By RAY KELLY
From the prestigious Venice Film Festival two week ago came word that the butchered The Magnificent Ambersons would be reshaped with the use of artificial intelligence into something more closely resembling Orson Welles’ original vision.
Welles lost control over the final shaping of The Magnificent Ambersons in February 1942. With editing underway on the movie, RKO Pictures ordered him to Brazil to direct the ill-fated It’s All True. In Welles’ absence, RKO drastically cut the movie down to 88 minutes, tacked on a happy ending and destroyed the trims.
Fable Studio’s Showrunner, backed with an investment from Amazon, seeks to demonstrate what AI can do for filmmaking. It has enlisted the help of Ambersons expert Brian Rose, who completed a partially animated version last year. Fable Studio has not licensed the rights from Warner Bros. Discovery, which owns the RKO release, and has no definitive plans for a commercial release.
Still, the use of AI is disturbing for some film lovers and Welles enthusiasts.
Fable Studio CEO Edward Saatchi and Rose generously agreed to discuss the controversy and their hopes for the project with Wellesnet. We talked informally over Zoom before conducting a structured interview by email.
Why did you choose The Magnificent Ambersons to demonstrate Fable Studio’s ability to complete or rework a finished film using AI?
SAATCHI: The lost 43 minutes of The Magnificent Ambersons is of course the lost holy grail of cinema, and there have always been some people who felt that one day, perhaps the technology would be there to help get closer to the original vision. Even in its compromised form, The Magnificent Ambersons is routinely included in Top 100 lists, including as the 7th best film in the 1982 Sight and Sound Poll and in the Top 100 in the 2012 poll.
In its true reconstructed form (meaning if the footage is found) perhaps we will all say, in the words of an audience card at the disastrous preview in Pomona that ended the film’s chances: “I think that this is the best picture that I have ever seen.” So much was removed that as an audience viewing the 88 minute cut you don’t simply lose a sense of the ending, but of the whole structure and meaning of the film.
This reconstruction can only ever be a ‘fake’ – but it is still worthwhile to give a clearer sense of the shape of the story of one of the great films. That’s the hope – not to equal the footage that may sit in Brazil, but to give everyone a sense of the shape and structure of the film.
In terms of why this is important for AI – it’s not simply a technical showcase, but challenging those who love AI to use AI to contribute to the art form of cinema. Even three years into AI media, the community of artists who are passionate about AI can’t really point to a work of art made with AI that will be able to stand the test of time. The technology is incredible, but it’s often used in service of generic ideas.
Worse, this incredible technology has been embraced by studios as a way to simply cut jobs – to make cheaper VFX shots and cheaper Pixar style movies. The use of AI in cinema has been distorted into simply a way to make the same sequel or VFX-heavy movie as always, but cheaper. Instead of that generic use of AI, using AI to contribute to the history of cinema seemed like the best possible use of the technology.
Ultimately the goal of this painstaking AI reconstruction over the next two years isn’t to show off technology, but to get as close as possible to Welles’ exact vision – as close as possible without finding the destroyed footage.
The cutting continuity of March 12. 1942 details Orson Welles’ cut at 131 minutes and 45 seconds. Using that as a guide, Roger Ryan did a reconstruction with surviving still frames and voice actors in 1993. Brian Rose recently did the same, but using animation to depict the movement and length of scenes. Neither one received the adverse reaction that your plans have. Why do you think that is so?
SAATCHI: From Roger Ryan’s work to Brian’s work there’s a clear line of trying to get closer and closer to the original shape of the film. Myself and Brian sometimes talk about it as a relay race of sincere and earnest lovers of the film trying to get closer to the original vision – and perhaps there will be a version that comes after this reconstruction too.
AI is of course controversial in and of itself, and this reconstruction throws up so many ethical minefields. Even though it is using AI, this reconstruction is being done with the same spirit as all the reconstructions – to get closer to the artist’s intent.
What perhaps is unique here is that the film itself is about the destructive power of technology. Using the new (potentially destructive) technology of AI to reconstruct a film about the unintended and, in some cases devastating, consequences of technology has a resonance for the world of today. A lot of people in the AI industry could recognize the exchange between Eugene, the Major and George on the negative effects of automobiles. Of course the Major is correct when he predicts: “So your devilish machines are going to ruin all your old friends, eh Gene?”
In Silicon Valley we move so quickly forward to embrace new technologies without thinking about the consequences. In its original form, Ambersons may be a film with a powerful warning for the AI moment:
EUGENE (laughs cheerfully) I’m not sure George is wrong about automobiles. With all their speed forward they may be a step backward in civilization. It may be that they won’t add to the beauty of the world, nor to the life of men’s souls. I am not sure. But automobiles have come, and almost all outward things are going to be different because of what they bring. They’re going to alter war, and they’re going to alter peace. I think men’s minds are going to be changed in subtle ways because of automobiles. And it may be that George is right. It may be that ten or twenty years from now, if we can see the inward change in men by that time, I shouldn’t be able to defend the gasoline engine, but would have to agree with him that automobiles “had no business to be invented.”
Anyone working in AI today, that is honest with themselves, would have to recognize in Eugene’s deep ambivalence their own fears about the impacts of AI.
Why have you chosen not to involve the film’s rights holder, Warner Brothers Discovery, in this project? Have they reached out to you following the announcement of your plans?
SAATCHI: Ultimately this is very novel technology – nobody could be expected to give support to a technological approach that is this unknown, even outside of the controversy that swirls around AI today. It’s their IP and their property so making this something commercial is completely up to them.
This project has been touted as both a scholarly endeavor and a means to demonstrate the abilities of AI in filmmaking. Does that mean there are no plans to make the end product commercially available?
SAATCHI: Because the film is owned by Warner Brothers any showing outside of a purely academic noncommercial setting would be up to them.
The use of AI has rankled the Writers Guild of America, Directors Guild of America and particularly Screen Actors Guild (SAG-AFTRA). How would you answer their concerns about AI replacing jobs since this project would not require the hiring of lookalike or soundalike actors to fill in the parts previously done by Joseph Cotten, Agnes Moorehead and Tim Holt?
SAATCHI: This will definitely require actors for voice and physical performance driving the AI, using ‘pose-to-pose’ and ‘voice-to-voice.’ In some ways this is similar to the work done on Alien Romulus and Robert Zemeckis’ Here using AI. There’s no plausible path to a purely AI acting performance or vocal performance with a human performance underlying it.
While this project is noncommercial we will still endeavor to obtain permission from the estates of the actors in the original film – and if there is ultimately a path to make this a commercial project – which would be up to Warner Brothers – making sure that those estates share in the royalties. This obviously is not a huge blockbuster like Star Wars or Alien, with large potential revenue, but making sure the estates are comfortable with this and compensated is important to us.
Purely AI voices and AI performances still lack emotion, even in short clips. The missing ending sums up an entire performance that an audience has watched for two hours, there is no world where AI alone could achieve the necessary emotional beats. This description by Welles of the final scene sets such a high bar of what will need to be emotionally communicated:
“Jo Cotten goes to see Fanny after all those years, [Fanny is] in a cheap boarding house and there’s just nothing left between them at all. Everything is over – her feelings and her world and his world; everything is buried under the parking lots and the cars. That’s what it was all about – the deterioration of personality, the way people diminish with age, and particularly with impecunious old age. The end of the communication between people, as well as the end of an era. Sure it was pretty rough going for an audience – particularly in those days. But without question it was much the best scene in the movie”
Joseph Cotten’s letters to Welles give a sense of the darkness that made people uncomfortable and that perhaps the reconstruction can bring to life by giving a feel of the shape of the film: “The picture on the screen seems to mean something else. It is filled with some deep though vague psychological significance that I think you never meant it to have. Dramatically, it is like a play full of wonderful, strong second acts all coming down on the same curtain line, all proving the same tragic point.”
How confident are you that an AI-generated Tim Holt and Ray Collins can be blended seamlessly with the existing footage of those actors in the trimmed train station scene? Can an AI-generated Jo Cotten and Agnes Moorehead convey the emotion of the lost boarding house finale?
SAATCHI: With actors driving those AI performances, we are confident. There will be many, many techniques used for different problems in different scenes but that final scene… that is the most important thing to get right and for that, the acting that drives the AI models will be make or break.
ROSE: I will answer this question by reflecting upon similar thoughts I had when I started work on the animated reconstruction, more than five years ago. At the time, I had no idea what I was capable of achieving. I only knew what was theoretically possible, based upon the materials that survived, and the tools I had available to me as an animator. The only way I knew how to find out what was possible was by trying.
I think the same can be said of this new endeavor. Having seen what already has been done with AI, I do think it is possible to achieve a seamless restoration, to convey the depth of the performances of Holt and Moorehead and Cotten and Ray Collins – who I believe deserved an Oscar for his performance. But even if the ultimate result falls short, and lands in the dreaded Uncanny Valley, it will be a success by what will be learned about what technology can, and cannot do.
I am reminded of a splendid documentary called Tim’s Vermeer, in which innovator Tim Jenison sought to prove the theory that Johannes Vermeer used optical aids to create his ultra-realistic paintings. His method was to recreate Vermeer’s studio and create a new painting using the supposed techniques Vermeer used. This was a controversial undertaking, as many in the art field took umbrage at the idea that Vermeer, or any master for that matter, used assistive tools to create their works, as if such revelations would undermine their genius, or the authenticity of their works.
Jenison succeeds in creating a marvelous painting, yet, there is a spark missing. Something undefinable. Jenison proves the practicality of his process and seemingly confirms his thesis, yet his greatest achievement is ultimately reaffirming the genius of Vermeer. If this AI project can show what is possible, and affirm the genius of Welles at the same time, the whole endeavor will be a glorious success.
Edward, how did you connect with Brian?
SAATCHI: We connected because we saw his work and then after talking to him and seeing the completed work we felt that this was genuinely going to be possible.
Aside from Brian’s even temperament, deep love for the work and high integrity, Brian’s brain has spent years both consciously and unconsciously absorbing the details of this film; someone could spend a week, a month a year ‘researching’ the film but that’s different to the many years that Brian has devoted consciously and unconsciously to the problems of the reconstruction. It means that we have a kind of ‘moral compass’ around the lost scenes that goes beyond questions of aesthetics and story – he knows the film’s lost scenes as well as anyone and it’s a complete honor to work with him on it.
You only need to talk to Brian for 10 minutes to feel the mix of earnestness and determination that made us excited to work with him. Carringer’s book brought the world a little closer to the original vision for Ambersons; Bogdanovich’s interviews brought the world a little closer, Roger Ryan’s work and Brian’s work brought the world a little closer, and hopefully our work can also bring the world just a little closer to that original artistic vision. That’s the goal.
Brian, I was fortunate to present your animated ending during a talk I gave in 2023 on The Magnificent Ambersons at the Philadelphia Free Library, where it was well-received. Why did you decide to get involved with an AI-generated reconstruction?
ROSE: I feel like Ambersons is a special film, in that its unfinished nature is like an invitation to engage with it, whether it is to ponder what might have been, or to take a more active, participatory role.
Welles himself was the first to posit a reconstruction of his own work. In the 1970s he envisioned reuniting the surviving cast members to film a new ending, which would incorporate their having aged into the story (this incorporation of real time compression into narrative cinema would later be masterfully employed by director Richard Linklater). Sadly it did not come to fruition.
Then, Robert Carringer did the first major scholarly work on the film by presenting the 131 minute version in a detailed written analysis which enabled one to grasp was had been removed from the film. Nearly 40 years later, this book is still the definitive text on the (un)making of the film.
Then, Roger Ryan did the first cinematic reconstruction, creating a two hour version that restored the original plotting and dialogue. I remember watching it in a semester-long seminar course on Welles’ films, and being captivated by its revelations, and struck by what might be possible as technology advanced.
Working with the tools available in 2019, I sought to building upon the existing work, to recreate the entire 131 minute version, not only in terms of dialogue and music, but as well as action and tone, through a detailed reconstruction of the sets in a 3D environment, and hand drawn, rotoscope animation based upon live-action footage created with exacting precision, all based upon the reams of surviving material, including never-before-seen images.
I never thought my film would be the last answer, but the next step in an on-going, iterative process of interaction. That Ambersons could become something special, a living film that each generation might put its stamp upon, like an epic Greek poem receiving a new translation, or a Shakespearean play being recontextualized on stage.
All of this is to say that I had imagined the next step could well be to use advanced computing technology to achieve a photo-realistic reconstruction. That it might even be possible for a seamless restoration that achieves the look and the texture of the surviving material, to offer audiences a cohesive viewing experience.
What I did not imagine was that the technology would advance so quickly that such an endeavor could be possible so soon, when I had just completed work on my animated reconstruction.
AI is understandably a controversial technology, because of the great potential it has for impacting jobs, creativity and the way we perceive reality. My choice to participate in this new, AI-driven endeavor comes from a desire to be an active participant in how the technology may be ideally and ethically utilized. What better way is there for such a powerful tool, than to utilize it to potentially correct a great injustice done to a man and his artistic vision? To allow people to fully appreciate his vision, rather than be left wondering what might have been?
You spent years researching what was trimmed from Orson Welles’ lost, longer cut for your animated project. Obviously that is a major help in tackling this, but what additional research needs to be done?
ROSE: I would love to do an “audit” of the various archives to see what, if any, new materials may have emerged since I began work back in 2019. New discoveries are still being made. Just the other year, a new frame enlargement turned up for one of the missing scenes, which revealed one of my shots was incorrectly composed, and required a complete redo. These are the kinds of discoveries I hope can be made. There are some frame enlargements that were known to have existed into the 1970s, but have since gone missing, which could tell a great deal about some missing scenes, how they were composed and lit. I want to be sure I’ve done everything I can, utilized every possible source of information, to achieve the best-educated approximation of the original.
I also hope this project could be an example of transparency in rebuttal to one of the main criticisms of AI, that it is opaque and lacking in human input and guidance. I want to write a detailed analysis of every recreated shot in the film, describing the surviving materials, and outlining the process and logic behind its reconstruction. It is important that fans of this film understand that there is no guesswork here, no liberties taken or assumptions made, except those informed by the available documentary record. This is art in service of scholarship.
Why is presenting a completed Magnificent Ambersons so important to you?
ROSE: The history of our culture is littered with examples of loss, of disregard for preservation, of short term thinking over long term. Silent films were junked because studios saw no value in them in the era of sound. Television broadcasts were erased so studios could reuse the tape. In 2024, the entire archive of MTV News was wiped from the internet by Paramount. Unless you own a physical copy of your favorite film, or television program, there is no guarantee that it might not disappear from whatever streaming platform it resides on.
The cycle continues, and it makes the story of The Magnificent Ambersons all the more timely and immediate, by reminding viewers of what has been lost, and hopefully awakening in them the realization of what can still be lost.
And all this belies the point that the completed version of the film is a profound work of art. It poses highly relevant questions about the relationship between people and technology. It is not a film that is strictly opposed to innovation, but rather deeply concerned about the role it plays in our lives, and what will be both gained and lost by its adoption. The film is about reflecting upon that trade off, urging us not to lose our ties to our communities and to one another as the velocity of our lives accelerates.
____________
Post your comments on the Wellesnet Message Board.