Ghosts in the Machine: How AI Is Quietly Rewriting the Performances You Thought You Knew
There's a version of a movie you watched last year that doesn't quite match the one streaming today. Not because someone caught a plot hole or fixed a continuity error — but because an algorithm decided an actor's eyebrow moved wrong in a pivotal scene, and now it doesn't anymore. You'll never know. That's kind of the point.
AI-driven performance alteration is one of the quietest revolutions happening inside Hollywood right now, and the industry isn't exactly rushing to hold a press conference about it. Studios are using machine learning tools to clean up dialogue, smooth out facial expressions, extend scenes with actors who were unavailable for reshoots, and in the most extreme cases, reconstruct entire performances from people who are no longer alive. The technology is real, it's already deployed, and most of us are consuming its output without a single disclaimer.
What's Actually Happening on Screen
The most visible version of this conversation happened with Indiana Jones and the Dial of Destiny, where de-aging technology gave us a younger Harrison Ford in the opening sequence. People talked about it. Critics wrote about it. Ford himself had opinions. That's the sanctioned, marketed version of AI performance work — the kind studios are comfortable letting you see.
What they're less comfortable discussing is the subtler stuff. Dubbing and lip-sync correction has become a major AI use case, where tools can reshape an actor's mouth movements in a foreign-language dub so precisely that the performance looks native. Netflix and other streamers have leaned into this for international markets. The actor on screen didn't deliver that version of the scene. A machine did, using their face.
Then there's what industry insiders sometimes call "performance cleanup" — using AI to remove a momentary break in concentration, a slight mispronunciation, a blink at the wrong time. Things that would have required an expensive reshoot or a careful edit around are now fixed in post with tools that can reconstruct a few frames of facial performance from scratch. Subtle. Invisible. Done.
The Dead Don't Get a Vote
The cases that carry the most ethical weight involve actors who've passed away. The use of CGI to finish Paul Walker's scenes in Furious 7 was handled with visible care and family involvement. It felt like a tribute. But the industry has moved well past that level of craft and conversation.
With today's AI tools, studios can generate new dialogue in a deceased actor's voice, animate their likeness through existing footage, and create performances those actors never gave — for projects they never agreed to be part of. The legal landscape is a patchwork. Some estates have signed deals granting studios broad rights. Others are fighting in court. And in plenty of cases, the work is happening in legal gray zones that nobody's been brave enough to fully challenge yet.
California passed the Protecting Individuals from Digital Replicas Act in 2024, and SAG-AFTRA's recent contract battles brought AI consent language to the forefront of labor negotiations. But legislation and union contracts are reactive by nature. By the time the rules catch up, the technology has already moved three steps ahead.
What This Does to the Craft
Ask any working actor what they think about this, and you'll get an answer that sounds something like controlled panic dressed up as professionalism. The craft of acting — the thing that makes a performance worth watching — lives in the imperfections. The hesitation before a line. The way someone's jaw tightens when they're holding back emotion. The micro-expressions that happen because a human being is genuinely feeling something in a moment.
When AI smooths those out, or reconstructs them based on what a training model thinks "good" looks like, it's not enhancing a performance. It's replacing it with an average. A consensus version of what that moment should be, drawn from thousands of other moments. Which sounds fine until you realize that the greatest performances in cinema history are great because they don't look like anything else.
There's also a downstream effect on the actor's relationship to their own work. Imagine delivering what you consider your best performance in a film, then watching the finished product and realizing a machine quietly rearranged your face to match a different emotional register than the one you chose. Your name is still in the credits. Your face is still on screen. But the performance isn't entirely yours anymore.
The Audience Deserves to Know
Here's where I land on this: transparency isn't optional. It's a baseline.
When studios use AI to alter a performance — even slightly, even with good intentions — audiences have a right to know that what they're watching has been modified beyond traditional editing. We already have standards for this in other areas. Heavily retouched magazine photos carry disclosure requirements in some jurisdictions. Food advertising has rules about what can and can't be digitally altered. Why should the performances of human beings on screen get less protection than a burger in a fast food commercial?
The entertainment industry's current approach is to treat AI performance alteration as a production detail — no different from color grading or sound mixing. But it's not the same thing. Color grading doesn't put words in an actor's mouth or reconstruct their emotional expression. Sound mixing doesn't generate new footage of someone who died in 2013.
Where This Goes Next
The optimistic read is that AI performance tools, used transparently and with genuine consent, could open up interesting creative possibilities. Restoring damaged archival footage. Allowing actors to revisit and refine performances years later. Creating multilingual versions of films that feel authentically performed rather than dubbed.
The pessimistic read — and honestly, the more realistic one given Hollywood's track record — is that cost-cutting will drive adoption faster than ethics can govern it. Why pay for a reshoot when you can patch the scene in post? Why negotiate with an actor's estate when a lookalike's footage plus AI can get you close enough? Why invest in new talent when you can license the likeness of a beloved star who passed away decades ago and keep generating content indefinitely?
The stories that matter, the ones this site has always cared about, are the ones told by human beings with something real at stake. The moment we start outsourcing that humanity to a machine without acknowledging what we've lost, we're not making entertainment anymore. We're manufacturing the appearance of it.
And audiences, whether they can articulate it or not, will eventually feel the difference.