AI Can’t Resurrect Orson Welles — But It’s About to Rewrite How We Treat “Lost” Art
There’s something almost painfully human about trying to rebuild a destroyed masterpiece with code. A small AI startup now wants to digitally “restore” the missing 43 minutes of Orson Welles’ The Magnificent Ambersons — the most mythologised lost reel in Hollywood history. On paper, it sounds like pure sacrilege. In practice, it may become a blueprint for how we use AI on the cultural canon: not just to clean and remaster, but to invent what history erased. That’s exactly why this project matters far beyond cinephile circles.
The news in brief
According to TechCrunch, which summarises a recent deep-dive in The New Yorker, San Francisco–based startup Fable is working on an AI-driven reconstruction of the missing footage from Orson Welles’ 1942 film The Magnificent Ambersons. Founder Edward Saatchi, a long-time Welles enthusiast, treats the project as an attempt to approximate the director’s original cut before the studio notoriously removed around 43 minutes and ordered a new, upbeat ending.
Fable isn’t simply upscaling old frames. The team shoots new live-action material, then uses generative AI to replace performers with synthetic versions of the original cast and recreate their voices. They are collaborating with filmmaker Brian Rose, who previously tried a hand-animated reconstruction based on scripts, production stills and Welles’ notes.
TechCrunch reports that the AI output is still rough, with obvious visual artefacts and tone problems, and that Fable is now seeking approval from both Welles’ estate and Warner Bros., the current rightsholder, before releasing anything to the public. Reactions from family members, biographers and actors’ descendants are sharply divided.
Why this matters
The Fable project is a small experiment with outsized symbolic weight. If AI can convincingly fabricate “lost” Welles, it won’t stop there. Every truncated director’s cut, half-finished album, abandoned game and censored TV episode becomes a candidate for synthetic resurrection.
Who stands to gain? Startups like Fable get a high-profile demo for their AI “performance” stack. Studios gain a new way to monetise dusty IP without commissioning expensive shoots. Archives and museums may see a seductive tool for drawing audiences to heritage collections. And some fans will genuinely appreciate any extra window, however speculative, into Welles’ intent.
The losers are harder to see at first glance. Actors’ estates risk their likeness becoming raw material for experiments they never consented to. Audiences risk confusion between historical record and retroactive fiction. Most importantly, the notion of artistic finality erodes: a film is no longer what its maker finished (or what history allowed to survive), but whatever can be probabilistically approximated decades later.
This also blurs the line between restoration and revisionism. Traditional restoration tries to stabilise what’s there: repair colour, sound, missing frames. Fable’s work is closer to alternate-history fan fiction, only with more budget and a sheen of technical authority. Once AI-generated reconstructions circulate, they will inevitably be screened, taught, remixed and misremembered as “what Welles really wanted,” no matter how carefully the creators hedge.
In that sense, Ambersons isn’t just a passion project — it’s a test case for how far we are willing to let generative AI rewrite the cultural past.
The bigger picture
Fable’s experiment sits at the intersection of several ongoing trends in media and AI.
First, Hollywood has quietly normalised AI-assisted manipulation of performances: de-ageing actors, stitching together takes, rebuilding faces from stunt doubles, even cloning voices for ADR. The 2023 actors’ strike in the U.S. centred, in part, on studios wanting broad rights to reuse performers’ likenesses as digital puppets. Fable is essentially taking that pipeline and applying it to the dead.
Second, we already have a long history of controversial “restorations”. From colorising black‑and‑white classics in the 1980s to AI-upscaled anime and 4K revisions of Star Wars, each technical leap has produced its own purist backlash. The difference now is that generative models are not merely enhancing existing artefacts; they are inventing plausible alternatives where no footage survives.
Third, AI is rapidly becoming a standard tool for heritage work. European film archives are testing machine learning for scratch removal, frame interpolation and sound clean-up. Peter Jackson’s WWI documentary They Shall Not Grow Old showed how far such techniques can go while still claiming documentary status. Fable pushes this logic one step further: if we accept algorithmic colour and frame reconstruction, why not algorithmic acting and staging based on scripts and notes?
It also fits neatly into an IP-maximisation logic that dominates today’s media industry. For rights holders, a “new” Welles cut is less about artistic truth and more about another product line: director’s cuts, extended editions, AI reconstructions. Once the pipelines exist, they will be applied to everything from cult classics to mid-tier franchise entries.
The industry direction is clear: in a world of generative tools, nothing ever has to be definitively finished or lost again — at least in the eyes of lawyers and product managers.
The European angle
From a European perspective, the Ambersons project lands in a legal and cultural environment that is, on paper, far more cautious than Hollywood’s.
European copyright traditions, especially in civil law countries, emphasise the author’s “moral rights” — including the right to protect the integrity of a work and to object to distortions that harm the creator’s reputation. Welles was an American, but his legacy is global; if similar AI reconstructions were attempted for, say, Fellini or Bergman, European estates and collecting societies would almost certainly test these moral-rights provisions.
The forthcoming EU AI Act will also matter. Systems that manipulate biometric data, such as facial images and voices, fall under stricter transparency and risk-management obligations. An AI pipeline that fabricates the performance of a deceased actor may not be illegal, but it will need clear documentation and labelling if exploited in the EU market.
European audiences are, on average, more sensitive to authenticity and privacy than their U.S. counterparts. That is visible in debates around deepfakes and synthetic influencers. If and when a reconstructed Ambersons is shown at European festivals or on public broadcasters, expect curators to frame it explicitly as a speculative experiment rather than a definitive restoration.
Finally, Europe has its own rich ecosystem of film archives — from the BFI in London to the Cinémathèque Française, the Deutsche Kinemathek, and smaller institutions in Central and Eastern Europe. Many are digitising fragile nitrate prints from the silent era. For them, AI offers both an exciting toolkit and a potential ethical headache: do they stop at stabilising the surviving material, or follow Fable into the realm of educated guesswork?
Looking ahead
What happens next with The Magnificent Ambersons specifically will likely be slow and messy. Fable still has to solve hard technical problems (avoiding uncanny faces, matching Welles’ famously intricate lighting and camera movement) and, more importantly, secure legal and moral legitimacy from the estate and studio. The most plausible outcome is a limited, clearly framed release: perhaps a festival screening or museum installation that presents the reconstruction as one hypothesis among many.
But the broader pattern is easier to predict. The tools Fable is building for this one film will be reusable. Within a few years, cheaper, more accessible versions will sit in consumer-grade software. At that point, you won’t need a well-funded startup to “restore” a missing scene; a determined archive, or even a fan community, could generate its own versions from scripts and stills.
Expect the conversation to shift from can we do this to how we label and govern it. Clear on-screen disclaimers, technical transparency reports, and separation between archival records and speculative reconstructions will become essential. Professional bodies for archivists and film historians in Europe and beyond will likely issue guidelines, just as photojournalism has developed rules for digital manipulation.
Unanswered questions remain thorny: Should heirs be able to veto any AI use of a deceased relative’s performance? How much training on private rehearsal footage is acceptable? At what point does a speculative reconstruction stop being “the film” and become a new derivative work that needs its own credits and responsibilities?
For creators working today, the implications are unsettling. You can no longer assume that your “final cut” will stay final once models can extrapolate infinite variations from whatever you leave behind.
The bottom line
Fable’s AI-driven Ambersons is fascinating precisely because it fails the impossible promise in its marketing: no, we are not resurrecting Welles’ lost masterpiece, only fabricating a clever, data-driven homage. As long as it is presented that way — as a speculative side branch, not a canonical restoration — it can be valuable and even moving. The real danger lies in the precedent: if we allow AI to quietly overwrite the boundary between history and hypothesis, what other cultural “corrections” will follow?
