The
Actor Who Never Came to Set
How AI De-Aging, Digital Replicas and Synthetic
Performers Could Rewrite Hollywood
A Next Horizon longread about the moment an
actor’s face, voice and performance can become a licensable digital asset.
| AI is turning an actor’s face and performance into something that can be scanned, preserved and recreated with remarkable realism. |
Imagine a major action film announced for
2032. A world-famous actor dominates the poster. The face is unmistakable. The
voice sounds right. Even the tiny habits are there: the half-smile, the pause
before a line, the familiar turn toward the camera.
Except the actor never came to the set.
There was no costume fitting, no twelve-hour shooting day, no reshoot months
later. Maybe the star licensed a digital replica years earlier. Maybe a body
double supplied the movement while AI supplied the face and voice. Maybe the
actor died before production even began, and the estate approved a new
performance.
That scenario is no longer pure science
fiction. Filmmakers already de-age actors, rebuild faces, clone voices and
negotiate rights to digital replicas. In 2026, an AI-rendered version of the
late Val Kilmer pushed posthumous performance into the present rather than the
hypothetical. At the same time, Hollywood's labor agreements and state laws are
becoming much more specific about who can authorize a digital likeness and how
it may be used.
So the real question is no longer whether
cinema can copy a face. It is how far that copy can go. Can an eighty-year-old
actor convincingly play thirty? Can a star license a younger digital self to
work without them? Can a deceased performer take on a new role? And could a
future studio build an entire film around famous faces whose owners never
physically appear on set?
First, “AI Actor” Can Mean Three Very Different Things
The phrase AI actor sounds simple, but it
covers several very different things. Keeping them separate matters, because a
digitally altered human performance raises different creative and legal
questions from a licensed digital replica or a performer who never existed at
all.
1. A digitally altered human performance
A real actor performs the scene, and
technology changes the appearance afterward - or increasingly, while the scene
is being shot. The face can be de-aged, aged, repaired or transformed while the
body language, timing and emotional choices still come from a human performer.
This is the least radical version of the idea because the acting begins with a
person.
2. A digital replica of a specific person
A digital replica is a realistic
representation of an identifiable person: their face, voice or both. It can be
built from scans, recordings, photographs, archived performances or material
captured specifically for that purpose. The important difference is that the
replica can potentially generate something the person never recorded.
A performer might record ten lines and
later authorize a model to create one more in the same voice. A body double
might play a scene while the star's licensed face is applied. With enough
source material, technical control and permission, a production can create
shots in which the celebrity audiences recognize was never physically in front
of the camera.
3. A fully synthetic performer
The most radical category is a performer
who is not a replica of any identifiable person. The face, voice, expressions
and public persona can all be designed. In July 2026, the synthetic character
Tilly Norwood was announced as the lead of the feature film Misaligned, moving
the idea beyond demos and social-media experiments. The creative limits are
still obvious, and the backlash from actors is real, but the debate has
changed: Hollywood is no longer asking only how to copy performers. It is
starting to ask what counts as a performer in the first place.
If you want to see how these digital performers fit into the broader VFX pipeline — from cleanup and relighting to generative video and digital humans — read our guide: The Invisible Crew Behind the Screen: How AI Is Rewriting Visual Effects.
AI De-Aging: The First Version of the Digital Actor
De-aging is the clearest bridge between
familiar visual effects and the idea of a digital actor. Cinema has made
performers look younger for decades; what is changing is the speed, flexibility
and amount of human labor required.
Traditional de-aging could demand
painstaking frame-by-frame work: skin cleanup, face replacement, 3D tracking,
compositing and endless manual correction. The results could be impressive, but
they were expensive and slow. They also exposed a stubborn problem: a young
face could still sit on a body that moved like an older person.
Martin Scorsese’s The Irishman became a
major milestone. Industrial Light & Magic developed a markerless
facial-capture system called FLUX so Robert De Niro, Al Pacino and Joe Pesci
could perform without facial markers or head-mounted cameras. Machine learning
assisted the VFX pipeline, but the acting itself remained theirs. The system
did not invent De Niro’s performance; it transformed his appearance while
preserving what he did on set.
Here pushed that idea further. Metaphysic's
real-time system let Tom Hanks, Robin Wright and other performers see younger
versions of themselves during production. That feedback matters. If the face
reads as thirty but the posture or timing still feels sixty, the actor can
adjust on set instead of discovering the mismatch months later in
post-production.
| AI de-aging and aging tools allow filmmakers to show the same performer across multiple decades without recasting the role. |
Why de-aging is more than a visual trick
Once a face can move convincingly backward
and forward through time, age becomes less rigid as a casting constraint. One
actor can play a character at 25, 45 and 75. Flashbacks do not automatically
require a younger look-alike. A sequel can revisit an earlier period without
replacing the face audiences associate with the role.
But age is not stored only in skin. It
shows up in posture, muscle tension, breathing, timing and energy. That is why
the best de-aging still depends on performance. A flawless young face on an
older physical performance can feel stranger than a slightly imperfect visual
effect on a body that moves believably.
The logical next layer is to alter more
than the face: body motion, posture and voice as well. When face, body and
speech can all be changed coherently, de-aging stops being a cosmetic effect
and starts looking much more like a reusable digital double.
The Digital Twin: When an Actor Becomes a Reusable Asset
A modern film production already captures a
remarkable amount of data from its stars: high-resolution photography, 3D
scans, facial scans, motion capture, dialogue recordings, alternate takes and
promotional footage. AI changes the value of that archive because it can turn
recordings that once sat on a server into material from which new shots, voices
or expressions can be generated.
That shifts the economics of performance. A
studio is no longer negotiating only for an actor's time. It may also be
negotiating for controlled access to a digital replica: what can be generated,
for which project, for how long, in which territories, for what fee and under
whose approval.
Consider a routine production problem. A
star finishes the film and leaves the country. Months later, one pickup shot
needs a new sentence. An authorized voice model could generate it. A stunt
professional could perform a dangerous sequence while the licensed face remains
consistent. A dubbed version could preserve more of the original vocal
identity. None of these uses requires handing the studio unlimited rights to
the actor; each can be separately negotiated.
The result is something cinema has never
really had before: an actor who can become partly asynchronous. Pieces of the
performance can continue after the person has left the building.
Voice cloning is already the clearest example
Voice is easier to separate from the body
than a complete screen performance, so voice cloning has moved faster.
Respeecher's work with Darth Vader showed how an authorized model could
preserve a recognizable character voice without asking the original performer
to record every future line in the usual way. Similar systems can reconstruct a
voice that has been lost, unavailable or deliberately licensed for new
material.
Voice also makes the idea of a licensable digital self easy to grasp. An actor could permit a model for dubbing, games, animation or one specific sequel while forbidding advertising, political speech or unrelated characters. The technology answers whether copying is possible. The contract answers what that copy is allowed to do.
Voice may be the most intimate part of a digital replica. Modern AI can reproduce not only a person’s tone, but also rhythm, accent and elements of emotional delivery. We explore the implications in more detail in When Anyone Can Sound Like You: AI Voice Cloning, Deepfake Songs and the Future of Music.
Can We Use an Actor’s Face and Voice After They Die?
Technically, yes. The harder questions are
who has the authority to say yes, what exactly they are approving, and how far
that approval should extend.
Cinema has been digitally resurrecting
performers for years. Peter Cushing's Grand Moff Tarkin in Rogue One is a
useful precursor, although calling it a modern generative-AI performance would
be misleading. Actor Guy Henry performed the scenes, and visual-effects artists
recreated Cushing's appearance on top of that performance. It was sophisticated
CGI and performance transfer, not today's text-to-video generation.
What has changed is scale. Machine learning
can automate more of the reconstruction, while generative systems can
potentially create speech, expressions and shots that were never captured in
the original archive.
Val Kilmer turned the hypothetical into a current legal test case
In 2026, the late Val Kilmer was digitally
recreated for As Deep as the Grave. SAG-AFTRA publicly stated that the
applicable union agreement and state law required estate consent, and that the
family’s permission had been obtained. Reports said the production used
archival images, recordings and other source material to create the new
performance, with Kilmer’s estate involved and compensated.
The important part of the Kilmer case is
not simply that a deceased actor appears on screen; cinema has done that
before. What is new is the emerging expectation that a posthumous digital
performance is something to be licensed, negotiated and governed, not treated
as a one-off VFX exception.
That creates room for much finer choices. A
family might approve one film because it aligns with the actor's known wishes,
then reject every later request. Performers may increasingly leave instructions
about which characters can continue, whether the voice can be licensed
separately from the face, which commercial or political uses are off-limits,
and who gets final approval after death.
| Once an actor’s likeness and voice become digital assets, questions of consent, licensing and inheritance become central. |
The law is becoming much less permissive
California now has some of the clearest
guardrails. AB 2602 restricts contracts that grant overly vague rights to
create new performances from a living performer's digital replica, especially
when the intended use is not reasonably specific and the performer lacked
appropriate representation. AB 1836 addresses certain uses of digital replicas
of deceased performers without consent from the relevant rights holder, subject
to statutory exceptions.
New York has also strengthened postmortem
publicity protections around digital replicas. The practical message for
studios is simple: an old film, photograph or recording may be technically
useful source material, but it does not automatically make the person's
identity free to reuse.
There is still no single U.S. federal rule
covering every case. The NO FAKES Act of 2026 would create a nationwide right
over highly realistic digital replicas of a person's voice and visual likeness,
but as of September 2026 it remains pending legislation. The Senate Judiciary
Committee advanced the bill in June; it has not completed the legislative
process.
Europe adds another layer: disclosure.
Article 50 of the EU AI Act began applying on August 2, 2026 and introduces
transparency duties for certain AI-generated or manipulated content, including
deepfake-like material. Film and artistic contexts are not treated exactly like
deceptive impersonation, but the direction is clear: audiences and regulators
increasingly expect synthetic media to be identifiable as synthetic.
There is no universal answer yet. Rights
can change with the jurisdiction, contract, union agreement, estate plan, type
of production and even the way the replica is presented. A studio that can
technically rebuild a performer still has to establish that it is allowed to do
so.
Could Hollywood Make a New Film With Famous Actors Who Never Participate?
This is the question that sounds most like
science fiction, yet the answer is increasingly: yes, under the right technical
and contractual conditions.
Suppose a studio wants to make a new
detective film starring three famous actors, but none of them will come to the
set. There is not one way to do that. There are several, and they differ
dramatically in how much of the original performer remains in the final work.
Level 1: The actor performs once, then the replica extends the work
The star records reference material,
approves a scan and perhaps performs a handful of key scenes. AI and VFX extend
that work into pickups, foreign-language dialogue, stunt substitutions or
alternate shots. This is closest to current practice because the actor still
clearly anchors the performance.
Level 2: A body double performs while the star appears digitally
A stand-in provides blocking, physical
interaction and movement while an authorized digital face is applied. The
celebrity may contribute no physical acting to that particular scene. One
person supplies the body and timing; another person's licensed identity
supplies the recognizable face.
Level 3: The actor licenses a replica but performs nothing for the new film
Here the actor does not perform for the new
film at all. If a performer has already created a digital replica - or
explicitly authorizes one to be built from approved material - a contract could
permit new screen material to be generated without a fresh physical
performance.
The film would still need a source for
acting choices. Those choices could come from another performer, a
performance-capture specialist, animators, the director, or increasingly from
generative systems that create facial motion and body movement. The celebrity's
face and voice would become one layer in a performance assembled by other
humans and machines.
Level 4: An entire cast is synthetic
At the furthest end, there is no celebrity
to copy. Every performer is designed from the start. The director can change
age, accent, appearance or delivery without negotiating an existing star's
identity. That can be legally simpler than an unauthorized replica of a real
person, although questions about training data, resemblance and labor do not
disappear.
Short, controlled shots are already far
easier than an emotionally convincing two-hour lead performance. A feature film
demands consistency from every angle, believable interaction with props and
other actors, stable wardrobe and lighting, accurate eye-lines, subtle
expressions and a coherent dramatic arc. Generative video is improving quickly,
but cinema is an unforgiving test because audiences spend their entire lives
studying human faces.
The real technical challenge is no
longer producing five convincing seconds. It is preserving one believable
person - with recognizable intentions and behavior - across an entire story.
The Actor’s New Contract May Include a “Digital Self”
Once a digital replica can do useful work,
AI stops being only a VFX issue and becomes a contract issue. The central
question is no longer just how a replica is made, but what rights travel with
it.
SAG-AFTRA’s 2026 TV/Theatrical agreement,
effective July 1, 2026, strengthens rules around digital replicas and synthetic
performers. The union distinguishes between a replica of a specific person and
a synthetic performer that does not represent an identifiable individual. It
also emphasizes consent, compensation, security of biometric data and
restrictions on using synthetic performers as substitutes for human work.
For a major star, a future deal may
therefore contain two parallel businesses. One is familiar: payment for showing
up and performing. The other is newer: a tightly defined license for the
digital identity.
That license can become extremely specific:
may the studio generate new dialogue? Create a younger version? Use the replica
in sequels, games or advertising? Keep the biometric data after the project
ends? Transfer the rights if the company is sold? Let heirs approve future
uses? Use the replica during a labor dispute? The important shift is that
'permission to scan' and 'permission to perform forever' are not the same
thing.
For top stars, controlled replica licensing
could become a new income stream. For performers with less bargaining power,
the same technology creates the opposite danger: being pressured to surrender
face or voice rights cheaply, early in a career, under terms that are difficult
to reverse.
| A future production may rely on a body double, motion capture and AI enhancement while the final on-screen performer is a licensed digital version of someone else. |
Who Is Actually Acting When the Star Is Digital?
Once the layers of a performance can be
separated, the old answer - 'the actor did it' - becomes surprisingly hard to
defend.
Imagine a stunt performer supplying the
body, an animator refining the eyes, a director deciding the emotional rhythm,
a voice model generating the words and a generative system adding
micro-expressions. The final face belongs to a famous actor who never visited
the set.
Who, then, gave the performance?
The celebrity contributes identity,
reputation and a library of recognizable mannerisms. The body double
contributes physical acting. The director shapes intention. VFX artists and
model operators control execution. AI synthesizes details. What used to feel
like one person's performance can become a pipeline with authorship distributed
across many people and systems.
That does not automatically make the result
worse. Cinema has always separated pieces of performance through stunt doubles,
ADR, stand-ins, editing, animation and visual effects. The new psychological
break is that audiences are used to treating a recognizable human face as
evidence that this person actually performed what they are seeing.
Digital replicas loosen that link. A face
can remain proof of identity while ceasing to be proof of participation.
Would Audiences Accept a Dead Movie Star in a New Film?
The technical question may turn out to be
easier than the cultural one.
A posthumous appearance can feel respectful
when it completes a role, preserves a character or follows wishes the performer
made while alive. The same technology can feel exploitative when a person is
converted into an endlessly reusable commercial asset with no meaningful
connection to choices they ever made.
Scale and context matter. A ten-second
tribute is not the same as a new two-hour starring role. A performer who
explicitly licensed future use leaves a different ethical situation from
someone who never imagined that old footage could generate new dialogue. An
estate-approved dramatic appearance is different again from making a dead actor
sell a product or appear to endorse a political message.
There is a creative danger too: nostalgia
can become a business model. If studios can endlessly revive the most familiar
faces of the twentieth and early twenty-first centuries, they may have less
incentive to develop the next generation of stars. A technology built to
preserve film history could end up trapping popular culture inside it.
That is why the long-term alternative may
not be 'old stars forever' at all. Studios may eventually prefer synthetic
performers designed for digital production from the beginning: no estate
negotiation, no aging, no scheduling conflicts and no pre-existing biography to
protect.
Will AI Replace Actors? The Real Risk Is More Specific
'AI will replace actors' makes an
irresistible headline. It is also too vague to be useful.
Lead dramatic performances are among the
hardest jobs to automate because audiences care about authenticity, star
identity and microscopic emotional choices. Other layers of screen work are
easier to separate and automate: background crowds, temporary voices,
previsualization, stunt substitutions, digital doubles, pickups, localization
and some low-budget advertising.
At the same time, the technology creates
new work. Performers may drive the movement of digital characters. Actors may
license voice models. Motion specialists may perform for faces that do not
exist. Directors may cast one human specifically because that person's movement
will later be transformed into several digital identities.
The economic fight is therefore less about
whether humans vanish than about who gets paid when a human performance becomes
data, and who keeps control when that data can keep producing work.
From Invisible VFX to Licensed Performers
Near term: AI disappears into the workflow
The first wave is likely to feel almost
boring: cleaner de-aging, faster face replacement, better lip synchronization,
more natural dubbing, digital stunt doubles and cheaper pickup shots. Much of
it will be invisible. Audiences may watch more AI-assisted performances without
being able to point to the exact frame where AI entered the process.
Next phase: actors build parallel digital careers
Digital-replica clauses are likely to
become routine for major performers. Some actors will reject them entirely.
Others will license tightly controlled versions of themselves for a franchise,
game, dub, commercial or sequel. Estates will increasingly manage face and
voice rights as valuable assets, even though the precise legal category still
varies by jurisdiction.
Longer horizon: casting may include a set of identity rights
If long-form character consistency, motion
generation and synthetic dialogue continue improving, a celebrity could appear
throughout a film without giving a new physical performance. At that point,
casting may mean choosing not only a person, but a licensed package of face,
voice and approved digital behavior.
Fully synthetic performers could develop
audiences of their own. They need not age unless the story requires it. They
can speak any language and appear in multiple productions at once. Yet there is
a paradox here: the easier perfection becomes to manufacture, the more valuable
genuinely human unpredictability may feel.
| If digital replicas recreate real stars, synthetic performers go one step further—creating photoreal screen actors who never existed at all. |
The Bigger Question: Is a Face Still Proof of a Performance?
For most of cinema history, a familiar face
functioned as a kind of evidence. If Robert De Niro appeared on screen,
audiences could reasonably assume that Robert De Niro had performed the
material being shown, even if stunt doubles, editors and VFX artists helped
build the final scene. AI can break that assumption.
A screen actor can now be present without
having been present. A voice can say words never recorded. A thirty-year-old
face can belong to an eighty-year-old performer. A deceased actor can receive a
new role. A celebrity can license a digital twin while the human version stays
home.
Every one of those paths is becoming more
technically plausible. At the same time, the legal trend is toward more
permission, not less: clearer consent, narrower contracts, estate control and
disclosure. Culture is the unresolved part. Audiences have not yet decided
where assistance ends and impersonation begins.
That may be the question that matters most
for AI in Hollywood. Not whether a synthetic performance can look real - it
increasingly can - but whether viewers still care who actually made the choices
behind the face.
If they do, human performance may become
more valuable precisely because it is scarce and attributable to a real person.
If they do not, a movie star could start to resemble a franchise: a face, a
voice, a style and a bundle of licensed behaviors that can keep generating
screen appearances long after the person leaves the set - or the world.
The future of acting may be decided
less by a contest between humans and AI than by a negotiation over identity:
who owns the right to keep performing as you when you are no longer there?
FAQ: AI Actors, Digital Replicas and Hollywood
Can AI make an old actor look young throughout an entire
film?
Yes, within limits. AI-assisted de-aging is
already used in feature films, and real-time systems can show filmmakers
younger versions of actors during production. The difficult part is not making
one face look young in one shot; it is keeping body movement, lighting, hair,
skin, continuity and performance believable across an entire film.
Can filmmakers use the face or voice of an actor after the
actor dies?
Sometimes, but recordings alone do not
grant permission. Rights vary by jurisdiction. In major production centers such
as California and New York, laws increasingly protect posthumous digital
replicas, while union agreements can impose additional consent and compensation
requirements.
Could an actor license a digital twin to appear in films
without going to the set?
Yes, in principle, and parts of that
workflow already exist. The exact use depends on the contract and consent. A
licensed face or voice can be combined with a body double, performance-capture
actor, animator or generative system, allowing the recognizable star to appear
without physically performing that scene.
Is a fully AI-generated actor legal?
A synthetic character that is not
identifiable as a real person may avoid some likeness-rights problems, but it
can still raise questions about training data, accidental resemblance, labor
agreements and disclosure. In union-covered U.S. productions, SAG-AFTRA's 2026
agreement places substantial restrictions on using synthetics to replace human
performers.
Will AI actors replace human movie stars?
Probably not in a simple one-for-one way.
The nearer-term shift is likely to happen layer by layer: digital doubles,
dubbing, pickups, background work, stunt substitutions and some low-budget
content, while major human performances coexist with more sophisticated
replicas and synthetic characters.
Comments
Post a Comment