The AI Restoration of ‘The Magnificent Ambersons’
(From “Post-Scarcity Cinema: Essays in AI Film Theory” Volume 2 of The New Machine Cinema)
Hooroo Jackson presents the ultimate concept for AI cinema restoration, the most famous ‘almost’ in the history of cinema: The Magnificent Ambersons.
It has become one of Hollywood’s greatest tragedies. Welles follow-up to Citizen Kane (1941), unanimously chosen as the greatest film ever made, was never given light. For decades, film historians, Welles scholars, and cinephiles have longed for its return, despite the cold reality: the lost footage was likely burned.
The advent of artificial intelligence has ushered in a new age, one in which restoration is no longer limited to what physically survives. In fact, it presents an entirely new framework so big that it seems mundane: whatever can be thought, can be seen.
You see The Magnificent Ambersons is not just a case study in cinema history, or Welles himself; it is the very proof of concept for AI-driven restoration, a process that could redefine how we approach meddled and lost cinema.
To this point, AI is already being used in film restoration extensively, with upscales, sound enhancement and color–but this essay poses what it has not yet been used for, that of recreation.
The History
Welles’ original cut of The Magnificent Ambersons ran approximately 131 minutes. In his absence—he was in Brazil at the time, filming It’s All True—the studio panicked over poor test screenings and seized control of the film, metaphorically locking him out of the editing room.
They removed over a third of the film, including crucial character moments and a more thematically faithful ending. Entire subplots were erased, performances altered, and Welles’ signature long takes hacked into conventional coverage.
The situation cast a haunting shadow over the history of cinema. Only fragments of Welles’ vision remained: production notes, script notes, and recollections from cast and crew.
It was film history’s greatest ‘what if’. Now this has changed to: what if we can bring it back?
First, there is precedent for meddling with Welles’ unfinished projects. Filmmakers love getting in his business.
For Don Quixote, Welles worked on it for over 20 years (from the 1950s to the 1970s), but he never completed it. It was famously “restored” (or, more accurately, assembled) by Jesús Franco and released in 1992.
The result was controversial—Franco restructured the film in ways that many Welles scholars found deeply unfaithful to his vision. He also incorporated footage that was never intended to be used, and the final product felt more like an interpretation than a true restoration.
This is the inherent problem with the Magnificent Ambersons AI restoration—pedigree. On the surface, it would require historians, researchers, and a filmmaker to oversee it. It could not be done casually. Except there is nothing casual about the capabilities of the New Machine Cinema.
More successful was his 2018 posthumous release The Other Side of the Wind, the final film of Orson Welles. This was overseen by Peter Bogdanovich, Welles’s personal friend, who also acted in the film.
That effort, like Quixote, relied on existing footage—and a more prestigious auteur filmmaker with a Netflix budget behind him.
AI offers something radically different. From my essay The Living, Breathing Cinema:
“Eventually there might be no pure cinema of old left, only the best versions of them born from the collective imagination. Fan-fiction will become the dominant cinema. Collectors will be forced to compare versions of films to determine who has the real one. People will invent old films that never existed and try to smuggle them into the public record.
“Classics will be de-faced with alternate renditions with such frequency that opposition will defer to this as a formal artform, much like sampling in music. Who wouldn’t want to stand side-by-side with Welles?
“Consider the beginnings of AI where it wasn’t original works catching on with people but AI versions of popular franchises. It begins exactly how it ends, with mimicry.”
I take this further from the theoretical into the actuality. What is the best way to prove the power of AI? To go into the past and right a wrong? I bring forth, we bring The Magnificent Ambersons back to life. It was always building to this.
AI Restoration: What Can Be Done?
Performance Recreation
From my deep experience working with machine performance as outlined in my article The Infinite Performance, I realized AI voice synthesis and deepfake acting have reached a point that is not only at parity with human actors, but has in some ways, aesthetically surpassed them.
Machine learning can be trained on Joseph Cotten, Tim Holt, and Agnes Moorehead’s line deliveries, rebuilding the performances from the inside out. It will never be the performances they acted out, but no one would notice.
Visual Recreation
Here we get to the bottleneck as it stands, though it is closer. Black and white cinematography, with its unique qualities, could be achievable through AI cinema visuals. But as it stands—in this moment—it would just be slightly too uncanny. Believable for punch-ins, extending scenes, certainly, possibly even getting away with it for a scene. But not for forty minutes. Not yet.
Staging
From the outset of AI cinema, I attempted to stage my AI feature films through blocking as if they were real films. I was not prioritizing visuals but the performance as the very key to the motion picture experience. The mise-en-scene. This in retrospect made my films ahead of the field: they were performance-based, entirely machine movies.
It has only got better up to A Very Long Carriage Ride where full fledged staging happens scene after scene. So to say, all styles of blocking and coverage are possible in AI cinema, right now. Generative AI tools combined with existing stills and production notes could simulate the lost footage in a way that feels organic.
Editorial Integrity
The fingerprint of the auteur however, the intangible, cannot be recreated by a director overseeing the Magnificent Ambersons AI restoration. Naturally, he will bring his own stamp to the film instead. But AI as a collaborator can arrive us closer.
In my AI features, I experimented with AI-editing; that is editing more in the classic sense, the paper editor, during a time when film was cut with scissors and tape. We are not yet at a place we could hand the film over and tell the AI, have at it.
It will still require an auteur to oversee the restoration. It is why I did not pursue Carriage Ride as the first fully AI edited movie. When AI can watch the film over your shoulder and guide your hand. When AI can take control of your screen. That is where things open up in a big way.
That is where we can now approximate Welles’s actual directorial hand. Even then, it should be overseen. I never withdraw the importance of the auteur, else we enter in The Automated Film.
Now what would that look like? AI can analyze Citizen Kane, Touch of Evil, and even the surviving parts of Ambersons to reconstruct his pacing and shot structure. It will have it down to a mathematic science.
Unlike traditional restorations, this would no longer need to be a “best guess” at what Welles wanted. Every transition, dissolve, and camera move could be informed by his own patterns, rather than an external interpretation. We are closer than anyone realizes.
The Precedent: Caligula’s Ultimate Cut
The recent Caligula: The Ultimate Cut provides an instructive parallel, forecasting AI cinema restoration. We are basically already there.
The Caligula restoration took a notoriously compromised film, Caligula, funded by Playboy and directed by Tinto Brass. The film, in the 70s, was re-fashioned into nudy cult cinema.
The Ultimate Cut posited itself to refashion all the original footage to match Gore Vidal’s original screenplay exactly. Famously, the producer of the restoration reached out to Brass who wanted nothing to do with it.
At the time, I assumed this would be remarkably controversial. However, aside from grumbling online and a general feeling of distaste in a how dare you sense, it wasn’t.
In fact, the most controversial part of the restoration was not the restoration itself, it was the fact they utilized CGI, breaking the purity of the restoration by using tools not available in the original timeline.
The final result, Caligula the Ultimate Cut screened in Cannes alongside the endorsement of its star Malcolm McDowell, who remarked that he felt relieved his work could finally be seen.
This whole event was a notable anomaly forecasting what is coming. History will be malleable, available to all of us. A film reconstructed by people with no involvement in the original production will be common as rain.
But The Magnificent Ambersons AI restoration would bring the baggage of the Caligula Ultimate Cut times one thousand.
The visceral distaste, the how dare you, would be there ten-fold.
The purity argument, would lead to calls to boycott or worse.
The AI recreated machine performances, further outrage.
My first thought: How Wellesian.
There is no aspect of Ambersons that would not be controversial. At first. Until it’s normalized, perhaps starting smaller, and even bigger than that.
Beyond Ambersons: The Future of AI Restoration
The article presents the concept through its Holy Grail. We are setting the precedent for AI-driven film restoration as a discipline, bringing the question: if Ambersons can be reconstructed in a way that respects its legacy while embracing the inevitable future of AI, it paves the way for other lost or mutilated films. Imagine:
Erich von Stroheim’s Greed restored to its nine-hour cut.
Eisenstein’s ¡Que Viva México! completed as he envisioned.
Hitchcock’s The Mountain Eagle, a lost silent film, reborn.
Historic wrongs could be righted, such as films lost through political censorship:
Andrei Tarkovsky’s The Confession, partially shot in the late 1970s, but confiscated by Soviet authorities.
What about famous unmade films? Directors such as David Lynch and Peter Greenaway have a tranche of unmade films (Lynch’s “Ronnie Rocket”; Greenaway’s brilliant unmade scripts including “The OK Doll”, “Four Storms & Two Babies”, “Eisenstein in Hollywood” and “Eisenstein in Switzerland”).
Steven Spielberg is mounting a massive television miniseries production of Stanley Kubrick’s Napoleon, known as the greatest movie never made, perhaps missing that Kubrick was not a television director but a filmmaker. I wonder if such endeavors were premature, that AI could give us the actual Kubrick film, not the watered down Hollywood facsimile.
Cinema has been discretely moving along these lines for years, with techniques that have been strangely accepted with almost no resistance by traditionalists. De-aging, CGI reconstruction, and even AI voice replication for actors are commonplace. The logical next step is using these same tools to reclaim films thought lost to history.
The Magnificent Ambersons proves as the ultimate case study for AI cinema restoration.
As always, there is a fringe. Every piece surrounding such an endeavor is being rapidly chipped away. And as I spoke about the concept of the malleable film a new question emerges not of historic preservation but that of cinema existentialism: Why would we need Welles’ Ambersons when we can have AI’s Ambersons?
Conclusion: The Inevitability of AI Restoration
In understanding the continuum, I keep returning to all the lines in the sand that were once drawn only to be crossed with almost no resistance. Audiences quite generously offer artists and technicians carte-blanche to move the evolution of cinema forward. So the question is not whether The Magnificent Ambersons will be restored using AI, the question is how soon.
We are field-testing the future of the New Machine Cinema, and with post-aesthetic machine pleasures, I see no reason why every corner of possibility should not be explored to the fullest extent. This also means that in evolving forward, the living breathing cinema takes us backward too.