LAS VEGAS – A new category of “restoration” is moving from postproduction suites into ticketed exhibition: the use of generative artificial intelligence to expand, alter, or “complete” legacy films-sometimes by creating new images and performances intended to sit alongside, or directly on top of, the original work.
The practice has been propelled into mass visibility by The Sphere, the Las Vegas venue whose AI-modified presentation of 1939’s The Wizard of Oz was designed to fill its 160,000-square-foot interior display plane, part of a 16K wraparound LED system built to immerse an audience of up to 20,000 people.[2] Since the show opened in August 2025, the venue’s Oz presentation has sold more than 2.2 million tickets, a level of demand that has reframed what was once a specialist debate-archival preservation versus artistic revision-into a question with clear commercial implications.
At the same time, an ambitious, yearslong attempt to reconstruct missing portions of Orson Welles’ 1942 film The Magnificent Ambersons is underway outside the participation of the film’s rights holder, raising a different set of issues: estate engagement, studio consent, deceased performers’ likenesses, and what “completion” means when the underlying footage was destroyed.
The conversation is not new. It is, however, newly pressurized by a technology that can generate credible moving images at scale-and by an industry that increasingly monetizes libraries through reissues, remasters, premium experiences, and venue-based exhibition. It is also unfolding into a regulatory environment that, in Europe at least, is beginning to treat AI-generated or AI-manipulated moving images as a distinct category that must be disclosed to viewers under the EU’s emerging AI Act transparency rules.
A recurring argument, revived by a different tool
In 1986, the late critic Vincent Canby published a sharply worded critique of the then-rising trend of film colorization, arguing that applying modern visual flourishes to black-and-white classics “desecrated” those films. He wrote that “nobody connected with the original[s]…had anything to do with this artistic revisionism” and that “of the half-dozen [colorized] films I’ve seen to date, all but one were virtually unwatchable.”
Canby’s objection was ethical as well as aesthetic: a film is an artwork anchored to its time of creation, and revision-especially revision executed without the original artists-risks recasting history as a product update.
He also pointed to money as a driver. Canby wrote that a colorized, or “tinted,” version of the 1937 film Topper had made $1 million, mostly in TV syndication dollars, by the time his essay was published-an early signal that commercial incentives could push alteration from novelty into standard practice.
Colorization later fell out of favor as a formally accepted practice. The backlash was a reminder that industry-wide rejection of a technique can be swift when a critical mass of creatives and audiences conclude that the intervention has crossed a line.
Generative AI presents a harder case, in part because it has uses well beyond image manipulation-and because its adoption is arriving amid heightened economic expectations about what the technology can do for production, marketing, localization, and catalog exploitation. For studios and streaming platforms that now treat libraries as balance-sheet assets, the pressure to “enhance” or “unlock” value from familiar titles is far greater than it was in the era of one-off colorization deals.
“In moving-image history, these debates about technological change and its impact on creativity or labor or our understanding of the past have resurfaced at various times,” said Dr. Charles Acland, a distinguished professor of cultural theory and film studies at Concordia University. “But we also live in an economy where there is such extraordinary hype around what gets called AI … that it puts a different kind of pressure on these discussions and debates. Colorization is a good comparison, but it didn’t have the same sweeping social and economic impact of something like generative AI – so there’s more at stake in how we sort through what we’re going to accept and valorize.”
The Sphere’s Oz: immersive exhibition meets catalog IP
The Sphere’s project is not positioned as a conventional remaster. It is a venue-specific presentation built around the practical problem of scale: how to adapt a film composed for a 4:3 screen to a display environment with radically different dimensions and viewing conditions, including a hemispheric screen and synchronized in-seat effects designed to behave more like live spectacle than like a traditional screening.[2]
A source familiar with the venue’s Oz presentation said that “using AI was the only way to maintain the integrity of the original film.” According to that account, the endeavor began with the challenge of translating the original material into the venue’s unique format, and AI models were trained on the original source material.
Critics and cinephiles have been split on the outcome, with some objecting to the digital addition of new performances and visuals. The same core concern raised in earlier debates about colorization has resurfaced here in updated form: director Victor Fleming and other late artists who made the 1939 film had no say in how their work would be used as training material for the generation of new images and performances, or in how prominently those generated elements would be interwoven with the film’s original frames.
The source familiar with the presentation said that “respecting the original was a priority.” While the venue frames the show as an “experience” rather than a replacement of the canonical film, the distinction can be hard for audiences to parse once new sequences are projected at the same scale and apparent resolution as studio-era images.
Whatever the artistic verdict, the audience response has been measurable. Selling more than 2.2 million tickets since August 2025, the Sphere’s Oz underscores how a widely available film can be re-monetized as an event when the presentation format is sufficiently differentiated-and when the venue itself is part of the attraction.
From an industry perspective, that matters because it offers a template that sits adjacent to theatrical distribution and traditional repertory exhibition. Rather than relying on bookings across multiple cinemas, the model centers on a controlled, premium environment with ticket revenue tied to a specific technology-forward experience. That commercial proof point is likely to motivate further experimentation, particularly as rights owners and exhibitors look for new ways to extract value from mature intellectual property-and as governments and regulators begin to ask whether viewers should always be told when what they are watching is, in part, synthetic.
A different kind of “completion”: rebuilding The Magnificent Ambersons
Running parallel to venue-based reinvention is a more granular, historically loaded project: an effort to reconstruct missing parts of The Magnificent Ambersons, the 1942 Orson Welles family drama that was cut down and reshot by RKO against Welles’ wishes, with more than an hour of unseen footage eventually destroyed.
Welles spoke decades later about wanting to reshoot the original ending-RKO’s released version was “decidedly sunnier,” according to the account-and to revive the dismantled final act. Admirers have long brainstormed how that might be done on his behalf. The existing version of Ambersons is still revered, even as Welles’ preferred cut is considered among cinema’s great “lost films.”
Edward Saatchi, founder of Fable Studios, is spearheading the current reconstruction effort, drawing on a body of surviving evidence. That includes set photos and a “cutting continuity,” described as a document that lays out how each shot leads into the next, alongside Welles’ comments over the years.
Before this AI-driven approach took shape, filmmaker Brian Rose spent years creating the lost scenes through animation, using the available materials as a blueprint.
“The thought was always in the back of my mind that, ‘Yeah, this will be my thing, and then somebody else will come along and do something else, or maybe technology – AI – could do a seamless re-creation,’” Rose said. “The only thing that I completely missed the ball on was how quickly the technology would come around.”
Saatchi, who has said he has been obsessed with Ambersons since childhood and now runs a generative AI company, contacted Rose about combining efforts. The goal, as described by those involved, is to use the existing blueprint to recreate the missing sequences with precision rather than inventing new material.
The project is proceeding without the participation of Warner Bros., described as the owner of the property and much of RKO’s back catalogue. That absence is not just symbolic: it means the work is unfolding without the normal studio restoration pipeline, insurance structures, or contractual guardrails that tend to govern how back-catalog films are handled.
One filming session with real actors has already been completed, with missing shots re-created. Two additional filming sessions are planned. Performers’ work is intended to be superimposed onto the original actors’ likeness in the film with the help of AI. The stated hope is for the final filming portion to be completed within “a year or so.”
Saatchi frames the effort as a test with a high bar for success and an obvious threshold for failure. “Some people are going to be like, ‘Oh no, this is terrible,’ and some people are going to be like, ‘OK, so wait, I’m going to be defending the butchering of this person’s vision and not even think about how to actually show what he was intending?’” he said. “If I was to guess, the majority of the people that I’ve talked to across many different areas within Hollywood are of the view that … if it’s genuinely seamless and you can completely justify what you’ve done in terms of the decisions, maybe it is a service to cinema to see what the greatest filmmaker of all time at the height of his powers made.”
He added that “established directors” have expressed interest, though he said he could not name them at this stage. Saatchi also said he has not heard from Martin Scorsese, noting that film historian Robert Harris has previously said Scorsese once expressed interest in reconstructing Ambersons. Saatchi said he hopes to attract further filmmaker participation as the work progresses, as a way to build legitimacy within cinephile circles.
Estates, rights holders, and a consent problem the technology cannot solve
The Welles estate was not approached about the project before it was announced, a lapse Saatchi has said he regrets. Orson Welles’ daughter Beatrice Welles runs the estate and issued a statement that captured both broader anxiety about generative technology and a narrower skepticism about intervening in her father’s work.
“Like most people I’m quite terrified of AI and in many ways wish it had never been invented.”
In the same statement, she said: “As far as Ambersons is concerned, I’m a purist and wish that originally it had never been tampered with. Nobody and nothing can think like my father. In regards to what Fable Studios is doing, while I am skeptical I know they are going into this project with enormous respect towards my father and this beautiful movie and only for that I am grateful.”
Saatchi has said he has since spoken with the estate and that “they’ve been really open-minded about it. I think, for them, it’s about the intention and how it works structurally.” He described the effort as “an academic exercise” rather than a commercial one like the Sphere’s Oz presentation. “It’s kind of a terrible turning point in cinematic history that we’re trying to undo to some extent,” he said.
Even with estate dialogue, the project surfaces a dilemma with limited precedent: the use of deceased actors’ likenesses to generate new performances without their consent.
“Even this, which I feel is clearly done with the best of intentions, has things that are kind of ethically indefensible,” Saatchi said, referring to generating new performances using deceased actors’ likenesses without consent. “There is no argument for why that’s defensible – other than that’s the only way to do it.” He added that anyone pursuing AI-driven restoration should acknowledge upfront: “This is not a wholly good thing.”
The tension here is not simply philosophical. It is structural. Legacy films have multiple stakeholders-studios or other corporate rights holders, estates, archives, and audiences-and the clearance logic that governs traditional restorations does not automatically map onto AI-mediated “completion,” where the intervention can include newly generated imagery that appears to be part of the original. In the United States, recent union negotiations have begun to formalize consent and compensation rules for AI replicas of living and deceased performers, but those agreements sit alongside, rather than inside, wider copyright and estate law. In Europe, the AI Act’s deepfake disclosure obligations signal that lawmakers are starting from the audience side of the equation: making sure the public knows when what it is seeing has been synthetically altered.
Creative opposition: preserving history versus rewriting it
Not all filmmakers and film scholars accept the premise that technical capability should be used to reverse historical decisions, even ones widely viewed as damaging.
Daniel Roher, the Oscar-winning filmmaker who directed The AI Doc: Or How I Became an Apocaloptimist, criticized restorations made without a late filmmaker’s consent, citing Oz and Ambersons as examples.
“To be like, ‘I know that the artist doesn’t have any agency over this, but I’m just going to do it, I want to do it,’ is just a dystopic, selfish, postmodern dumpster fire of a use case for the technology, in my opinion,” Roher said. “Do we have to be fucking with everything that was made in the past? Can we just let things exist? … It is kind of inconceivable that you’d go to the Sistine Chapel and they’d be like, ‘Hey, yeah, we decided we wanted to change some elements to the ceiling – we zhuzhed it up a little bit.’”
Acland, meanwhile, argued that the compromised studio version of Ambersons carries historical meaning precisely because it reflects the period’s studio power and the specific struggle between Welles and RKO.
“The fact that the studio back in 1942 insisted on re-edits and a happy ending and taking the film away from Orson – well, that’s part of Orson Welles’ story, and makes the film interesting to watch for all of its flawed components,” Acland said. “The idea we’re going to go back and fix that is a historical absurdity. … What you’re going to get is a shell of that technological component on top of it.”
Rose offered a different rationale: the first widely recognized, milestone uses of AI in moving-image culture could easily be exploitative, and he sees value in attempting a use case intended to “give something back.”
“When I think about the first milestone use of AI in this medium, it could be really without redeeming value – propaganda to foment disunity and confusion, something that exploits a person’s likeness, something pornographic,” Rose said. “Edward and I are trying to use AI to give something back. That’s also a motivating factor: This could be a really beautiful, redeeming way to employ this technology, which is still being worked out and can be scary and leave a lot of people with uncertainty.”
The business question beneath the aesthetics
The industry impact of AI-driven restoration will be shaped less by any single project than by repeatability: whether the approach can be standardized into a cost-effective pipeline and whether rights holders, estates, and audiences accept the premise that “integrity” can include newly generated material.
The Sphere’s Oz has already demonstrated demand for an immersive, eventized version of a canonical film. The Ambersons effort is different: it is not presented as a venue attraction but as a reconstruction of missing work, built around a documentary trail and a stated aspiration toward Welles’ intentions.
In both cases, the immediate constraints are concrete. Venue presentations require technical adaptation to nontraditional formats. Reconstruction efforts require source materials, a defensible blueprint, and some form of stakeholder engagement-even when that engagement arrives late.
For now, the Sphere’s AI-modified The Wizard of Oz presentation continues as a ticketed live-and-edited show built to fill the venue’s interior display plane, while Saatchi and Rose’s The Magnificent Ambersons reconstruction remains in process without Warner Bros.’ participation, with two additional filming sessions planned and a stated hope that the final filming portion will be completed within “a year or so.”
What happens next will be decided less in any one cutting room-or on any one Las Vegas screen-than in the frameworks that emerge around them: the guild contracts, estate policies, platform standards, and regulatory rules that will determine when an AI-enhanced classic is treated as a restoration, when it is treated as a derivative work, and when audiences are entitled to know that the past they are seeing has, quite literally, been rewritten.