AI and Magnificent Ambersons restoration

[sherbert]

Thanks for the details! I remain cautiously optimistic about this project. It won’t be Welles’ Ambersons, but it could be an interesting “what if."

[JMcBride]

I sent this letter to the editor to The New Yorker today about their article:

To the Editor:

Michael Schulman’s article about Orson Welles’s mutilated film, The Magnificent Ambersons
(“Deepfaking Orson Welles,” February 9), contains unfortunate distortions about why he was fired by R.K.O. in 1942.

Most importantly, the studio lied to Welles and claimed he was fired for going overbudget on his Brazilian documentary, It’s All True. As my research in R.K.O. and U.S. government documents demonstrated in my 2006/2022 book What Ever Happened to Orson Welles? (University Press of Kentucky), Welles was actually under budget by $447,452 at the time of his firing. Studio executives deliberately concealed from Welles that the actual budget was $1.2 million, of which $752,548 was spent by the time Welles was doing the last filming in the fishing village of Fortaleza.

Furthermore, Schulman repeats the old racist, neocolonialist stories about Welles having sex with Brazilian women while claiming that was why he had problems finishing It’s All True. In fact, R.K.O. sabotaged the goodwill project, which was causing consternation both in Hollywood and Brazil for showing poverty and mixing races of performers.

The Pomona preview of Ambersons was indeed disastrous, as Schulman writes, but he leaves out some highly positive responses from attendees, including “I think it was the best picture I have ever seen” and “The direction, acting, photography, and special effects are the best the cinema has yet offered. It is unfortunate that the American public, as represented at this theater, are unable to appreciate fine art.”

Welles was not “mostly ignoring the studio’s panic” over Ambersons, as Schuman writes, but could not leave Brazil due to his obligations to the U.S. government and tried vainly to save the film in phone calls and telegrams to the studio. Schulman contradictshimself and further blames Welles by writing that he “wired back reams of cumbersome changes, which R.K.O. disregarded." Welles's later career, which Schulman dismisses in clichéd terms as trivial, was largely and intentionally independent and included such major films as The Trial, Chimes at Midnight, and The Other Side of the Wind.

Joseph McBride

[Le Chiffre]

Thanks Joe, for posting your letter to the New Yorker. I hope it gets published.

I agree with you that the author makes a lazy appraisal of Welles' later career, and the impression I get from his article overall is that he is ambivalent about Welles, and whether the AI Ambersons project is worth the bother or not.

Most importantly, the studio lied to Welles and claimed he was fired for going overbudget on his Brazilian documentary, It’s All True. As my research in R.K.O. and U.S. government documents demonstrated in my 2006/2022 book What Ever Happened to Orson Welles? (University Press of Kentucky), Welles was actually under budget by $447,452 at the time of his firing. Studio executives deliberately concealed from Welles that the actual budget was $1.2 million, of which $752,548 was spent by the time Welles was doing the last filming in the fishing village of Fortaleza.

I'm amazed that RKO would set a 1.2 million budget for an Omnibus film about Latin America, more than they budgeted for Kane or Ambersons. As I understand it, Rockefeller's office guaranteed 300K against losses suffered at the box office, which means the film would still have had to make at least 900K in order to avoid studio losses if the entire budget had been used. Just out of curiosity, I asked AI about the funding for the film:

The original $1.2 million budget for Orson Welles' unfinished film It's All True was allotted by RKO Radio Pictures.
Key details regarding this budget and its management include:

RKO Leadership: The budget was overseen by the studio's executive regime, which included President George Schaefer—initially a supporter—and Vice President Reginald Armour.

Government Subsidy: As part of the government's Good Neighbor Policy, the U.S. Office of the Coordinator of Inter-American Affairs (headed by Nelson Rockefeller) guaranteed RKO against losses of up to $300,000 for the project.

Budgetary Controversy: Despite a common myth that Welles exceeded his funding, internal RKO communications (such as a "smoking gun" phone call between Reginald Armour and Phil Reisman) revealed that Welles was actually approximately $500,000 under the $1.2 million budget when the studio pulled the plug.

Termination: The project was ultimately canceled after a change in leadership at RKO. New studio head Charles Koerner viewed the project as non-commercial and sought legal grounds to terminate Welles' contract if he exceeded a specific $500,000 spending threshold.

It would be interesting to know how Koerner could do that, but if true, that may be how they were able to fire Welles for "going over" the budget. Welles later said RKO turned the whole project into a tax write-off anyway after Rockefeller cancelled the 300K guarantee.

Furthermore, Schulman repeats the old racist, neocolonialist stories about Welles having sex with Brazilian women while claiming that was why he had problems finishing It’s All True. In fact, R.K.O. sabotaged the goodwill project, which was causing consternation both in Hollywood and Brazil for showing poverty and mixing races of performers.

Welles admits having affairs with Brazilian women in Barbra Leaming's book, and his secretary Shifra Haran jokingly says he had "quickies by the thousands", but how much that delayed the completion of IT'S ALL TRUE is probably impossible to prove. However, those stories, whether true or not, probably did give ammunition to Welles' enemies at RKO who were gunning for him and didn't care if they had to use lies or distortions to get him out.

I seem to recall an RKO memo saying that the footage of mixed races would preclude the film being shown anywhere south of the Mason/Dixon line. I'm not aware of George Schaefer having a specific reaction to any of the Brazilian footage, but in one of his memos to Welles, he describes Welles' 4-page plan to finish the film as disgraceful.

[Wellesnet]

Opinion videos are starting to appear on Youtube.

AI Restores Lost Films? Orson Welles’ Ambersons Debate:
https://www.youtube.com/watch?v=z_osYrn1qqg

AI recreating Magnificent Ambersons' lost footage?
https://www.youtube.com/watch?v=DzEpyb3joHM

Edward Saatchi: maybe not the Orson Welles of AI:
https://www.youtube.com/watch?v=60YBH2XP9R4&t=301s

Jon Oye on the AI TMA (Facebook Reel):
https://www.facebook.com/reel/1222407736663974

[Le Chiffre]

From The New Yorker article:

Attempts to recover the lost “Ambersons” are nearly as old as the film itself. At some point, Welles tried to reconvene the surviving cast members to shoot a new ending, with their characters aged twenty years, but, as he later said, he “couldn’t swing it.” Yet he left behind a substantial fossil record. There’s the “cutting continuity,” a document made by R.K.O. employees for the original film as a guide for editors and projectionists, with descriptions of each line, camera movement, and shot duration. There are publicity photos and frame enlargements—blown-up stills from the film reels, which Welles used while he was sending notes from Brazil—that provide visual clues. And there are Welles’s comments over the years about what he filmed and the effect he intended it to have.

Not to mention an almost complete recording of Bernard Hermann's original score for the film on CD, with Tony Bremner conducting the Australia Philharmonic. Hopefully Mr. Saatchi and his AI team will give proper due to this very important part of the film's sound design. As Orson Welles himself said, "A film is never right until it's right musically."

*

Good to see that Joseph Egan has started following the AI project at his great Ambersons website:
https://www.themagnificentambersons.com/appreciation/

[Roger Ryan]

The majority of the YouTube opinion video commentators seem to be under the impression that the Fable Studio reconstruction will have AI just invent 43 minutes of content since "no one knows what Orson Welles would have done". Given the detailed cutting continuity (identifying precise shot lengths, camera movement, dialogue and sound effects) for the 131 min. version, the numerous frame enlargements and production stills available, as well as, yes, the full Bernard Herrmann score, we actually do know what those 43 minutes of lost footage consisted of. It's true that we can only guess at how the Mercury cast might have performed in the missing footage, how they may have delivered their lines. But instead of viewing the recreation of deleted and trimmed scenes as a desecration of what Welles and his cast achieved, I believe a reconstruction can provide an improved context for the footage that survives, one that actually enhances the existing performances of the Mercury cast and the skill of Welles' story-telling, both of which are compromised in the released version of the film. This was certainly my belief and prime motivation when I undertook my own reconstruction in 1993.

I recently rewatched George Cukor's version of A Star Is Born (1954) incorporating several minutes of reconstructed footage. Fortunately, the original audio track survives for this material so we can hear Judy Garland and James Mason actually deliver their lines, but the visuals are a grab bag of production stills (including newly-created composites placing "cut-out" images of actors superimposed over backgrounds), alternate takes, and what appears to be behind-the-scenes footage. Somewhat surprisingly, this reconstructed version, completed in 1983, is the one that now routinely plays on TCM and is currently featured on the Criterion Channel. But, frankly, it's the version I most enjoy watching since I feel the surviving footage surrounding those reconstructed bits is enhanced and the additional context the reconstructed bits provide improves the story.

It's too early to say that I will feel the same way about the AI-assisted Ambersons reconstruction. A full third* of the film would need to be reconstructed for it to represent the initial 131 minute edit. There's always the chance that the "uncanny valley" aspect of the AI-generated characters will be too distracting to blend seamlessly with the surviving footage (although I think the tests shown so far look pretty good). There have been quite a few examples of AI being used in aesthetically displeasing ways (the Sphere version of The Wizard of Oz, background "enhancement" to Spielberg's The Sugarland Express and to I Love Lucy) that can easily be argued as unnecessary and damaging, but I think using AI as a tool for an Ambersons reconstruction may be a worthwhile and honorable use of the technology.

*Given that the released version of Ambersons is 88 minutes vs. the initial edit clocking in at 131 minutes, it's convenient to say that 43 minutes were cut. In fact, over 9 minutes of the footage in the released version consists of reshoots that often fundamentally changed Welles' intentions. A proper reconstruction would need to reconstruct/recreate the original Welles footage that the reshoots supplanted. Therefore, approximately 52 minutes would need to be newly created (as only about 79 minutes of Welles-approved footage remains in the released version).

[edmoney]

The majority of the YouTube opinion video commentators seem to be under the impression that the Fable Studio reconstruction will have AI just invent 43 minutes of content since "no one knows what Orson Welles would have done". Given the detailed cutting continuity (identifying precise shot lengths, camera movement, dialogue and sound effects) for the 131 min. version, the numerous frame enlargements and production stills available, as well as, yes, the full Bernard Herrmann score, we actually do know what those 43 minutes of lost footage consisted of. It's true that we can only guess at how the Mercury cast might have performed in the missing footage, how they may have delivered their lines. But instead of viewing the recreation of deleted and trimmed scenes as a desecration of what Welles and his cast achieved, I believe a reconstruction can provide an improved context for the footage that survives, one that actually enhances the existing performances of the Mercury cast and the skill of Welles' story-telling, both of which are compromised in the released version of the film. This was certainly my belief and prime motivation when I undertook my own reconstruction in 1993.

I recently rewatched George Cukor's version of A Star Is Born (1954) incorporating several minutes of reconstructed footage. Fortunately, the original audio track survives for this material so we can hear Judy Garland and James Mason actually deliver their lines, but the visuals are a grab bag of production stills (including newly-created composites placing "cut-out" images of actors superimposed over backgrounds), alternate takes, and what appears to be behind-the-scenes footage. Somewhat surprisingly, this reconstructed version, completed in 1983, is the one that now routinely plays on TCM and is currently featured on the Criterion Channel. But, frankly, it's the version I most enjoy watching since I feel the surviving footage surrounding those reconstructed bits is enhanced and the additional context the reconstructed bits provide improves the story.

It's too early to say that I will feel the same way about the AI-assisted Ambersons reconstruction. A full third* of the film would need to be reconstructed for it to represent the initial 131 minute edit. There's always the chance that the "uncanny valley" aspect of the AI-generated characters will be too distracting to blend seamlessly with the surviving footage (although I think the tests shown so far look pretty good). There have been quite a few examples of AI being used in aesthetically displeasing ways (the Sphere version of The Wizard of Oz, background "enhancement" to Spielberg's The Sugarland Express and to I Love Lucy) that can easily be argued as unnecessary and damaging, but I think using AI as a tool for an Ambersons reconstruction may be a worthwhile and honorable use of the technology.

*Given that the released version of Ambersons is 88 minutes vs. the initial edit clocking in at 131 minutes, it's convenient to say that 43 minutes were cut. In fact, over 9 minutes of the footage in the released version consists of reshoots that often fundamentally changed Welles' intentions. A proper reconstruction would need to reconstruct/recreate the original Welles footage that the reshoots supplanted. Therefore, approximately 52 minutes would need to be newly created (as only about 79 minutes of Welles-approved footage remains in the released version).

That's a really good point about how the reconstructed footage both provides context for and also affects perception/interpretation of the surviving release footage. I had that experience when previously watching your reconstruction and also Brian Rose's animated reconstruction.

That reconstruction of A Star Is Born is also the only version of the film available on Blu-ray and the only version of the film I've ever seen. I've been curious to also see the film without the reconstructed bits to compare how the film viewing experience changes, but I don't know if it's readily available anywhere.

And thank you for answering a question that I've long wondered about - how much of the footage from the Welles cut of Ambersons was actually removed for the release version. I commonly see the "43 minutes" figure cited in articles, books, etc., which seems to result from simply subtracting the two run times: 131 min. - 88 min. But as you say, not all of those 88 minutes were directed by Welles. I've thought about using Carringer's book as a guide to identify the reshoots and a stopwatch to time them, but now I have the answer...52 minutes!

One other thing to mention: I know of at least one instance in the release version where the footage is directed by Welles, but a brief bit of dialogue was later overdubbed without his involvement. When Jack goes to visit the Morgans' home, he and Lucy exchange light pleasantries ("...nice of you to have me over at your new house...") in the release version, while the original contained "I wonder, Lucy, if history's going on forever repeating itself...").