One Film, Three Ways: AI and CG Animation
Choose Your Protagonist: A study of My Boyfriend is a Superhero!?
*From Generative Animation: Volume 4 of The New Machine Cinema*
A Very Long Carriage Ride proved that style is a parameter and that a film can survive a shift in visual ontology while furthering deeper into a directorial identity than previous forms made possible. That discovery would become an innate part of the future of storytelling. Here we ask, if style can vary while the film remains the film, what else can vary?
The answer arrived with 2025’s My Boyfriend is a Superhero!?
Five versions of the same film were released: two protagonists and the True Line Cut. The project became the first Choose-Your-Protagonist feature, the first solo made fully AI 3D CGI animated feature, and the first practical deployment of agentic AI editing in a film. Around 15 minutes of live action AI footage also became a notable demonstration of early live action footage, that I would go on to prove at feature length with 2026’s Strings.
Superhero also generated, through its own production confusion, the counter-theory of the True Line Cut.
Entering the Dominant Register
Superhero!? was produced in 3D CGI as we know it. 3D animation is the dominant commercial animation form. It is the language of Pixar, DreamWorks, Illumination, and Sony Animation, films that seem to run an elaborate juggling act year after year around which will dominate the box office with its talking animal films and anthropomorphic inanimate objects brought to life features. The point stood, As long as AI animation operated in stylized or handmade-feeling registers: vintage anime, classical 2D, stop-motion, the gatekeepers could quarantine it as a curiosity, while ignoring its true industry cash cow: CG animation.
Enter AI. Producing end-to-end AI in the register where audience expectations are highest and tolerance for deviation is lowest closes the quarantine of CG animation as a gatekept frontier. In fact, CG animation was one of the earliest breakthroughs in AI capabilities.
The choice became an ontological necessity. Superhero!? required a visual language in which the whole could be generated from intent rather than through simulation. The storytelling, as always, was king. There were chase scenes across rooftops, robot gun battles, government and corporate conspiracies, crossed with quiet domestic moments in cluttered apartments. Every point in the spectrum was touched across this film, perhaps creating the most varied and complete AI feature film to the moment of its release.
These sophisticated register with so many ups and downs required a technology that could sustain anything that was thrown at it, like a blender that could slice and dice hard chunks of ice with ease. CG was the the best test case for the architecture a blockbuster superhero film required.
The decision to enter CG carries implications beyond the strategy of the needs the film required. CG is, right now, more than anything, the sacred cow. It is the animation branch defined most by its pipeline, most operating in an industrial scale, and most ideologically committed to a specific process, creating a critical battleground: if CG fell, the whole thing fell.
But as I saw with Superhero!? the response to such a significant existential threat was not to fight it, it was merely to ignore it outright and pretend the film in front of them simply did not exist.
The Pipeline Collapses
Traditional CG animation is chain of technological dependencies. Each stage feeds the next. Each stage is staffed by specialists. Each is a bottleneck. The entire organizational structure of a studio like Pixar or DreamWorks operates as a sophisticated cogset handing off the tools between the other. A CG animated feature is a manufacturing process with extremely long lead times and extremely high coordination costs.
CG alone had built an industrial apparatus so complex that the pipeline became the medium’s identity in the eyes of practitioners. A technology requiring thousands collapsed into something one single person can do trivially. The image, characters, environments, storyboard, writing, arrives with basically no effort. There are no lighting department. Most important, there is no creative oversight telling you what can and can’t do.
To work in CG meant to work inside the pipeline. To master CG meant to master one’s station within the larger system. The entire professional culture organized around pipeline fluency.
When the pipeline collapses, that professional culture loses its basis. The individual who generates a CG film through AI has no station in the old sense. And comparing the end result, landing at 1:1, even then, still reading as professional caliber CG animation, is no longer computed through the same craft, tradition, and industry. As stated in my essay Breaking the Caste System of Independent Cinema, the appraisal comes before the product, or the product subsequently is defined on retroactive preconditions.
The Uncanny Valley
CG spent decades fighting the uncanny valley through sheer brute-force. The market necessitated the new subset of animation to work on its own terms, before the output was even ready. The trajectory from Toy Story to Elemental is the story of technology itself. Each generation of films closing the gap between what a simulated image could achieve and what the human eye expected.
AI asks us to merely consider the trajectory as it already existed: statistical convergence on what CG looks like, learned from the entire existing corpus of CG imagery. The AI model thereby knows what a Pixar-quality specular highlight looks like because it has seen millions of them. It reproduces the visual signature without computing the physics that originally generated it. It, like any artist at all, takes from what it knows and what has inspired it, and in all fairness, we must defined these words in a new machine light: knows, and inspired.
Naturally, what follows with AI CG is its own structural uncanny logic, mapped completely to different coordinates than a traditional image set.
AI, in its initial stages, struggled with things traditional CG solved early: consistent topology, physically accurate reflections and refractions in complex scenes. Tthe precise movement of rigid-body collisions.
For the director, this means learning the personality of the model itself. A new kind of material substrate emerges that is neither stop motion, CG or 2D animation. Think of the stop motion animator’s knowledge of each type of figurine’s motion. Or the 2D animator’s knowledge of space.
Every medium teaches its practitioners its strengths and weaknesses. AI CG has its own geography of strengths. Of deceptions. The director who works in it must map that geography through practice. This is intuitive. Every new artform has its consideration. Even the CG practitioner will immediately find it, while operates in parallel, brings an entirely new canvas to the imagination.
The Guardrails of CG Animation
Of all the animation branches, CG was the most inaccessible to independent filmmakers and most baffling to even attempt it with an independent film budget. 2D animation had always permitted solo practice. Stop motion required just a camera and a figurine. For CG, the materials cost too much money, the labor was too intensive. And the output was more strictly guarded.
But the barrier of CG ryhmes with the barriers of AI, ultimately: it is the study of computational infrastructure; for AI, that is algorithmic expenditure. The software tranch of CG animation: Maya, Houdini, RenderMan, costs thousands of dollars to upkeep subscriptions annually. Just one single frame of a DreamWorks animated feature required hours of render time on server farms.
The knowledge base of CG animation spans multiple engineering disciplines. A solo CG feature at studio quality was, prior to AI, a structural impossibility.
Superhero!? breaches that wall wholly. It is the first time that a single director, working with generative tools, produced a feature-length 3D CGI animated film that operates in the same visual register as studio productions.
The access gap was widest in CG and the collapse is therefore most dramatic of all my productions. DreadClub: Vampire’s Verdict proved that a fully AI pipeline produce a complete animated film. A Very Long Carriage Ride proved that generative AI could operate across a multitude of animation traditions simultaneously. My Boyfriend is a Superhero!? proved what exactly? It stood as perhaps the most complex proof of the three animated films: it proved that AI could enter the one tradition that had been genuinely closed to independents.
This, beyond the technical proof, beyond the interaction, beyond the entertainment, is the democratization argument of generative AI cinema at its incisive. Every prior branch of animation had some version of the outsider artist. CG had almost none. No rebellious practitioner. No outspoken ideologue. No counter-culture to speak of. The economics and infrastructure requirements excluded such indulgences of the artist.
While AI CG creates the possibility of independent CG cinema for the first time: a certain kind of cinema that engages with, counters, and remarks upon the dominant commercial animation register from outside the institutional structure that stood in full control.
Choose Your Protagonist
AI CG severs the bond between process and appearance that sustained the teleology of photorealistic CG animation. A CG-looking image now, no longer implies a CG pipeline was behind it. The aesthetic of computational imagery traditionally associate with Pixar or DreamWorks can now be generated without computation in the old sense. But the new sense, algorithmic expenditure is born. That is more efficient at the level of, on a budgetary scale, between 10,000x to 30,000x. I use the latter figure to describe the difference between the budgets of Toy Story ($30 million in 1995 dollars) and My Boyfriend is a Superhero!? ($1,000 in 2025 dollars).
This connects to the “style as variable” finding from A Very Long Carriage Ride, and further extends it into CG animations’s specific territory. That is what I call, choose your protagonist.
At the title card, the viewer makes a decision: Abigail A or Abigail B. Both versions of the protagonist of Superhero!? have the same name, the same dialogue, same blocking, same plot. But there instead, the viewer is given a choice between the two: one white, one Black.
The concept emerged from a cultural observation that had been building for years about race and popular entertainment. This culminated in a firestorm around Disney’s casting of a Black actress as the Little Mermaid, revealing how a single casting decision could become a proxy war for culture politics, tradition, marketing, and belonging. In no sense am I attempting to validate the controversy, rather I put forward that audiences who took on the matter as one meaningful to them, treated the protagonist’s race as if it were the film’s primary meaning, its purpose. That the question of who you see on screen overwhelms every other element of the work.
The discourse was ugly, reductive, and toxic all around, but in it I saw a seed of value that identity matters, and that inherently, no matter what category, identity, or framework is included, it explicitly implies that another category becomes excluded in the process. To me the question of the protagonist’s identity becomes one of the most powerful and pressing variables that modern cinema was struggling to address in any meaningful form.
Superhero!? was designed to address the question. Not to provide answers, but to remark that the question existed at all required some engagement with the question. By isolate the one variable of identity, we look forward toward a future customizable cinema. Two versions of the same film, identical in every respect except the protagonist’s race. In the process of the experiment however, I found a certain richness opened. Like A Very Long Carriage Ride, the formal experiment took on a life of its own.
It becomes impossible to view either version without comparing them. Who was better? Who was different? Which beat landed how, and differently? Directorially, this was fascinating. I had gone in expecting a mere character change. I did not expect that both Abigails would bring about two completely different personalities. That I would form different relationships to both performances, yield to different strengths, and build around different weaknesses. What, on the surface, seemed to be aesthetic, became a richly rewarding sociological study.
Therein the Black Abigail takes on an entirely different light. What began as diversity introduces a racial element to the film which did not exist in the white Abigail version: she is the sole minority in entirely white corporate buildings and offices after all. Tension lands in stretches that went smoothly for Abigail A, now, with Abigail B, the power imbalances became more pronounced. Little bits of social minutia began to feel like injustices. Standard conversations became awkward.
A rewarding artistic framework emerges. I came to see that the experiment in character identity became in its own right, an experiment in seeing the world through another character’s eyes through contrast alone.
The flirtation scenes for instance. In one version, the early banter between Abigail and her boyfriend plays as uncomplicated romantic comedy. In the other version, the same exchange carries a faint undercurrent of public visibility. The café where they meet is the same café, the extras are the same extras, but the couple at the center of the frame now registers as interracial. The flirtation now exists in an entirely different frame than a white with white romantic comedy. The audience holds its breath even if only for a half-beat longer.
In a sense, cinema has always known this in practice. Casting changes films. This much is obvious on the surface. Presence changes films. This is obvious.
Films have always been built on identity, unknowingly. What had not been crystallized was the possibility of building those alternate bodies into the same work, preserving the same architecture, and letting the viewer witness how the variable bends the emotional and political weather.
This, beyond any aesthetic question in Superhero!? generative AI CG animation practice, expands it into the illusion of inevitability. The body is shown as a chosen vector through the same work. The film thusly expands what cinema can be as a technicality, while stripping it bare, shining a light on what cinema has always been doing culturally, failing and succeeding in different parts and measures.
The proposal is not without its own set of risks. I was fully aware that treating racial identity as a parameter invites the accusation that the film reduces race to a cosmetic toggle. If the script stays the exact same across both version, is it not implicitly arguing that race itself is a mere superficial aesthetic construct? That a Black woman’s experience in a boardroom is interchangeable with a white woman’s as long as the lines match?
But as a work of art, I resolved that the film is not meant to answer the questions that arise from its practice. It is merely to document the question itself. There I found the controlled environment reveals instead of flattens racial identity questions. By holding every other variable constant, the film implicitly becomes about how a character’s identity changes the meaning of the work around them.
The experiment becomes bigger than the film itself by every measure. There, we must also give credit to the careful construction of the performances. The danger might have been more pronounced if the performances registered as flat skin changes with robotic performance. They do not. Abigail A and Abigail B are both delightful performances with entirely different personalities and registers.
The Adaptive Cinema Engine
The hidden breakthrough inside Superhero!? is in its experiments with agentic AI editing.
In DreadClub I experimented with paper edits with LLMs, describing scenes and storyboards and asking it to time out sequences, after which I matched their edits exactly as specific. This was used as a proof of concept for AI-enhanced editing across several scenes in the film.
On Superhero!?, I allowed the AI agents to begin collecting and assembling footage on its own. The machine did not wait for instruction on which angles to generate, instead, it offered coverage through its automated systems. The experience was like playing ping-pong. I would place a shot decision into the agent’s hands. It would return a net result through its own algorithmic methodology. I would build on the proposal and return the iteration and receive another echo of the shot.
The agent would respond. Across the film, a pattern emerged where machine cuts lived inside the directorial picture.
The agent never directed the film. It collaborated in the film’s assembly in a way that no prior tool in my pipeline had done.
By Strings, the agentic component had grown to somewhere between fifteen and twenty percent of the film’s cutting. The trajectory is clear. Paper edits to scenes to entire features were now being entrusted upon machine agents. The agent’s role expands as the director’s trust in the collaborative feedback loop deepens. The director gains a collaborator whose proposals are both surprising and architecturally sound. It is never not in the bounds of what a director themself might come up with. Occasionally, it would propose something completely out of left field, and occasionally, that would go into the movie.
But Superhero!? was the production where the motivation started to sour. I had never directed two films in a single year before, let alone two dual releases, and the production was grueling. Directing the second Abigail was the hardest thing I had done in filmmaking, because I had poured everything into the first version of the film and left nothing for the return journey, and there I was, forced to recreate the entire protagonist from start to finish.
Parametric model must eventually resolve this. The machine’s capacity to generate and maintain versions is limitless. A director’s is not. Eventually algorithmic machine filmmaking will exceed any single director’s capacity to oversee a production. The system scales with the technology, not with our human ability to engage with it.
This will eventually require new models of directorial organization, or new forms of trust between director and agent, that do not yet exist. The question as it stands is too large. The theoretical elegance of the variable film meets, in practice, the finite energy of the person who must hold every version in their head at once.
The True Line Cut
The third version of My Boyfriend is a Superhero!? was born from a mistake.
Early in production, I could not decide on the character model for Abigail. I committed to one, then replaced her with a second, then a third. Each version locked into the film at the moment I was certain she was correct, and each was abandoned days later. On a traditional production, this indecision would have been resolved months before shooting at a cost that enforced decisiveness. On an AI production moving at the speed of the mind, the indecision played out in real time, on screen, inside the film itself.
The realization arrived through the mistake: the process of making the film could be the film.
I call this the True Line Cut. Miles Davis walked into the studio with sketches, not finished compositions. What the musicians performed, in that session, on that day, went onto the tape. The result was Kind of Blue—a work whose value derives precisely from the fact that it was immediate, unedited, and lived in real time.
AI production moves fast enough to achieve something analogous in cinema. A film can record the artist’s decisions as they happen — the hesitations, the false starts, the moments of conviction and revision, and present that record as a legitimate version of the work. The True Line Cut is this record: the unbroken press of the pencil on the page, the line that never lifted.
The strangest thing about Superhero!? is that the True Line Cut emerged as the exact counterforce to the adaptive system. The same project that pushed branching identity, editorial agents, and version parity also generated its own opposite. This is, for all its complexity, a logical development.
Once the film can multiply versions and support increasingly fine adaptive structures, another pressure appears: the need for event-value, risk, and finality. Abundance creates two corresponding desires: the desire to explore all valid possibilities, and the desire to impose consequence.
The most important resource in post-scarcity is scarcity.
For Superhero!?, the True Line Cut functions as version 0.1 of a concept that will deepen in future productions. The viewer sees the indecision about Abigail’s character model playing out inside the film itself: the protagonist shifting appearance as the director searches for the right face. In a traditional film, this would be a continuity error. In the True Line Cut, it is the entire point. The process is the product.
If the process version and the polished version both exist as legitimate releases of the same film, then the notion of completion itself becomes an entirely parallel variable category.
Superhero!? proved that five parameters can vary within a single production while the film remains recognizably one authored work: protagonist identity, editorial process, visual register, and degree of finish.
The traditional animated feature is a fixed object. One cast, one cut, one score, one final version delivered to the audience as a fait accompli. The final cut conceals the lost possibilities that hovered around it. Superhero!? brings alternate possibilities to the surface and organizes them into a formal system. The work now has a core architecture and a set of variable states through which that architecture can be instantiated. The viewer activates one configuration at the start. The film runs. The architecture holds. The Adaptive Cinema engine thusly operates in practice.
The system forces new meanings. The protagonist variable proves that identity reshapes meaning even when the text stays fixed. The agentic editor proves that rhythm and coverage can be maintained across versions by a system that holds the whole runtime in view. The True Line Cut proves that the degree of finish itself is a choice, not a destination. Together they demonstrate the film has migrated from object to an architecture. The director does not make a film. The director defines the space within which a family of films can exist.
Conclusion
Superhero!? is the earliest sustained demonstrations of cinema as architecture. This system is reproducible and capable of sustaining emotional depth at feature length. The counterculture is here. It is decided. But even at its endpoint, the process has only just begun.
