknowledge that a source-level division of training and testing material would offer a stricter measure of generalisation. The research is promising precisely because its limitations are specific enough to test. [39]

There is a corresponding social-science answer to the assumption that usable AI must simply sweep away an existing industry. A peer-reviewed July 2026 study by AD Narayan, Angelique Nairn, Justin Matthews and Duncan Caillard analysed twelve interviews with experienced visual-effects workers and twenty-six articles on AI in visual effects. It found that practitioners recognised benefits in brainstorming, rapid visualisation, prototypes and selected technical operations, while remaining cautious about final-pixel production, intellectual property, provenance, consistency, creative control and compatibility with existing systems. Twelve interviews do not constitute a census of the workforce; their value lies in showing the practical questions that abstract predictions about automation tend to conceal. The report's historical resonance is strong. In 1993, computer animation required older motion craftsmen to discover a new professional position within digital production. In 2026, working VFX artists are again trying to decide what knowledge should migrate into a changed toolchain, what contractual safeguards are necessary and which categories of human judgement must remain readily accessible. [40]

Terminology matters because public debate will otherwise conflate incompatible activities. Computer-generated imagery, or CGI, denotes the broad family of images created or manipulated by computer, regardless of whether a contemporary generative model is involved. Computer graphics is not automatically generative AI. A machine-learning denoiser, a procedural particle simulation, a traditional character rig and a diffusion-based video synthesiser operate differently, even if they all contribute to a final digital shot. 'AI-CGI' and 'CG-AI' express a cultural intuition but lack a settled technical definition. 'Generative visual effects', abbreviated here as generative VFX, is a more useful description of the production category being discussed. 'Neural rendering' may be apt for particular methods, but it should not become a catch-all for text-to-video, machine-assisted cleanup, synthetic performance or conventional simulation. The article's forecast concerns the controlled integration of generative methods into high-end VFX, not the proposition that every effect in every future movie will be made from a written prompt. [43]

XII. WHY CAMERON, AND WHY HE MIGHT NOT BE FIRST

The case for Cameron is stronger than an appeal to his reputation. He has repeatedly combined an extreme willingness to invest in unfamiliar technologies with an equally strong insistence on shaping their final creative use. His interests bridge engineering and narrative spectacle. He has experience establishing entire production systems around an effect that conventional workflows cannot accomplish efficiently. He has a continuing franchise demanding vast quantities of controllable synthetic imagery. He has publicly articulated a precise productivity target for the next two instalments and has sought direct contact with generative technology developers. A studio can hire engineers to make a new system work, but only a filmmaker whose ambitions require the system can give it artistic necessity. Cameron's career offers numerous examples of that combination. If a director were to insist on a breakthrough that made generative VFX respectable to the rest of the industry, he would be a plausible candidate. [5][6][16][20]

Age and legacy sharpen the story without proving what he will do. Cameron was born in 1954 and has already spent decades defining and revising the visual possibilities of blockbuster cinema. The temptation is to imagine him telling sceptics that he has nothing left to prove, that whether his remaining Avatar pictures succeed no longer matters and that he is prepared to take a technological gamble simply because he can. This would make splendid dialogue in an invented dramatic scene. It is not an established account of his intentions. His repeated public concern about the economics of continuing the franchise indicates that commercial viability matters a great deal. Indeed, reducing cost and time is meaningful primarily because it may help the films get made. A more defensible interpretation is that Cameron might accept unusual technological risk precisely because he wants to preserve the possibility of making large, ambitious films without allowing the production model to collapse under its own weight.

For this reason, I would frame the central prediction narrowly: if Avatar 4 and Avatar 5 proceed on a new technical footing, they could become major demonstrations of a hybrid generative VFX pipeline developed under the supervision of traditional effects artists, while performances remain grounded in human actors. I would not predict, on present evidence, that Cameron will invite a public text-to-video platform to generate the final films in place of Weta and his performance-capture teams. Nor would I describe the films as already shooting back-to-back in their entirety. Some material intended for the fourth picture was previously captured in connection with the long production of the earlier sequels, and subsequent scheduling remains subject to confirmation. As of October 2026, their commonly reported target release years are 2029 and 2031, but those dates are planning markers rather than guarantees. [23]

Even if a major technical change arrives, the first blockbuster beneficiary may not be Avatar. The franchise's elaborate visual grammar, detailed creature designs and vast existing library of assets could make a radical transition especially difficult. Much of Pandora's current production expertise represents an enormous investment that cannot lightly be discarded. Cameron's ambition for reduced cost might instead encourage the familiar pipeline to become leaner, making better use of automation, distributed rendering, procedural systems, refined performance capture and carefully bounded machine learning. Another studio, unburdened by a pre-existing world, might construct a new production around generative assistance from the first storyboard. A director making an original science-fiction adventure could design creatures, environments and scene structure specifically to favour the technology's strengths. The historical precedent of Young Sherlock Holmes suggests that a smaller, bounded experiment can precede the giant movie that receives the historical credit.

Rob Minkoff's planned Storm Dogs offers another example of how the race might be redefined. Reporting in October 2026 described a family-oriented project combining live-action material and AI-generated animal characters, with production anticipated in 2027. It remains a project rather than a proven box-office result, but the premise is revealing because AI animals are closer to the class of effect that once made Jurassic Park revolutionary. A convincingly performed animal character may provide a clearer demonstration than an elaborate cloudscape or a cityscape rendered behind human actors. If a future family adventure presents sustained interactions between photographed people and AI-generated creatures, the result could provide both a technical benchmark and a test of audience acceptance. The technological prize is distinct from the ethical challenge posed by simulating real performers. Animals and fantasy creatures may offer more flexibility, although their design and animation still involve the work and rights of artists. [33]

The historical winner may therefore be unexpected. The movie that first uses generative VFX extensively might be a modest success in a regional market. The first global commercial hit might be released by a streaming company and watched without an exhibition campaign emphasising technical novelty. The film that becomes shorthand for the revolution might arrive years later and use only a portion of the technique its predecessors developed. Historians of CGI already distinguish the first generated character from the first fully digital feature and the first truly influential photoreal blockbuster. The new era will require equally careful categories. The strongest wager is not that Cameron will certainly be first to use generative methods. It is that he is among the directors most likely to turn them into an ambitious, visible claim about the future of cinematic spectacle, provided the technology can meet his unusually demanding standards.

It would be easy to cast Cameron as a technology evangelist and leave the case there. His history suggests a stranger role. He has spent decades pressing computers towards the expressive goals of live-action direction while insisting that actors' performances remain foundational, and his ambition to reshape the economics of Avatar is inseparable from those commitments. In that light, the most intriguing interpretation of his comments is that he could help invent a workflow in which generative VFX becomes the servant of precisely controlled motion, rather than a replacement for the people generating it. He may be among the artists able to ask the inconvenient question at the right moment: what would it take for the system to make this frame again, with one part changed and everything else intact? That question could matter more to the history of blockbuster cinema than the most spectacular image generated by a consumer tool in the preceding five years. It is also the boundary between the extinction of a technique and the extinction of a worker's creative agency.

XIII. THE COST OF A FASTER IMAGE

The economic argument deserves more scrutiny than the familiar headline that AI will make movies cheaper. A film's visual-effects budget includes people, facilities, software, computing infrastructure, management, storage, data movement, research and development, revisions, quality assurance and the costs of coordinating numerous departments. Increasing the speed of one operation does not automatically reduce the cost of the whole production in equal proportion. If a generated image takes one minute rather than ten hours to create, that may be transformative for an isolated task but irrelevant to a schedule delayed by an actor's availability, unresolved editorial decisions or a contract negotiation. A producer needs total cost per approved shot, adjusted for the probability of revision, rather than the advertised cost of producing an attractive first pass. Cameron's declared ambition to work in half the time suggests that he is thinking at the level of workflow and throughput, which is also how he described his interest in generative systems in 2025. [6]

There is also a paradox of demand. Historically, productivity improvements in effects have often allowed filmmakers to attempt more ambitious images rather than simply reduce spending. Once a city can be rendered, directors want more inhabitants, more detail, greater destruction, more elaborate lighting and longer sequences. Once a creature looks convincing, it becomes possible to give it a more demanding performance and more screen time. An effects facility that can produce a former week's work in a day may be instructed to explore seven times as many variations. Lower unit cost can increase total creative appetite, and final costs may remain high. The 1993 dinosaur was a scarce wonder because every second was difficult; later blockbusters could present entire digital armies and ecological systems because the technology expanded the feasible scale of spectacle. The generative equivalent may be a film with richer detail and greater freedom of iteration, not necessarily a film produced for pocket change.

The labour consequences cannot be waved aside by invoking earlier history as reassurance. Tippett's adjustment to CGI is an instructive example of skill migration, but it does not prove that everyone in a displaced specialty will find new, equally secure employment. Different workers have different capacities to retrain, different access to employers and different positions in a production hierarchy. Software can sometimes expand demand for human expertise while simultaneously diminishing the number of people performing entry-level tasks. The creative industry's apprenticeship problem is particularly acute. Junior workers learn the judgement required for supervision by engaging with complicated details, receiving corrections and discovering why a shot fails. If a generator eliminates much of that work, the industry will need deliberate ways to develop the next generation of artists. Promising that AI will free everyone to do 'more creative things' is an appealing ambition, but it should be evaluated against employment practices, pay and training rather than accepted as an automatic outcome.

The legal environment provides part of the response. During the 2023 Hollywood labour negotiations, performers secured contractual protections addressing digital replicas, consent and compensation. SAG-AFTRA's published explanations distinguish a performer's authorised digital copy from a wholly synthetic performer and outline obligations involving permission and collective bargaining. The Writers Guild also negotiated principles limiting the status and use of AI-generated literary material. Neither achievement guarantees that every present and future problem is solved, and contract provisions evolve. What they establish is that technological deployment in professional filmmaking operates inside negotiated labour relationships. The unionised workforce is not simply an audience for technical announcements; its members are participants in deciding what constitutes an acceptable production. A Cameron-led innovation that sought broad professional legitimacy would have to take these relationships seriously. [34][35]

Training data is another hidden cost. A studio may find it convenient to use models developed from broad internet-scale datasets, but doing so introduces legal and reputational uncertainty where copyrighted imagery, licensed designs or performer likenesses are implicated. The Seedance controversy makes that risk plain. A more controlled studio model could be trained or adapted using approved assets, licensed footage, proprietary motion data and production-specific references. Such a system would be expensive to establish and perhaps much narrower than a general model trained across many kinds of imagery, but it could provide stronger guarantees of ownership, confidentiality and reproducibility. There is no reason to assume these guarantees are technically effortless. Provenance information must survive through intermediate transformations, and consent may be limited to a particular project or type of use. The apparent cost of generating a frame leaves much of the cost of making it legitimate outside the calculation.

Environmental costs, often neglected in industry publicity, should also be included. Conventional CGI uses energy-intensive rendering farms and storage networks; training and operating generative models likewise require specialised computation, electricity, cooling and data infrastructure. The meaningful comparison is not whether either process uses computing, but how much energy and equipment each approved production task consumes, including repeated generations and failed attempts. A hybrid workflow could decrease one category of computational burden while increasing another. Any broad assertion that generative methods are automatically greener or dirtier than conventional effects would require task-level measurements that are not available for many commercial systems. More responsible studios will have reason to measure these impacts as part of production accounting. The arithmetic of visual progress should include the material systems hidden behind the apparent immateriality of a digital picture.

The dinosaur metaphor risks becoming sentimental unless it includes the costs of transition. Some craftsmen did lose familiar kinds of work as digital production changed; some adapted, and some never had the opportunity or resources to retrain. The equivalent transformation in generative VFX could compress entry-level positions, change the economics of outsourced effects work, alter contract bargaining and concentrate control within companies that own models and computing infrastructure. If studios simply celebrate an apparent reduction in the number of staff-hours required, they may discover that they have reduced the training ground from which their future creative supervisors would have emerged. A sector can become faster at delivering shots while becoming weaker at preserving the knowledge necessary to judge them. An extinction story is only triumphant from the perspective of the creature that survives. From the point of view of the animal that disappears, it is an account of vanished habitat, lost work and a future it cannot enter. This is why Tippett's adaptation is a precedent rather than a guarantee. [34][35][40][42]

XIV. THE 2030S, HERE THEY COME

My more specific wager is 2033. It is not a declared Avatar release date, and it certainly does not establish that Cameron intends to make Avatar 6 or Avatar 7. As of October 2026, Avatar 4 and Avatar 5 are the sequels publicly contemplated, with tentative release slots in 2029 and 2031. A sixth or seventh film is an imaginative extension of the franchise's long horizon, not a confirmed order or announced production. What makes 2033 useful is its position at the outer edge of a plausible technological development window: eight years beyond The Eternaut, forty beyond Jurassic Park, and one hundred beyond King Kong. A Cameron project arriving around that year, whether an accelerated fourth or fifth chapter, a further unconfirmed sequel, or something else he directs, could present an effects pipeline radically faster than today's while rendering results indistinguishable from photographed physical reality in the kinds of shots the story requires. That is a forecast to be judged in hindsight. [5][51]

The word indistinguishable must be understood as a filmmaking standard, not a promise that every generated image will be physically perfect or that audiences will lose the ability to detect manipulations in all circumstances. The relevant standard is a sustained dramatic sequence under scrutiny: characters who remain consistent, environments that behave coherently, surfaces that preserve material logic, performances that belong to consenting actors, and changes that can be directed precisely. If Avatar 4 and 5 achieve Cameron's efficiency target without compromising these qualities, the breakthrough might precede 2033. If the technology matures more slowly, the landmark might belong to another studio or a later film. The exciting part of the prediction is the convergence of the artist, the commercial problem and the incoming tools, rather than an exact date pretending to be a timetable.

What would a genuine Jurassic Park moment for generative cinema look like? Imagine a 2030 release that contains a seven-minute set piece in a spectacular fictional landscape. Its human performances are captured and directed in ordinary filmmaking terms. Its creature animation depends on approved rigs and performances. Its camera geography is designed in a three-dimensional scene that an editor can understand and an effects supervisor can inspect. Its environmental detail, certain simulations, atmosphere and intricate surface behaviours are generated or refined by models constrained to match that scene. Several shots are produced in a fraction of the time previously needed, but the filmmakers can still revisit individual actions without the model reinventing the rest of the world. The audience sees a sustained sequence of precise, emotional, coherent imagery. When the production details emerge, other studios discover that a substantial portion of the difficulty lay in a new hybrid process. This is an imagined example, not a report about an announced film, but it identifies a testable threshold: quality, duration, coherence, control and demonstrably improved production economics in the same delivered sequence.

There are several ways the equivalent moment could differ from Jurassic Park. A breakthrough might be invisible at first because the result is an ordinary street, crowd or environment rather than an impossible animal. It might arrive in a film that makes little money, making its historical influence clear only after other productions adopt the method. A movie might be celebrated for artistic invention while its generative workflow remains undisclosed because disclosure could cause unnecessary controversy. Conversely, a studio might boast about AI and generate an immediate backlash, even if the result looks excellent. The history of film technology warns against identifying technical achievement with critical acclaim, industrial adoption or commercial performance. Sound cinema, colour, widescreen formats, motion control, CGI, stereoscopy and virtual production have all moved along different curves. A successful prediction must be elastic enough to recognise whichever of these curves generative filmmaking follows.

If Cameron leads, I suspect the reveal will concern the process rather than the spectacle alone. He may show a sequence and then explain how its time to completion changed; describe artists iterating with unusual speed; present a comparison between a conventional pipeline and a new one; or emphasise the extent to which captured performances remain untouched while the surrounding image becomes faster to build. The most persuasive communication would be generous to the craftspeople who made it possible. The historian's analogy to Tippett would then become vivid. The specialists who once animated models may be working with generated detail, motion constraints and different tools, but they would still be judging movement, weight, expression and continuity. The word 'revolution' would describe an alteration in the workshop, not the disappearance of human intelligence from cinema.

The remaining Avatar films provide the strongest possible laboratory for such a demonstration because the franchise has already persuaded audiences to care about people who appear almost entirely as digital bodies in digital environments. If generative systems can be introduced while retaining actors' emotional performances, Pandora could become the bridge between the two technological eras. A visitor to a cinema in 2031 might encounter an environment denser than anything that could have been fabricated in 2025, with creatures behaving in complex, persistent relationships to one another and to the landscape, supported by a pipeline far more responsive to the filmmaker. Alternatively, a technically conservative Avatar 5 might remain a triumph of refined conventional methods while some younger director makes the decisive generative film elsewhere. Both are plausible. Cameron's own statements support a desire for technological and economic improvement, not the certainty of any specific solution.

Perhaps the most useful forecast is a sequence of stages. The late 2020s may be remembered for isolated use cases, competing standards, disputes over rights and increasingly capable systems constrained to particular professional tasks. Around the turn of the decade, one or more productions could demonstrate sustained generative VFX at scale, in a released film whose qualities withstand scrutiny. In the early 2030s, if the underlying methods improve and legal access becomes settled, hybrid generation may become an ordinary part of feature-film post-production, even when marketing campaigns choose not to mention it. Those dates are a reasoned scenario, not a clock set by technical destiny. Advances in controllability, costs, computing resources, labour agreements, rights clearances and public reception could accelerate or delay the transition by several years. The comparison with the past is a map of possibilities rather than a guarantee that history will repeat on schedule.

If the late 2020s give us an AI-generated stained-glass knight, the early 2030s may give us the next digital brachiosaurus. And the director who turns that creature towards the camera might very well be James Cameron. The wager is attractive because it joins technological ambition to a recognisable artistic temperament, a filmmaker who has spent much of his career refusing the suggestion that an impossible image must remain impossible. But the history of visual effects deserves the last word. The dinosaur in Jurassic Park did not persuade audiences because the computer finally became sufficiently powerful to boast about itself. It persuaded them because the camera held, the actors stared, the animal breathed and, for a few seconds, a manufactured image became the most natural thing in the world. The next revolution will have truly arrived when its technique vanishes into an equally unforgettable moment of cinema.

Imagine, at the beginning of the next decade, the first audience watching a sequence that could not have been made on anything like the same schedule five years earlier. They have paid for a story, not a seminar in machine learning. They see a creature moving among real actors, or an enormous imagined landscape responding to a performance, and the illusion holds. The credits will still include hundreds or perhaps thousands of humans, despite claims elsewhere that machines have made their work redundant. Later, the technical account may disclose that an animator controlled motion through a hybrid interface, that an environment artist directed the synthesis of a landscape through spatial constraints, or that entire layers of variation were generated quickly enough to allow experimentation previously abandoned as too expensive. The apparent simplicity of the final image will conceal an elaborate negotiation between two generations of tools. This would be a fitting successor to 1993. A technology might appear to make a profession extinct while simultaneously creating a new form in which the profession's knowledge endures. The coming wonder and the threatened predecessor would turn out to be different faces of the same dinosaur.

AFTERWORD: HOW TO RECOGNISE THE TURNING POINT

The prediction presented in this article can be assessed against specific evidence over the coming years. The most persuasive announcement would identify a released feature film and explain, with production-side testimony, that a substantial sequence of final visual effects was generated or transformed through controllable generative systems. It would describe how the technique integrated with conventional assets, who authorised the training and reference material, what happened to performances, and whether total work per approved shot fell under realistic revision conditions. An impressive video demonstration alone would not suffice. A claim that a film used 'AI' without explaining whether it referred to machine-learning denoising, production scheduling, animation support or final image synthesis would also be inadequate. The public history of cinema benefits when milestones are defined carefully rather than awarded to whoever possesses the most persuasive marketing department.

Success would not require unanimous approval. Jurassic Park did not settle every argument about practical and digital imagery, and contemporary objections to generative AI include matters of consent and livelihoods that visual excellence cannot neutralise. A genuine breakthrough could coexist with labour disputes, aesthetic scepticism or uneven audience response. The possibility of these conflicts is one reason transparent accounting will matter. The film that ushers in a new era should be able to say what the technology accomplished, what humans contributed, what rights were respected and what the audience actually saw. The terms are more demanding than a technical trick, but appropriately so. The promise of generative cinema is finally a promise about making films, not a promise about making pixels.

Until that moment arrives, we inhabit a familiar interval of persuasive fragments, ambitious claims and unresolved craft questions. In 1933, an articulated ape climbed New York's most famous building; in 1985, a stained-glass knight stepped out of a window; in 1993, a digital dinosaur walked into sunlight; in 2025, a building collapsed inside a Netflix series through a newly practical technique. By 2033, perhaps an imagined creature will move with such physical authority that the audience will have no useful way to identify which stages of its creation were modelled, photographed or generated. The dates offer a rhythm, not a destiny. The next dinosaur remains both the new wonder on the screen and the established technique facing its own historical sunset. What should survive is the discernment that makes images worth watching. Tippett's experience shows that an art can change its tools while retaining its memory; it cannot promise that everyone displaced by change will have an easy transition.

SOURCES AND RESEARCH NOTES

The references below support the historical and production details in the essay. Where a studio provides a cost or efficiency estimate, the text attributes that claim rather than presenting it as an independent audit. Industry forecasts, proposed terminology and imagined scenes are explicitly identified as opinion or literary reconstruction. For a simpler publishing copy, the notes contain no embedded external hyperlinks.

[1] Industrial Light & Magic. Breathing Life into the Dinosaurs of Jurassic Park. ILM retrospective.

[2] Academy Museum of Motion Pictures. Making Digital Dinosaurs. Academy history.

[3] Reuters. ByteDance pledges to prevent unauthorised IP use on AI video tool after Disney threat. 16 February 2026.

[4] Axios. Motion Picture Association sends cease-and-desist letter to ByteDance over Seedance 2.0. 20 February 2026.

[5] Variety. James Cameron Wants to Make Avatar 4 and 5 in Half the Time for Two-Thirds of the Cost. May 2026.

[6] Business Insider. James Cameron says the cost of blockbuster films needs to be cut in half, and AI is the answer. 10 April 2025.

[7] Variety. James Cameron banned generative AI use in Avatar: Fire and Ash. December 2025.

[8] British Film Institute. Celebrating Saul Bass’s centenary: 10 essential title sequences. BFI retrospective.

[9] Heritage (MDPI). Visual Heritage and Motion Design: The Graphic-Cultural Legacy of Saul Bass’s Title Sequences. 2025, scholarly article.

[10] Wookieepedia, citing crew testimony. Opening crawl: physical artwork and motion photography. secondary compilation, consulted with caution.

[11] Industrial Light & Magic. Young Sherlock Holmes: Milestones and Memories from an ILM Classic with Dennis Muren. 19 March 2026.

[12] Guinness World Records. First film character computer-generated. historical record.

[13] TechCrunch. Netflix starts using GenAI in its shows and films. 18 July 2025.

[14] PBS NOVA. Special Effects: Titanic and Beyond; 1980s chronology. historical timeline.

[15] Amblin. Young Sherlock Holmes: About the Movie. studio history.

[16] Industrial Light & Magic. Terminator 2: Judgment Day. ILM production record.

[17] Industrial Light & Magic. ILM Evolutions: Animation, from Rotoscoping to Rango. effects-history retrospective.

[18] Industrial Light & Magic. Jurassic Park: production milestones. ILM production record.

[19] Julie A. Turnock. The Empire of Effects: Industrial Light & Magic and the Rendering of Realism. University of Texas Press, 2022.

[20] Time. Why It Took So Long for James Cameron to Make Avatar: The Way of Water. 2022.

[21] Wētā FX. Our Work on Avatar: The Way of Water. 30 March 2023, visual-effects production report.

[22] The Walt Disney Company. Avatar: Fire and Ash Surpasses $1 Billion at the Global Box Office. 4 January 2026.

[23] People. Avatar 4 and 5 Are Full Speed Ahead After Third Movie Crossed $1 Billion. 2026, quotes from producer Rae Sanchini.

[24] The Guardian. Netflix uses generative AI in one of its shows for first time. 18 July 2025.

[25] The Washington Post. Secret Invasion creeps out fans with an AI-generated intro sequence. 22 June 2023.

[26] Variety Australia. After AI video of Tom Cruise fighting Brad Pitt goes viral, MPA denounces infringement. 13 February 2026.

[27] Runway. Runway and Lionsgate Expand Partnership. 11 June 2026, corporate announcement.

[28] Netflix. Innovation for Filmmaking, By Filmmakers: Why InterPositive Is Joining Netflix. 5 March 2026, corporate announcement.

[29] StarWars.com. 5 Groundbreaking Digital Effects in Star Wars. Lucasfilm retrospective.

[30] StarWars.com. Animating Star Wars: The Phantom Menace with ILM’s Rob Coleman. Lucasfilm interview.

[31] Entertainment Weekly. Every Upcoming Star Wars Movie and TV Show. 8 October 2026.

[32] Masahiro Mori / IEEE Spectrum. The Uncanny Valley: The Original Essay, translated by MacDorman and Kageki. original 1970; authorised translation 2012.

[33] The Times. Lion King Director: Why I’m Embracing AI; Rob Minkoff and Storm Dogs. 7 October 2026.

[34] SAG-AFTRA. 2023 TV/Theatrical Contracts: AI and digital replica protections. 2023 agreement and subsequent guidance.

[35] Writers Guild of America. Artificial Intelligence: WGA rights and the 2023 MBA. updated 18 December 2025.

[36] Los Angeles Times. The New Beastmasters’ Monsters. 6 June 1993, contemporary preview.

[37] Stephen Jay Gould / The New York Review of Books. Dinomania. 12 August 1993, contemporary essay.

[38] Ruihan Zhang and colleagues / CVPR Workshops. Generative AI for Film Creation: A Survey of Recent Advances. 2025, research survey.

[39] Kai Zhang, Di Wu, Yanan Xu and Baolai Bai / Discover Artificial Intelligence. Generative Models for Automated Visual Effects in Film Post-Production. 24 September 2026, peer-reviewed research.

[40] AD Narayan, Angelique Nairn, Justin Matthews and Duncan Caillard / Media International Australia. The Limits of Disruption: AI and the Persistence of VFX Labour. 31 July 2026, peer-reviewed research.

[41] Jurassic Park (1993), Steven Spielberg. Dialogue: Alan Grant, Ellie Sattler and Ian Malcolm. film quotation record, IMDb.

[42] Ian Failes / vfxblog. The Oral History of the Dinosaur Input Device: How to Survive the Near Death of Stop-Motion. 2018, interviews with Phil Tippett and original production artists.

[43] Visual Effects Society. Real Applications of Generative Artificial Intelligence in Visual Effects. VES technology and education resource.

[44] PBS NOVA. Special Effects: Titanic and Beyond, the 1930s and King Kong. PBS historical chronology.

[45] Turner Classic Movies. King Kong (1933): production methods and historical significance. TCM / AFI film history.

[46] The Film Foundation. Out of the Vaults: King Kong, 1933. Film preservation historical essay.

[47] British Film Institute. Celebrating Ray Harryhausen’s Centenary: 10 Essential Films. BFI, 28 June 2020.

[48] Turner Classic Movies. The Legacy of King Kong. TCM, 30 April 2011.

[49] Google Cloud. What Is Artificial General Intelligence?. 14 January 2026, terminology.

[50] The Matrix (1999), screenplay transcript. Agent Smith dialogue: The future is our world; the future is our time. Film quotation checked against dialogue transcript.

[51] Associated Press. James Cameron on Two Decades of Making Avatar and the Future He Sees for Movies. 2025 interview and sequel scheduling.

[52] Deloitte Insights. The Future of Cloud Gaming. Analysis of bandwidth, streaming and latency.

[53] Lyu, Wang and Sivaraman, Computer Networks. Systematic Assessment of Cloud Game Adaptability for Network Conditions and User Experience. Peer-reviewed research, May 2026.

COMING UP NEXT: THE DOWNLOADLESS WORLD

A related revolution is already forming beyond the cinema screen, and the possibilities for games may be even stranger. Imagine opening a game instantly from a cloud library, with no local installation, no visible updating, and a catalogue whose accessible scale feels effectively limitless. Every new texture, game world, patch or expansion could live on remote machines rather than occupying the player's storage. A screen and controller might become the principal hardware requirements, while remote computing performs the demanding work. This vision is not currently a guarantee of infinite physical storage or literally zero latency: cloud gaming still depends on network quality, the speed of light, regional computing facilities, operating costs and commercial rights. The attraction lies in approaching imperceptible delay and eliminating the download as an ordinary ritual for many players. That next essay will examine streamed play, responsive remote worlds, storage economics, the practical limits of bandwidth and what a genuinely downloadless culture might change about the idea of owning a game. Cinema's new method of creating images is only one part of the story. [52][53]

Something else will surpass AGI in time too…