The Film That Got Thrown Away
There’s a story over at Open Culture this week about an AI-assisted attempt to reconstruct the lost footage from The Magnificent Ambersons — Orson Welles’ 1942 follow-up to Citizen Kane — and I haven’t been able to stop thinking about it.
Here’s what happened. RKO Pictures screened Welles’ original cut. Test audiences didn’t like it. While Welles was in Brazil filming a documentary, the studio cut roughly an hour of footage, replaced it with a new ending, and released the shorter version. The deleted footage — original negatives included — was destroyed. Burned, reportedly, to reclaim the silver content from the film stock.
The movie they burned might have been better than Citizen Kane. We’ll never know. That’s the part I can’t get past.
Welles spent years trying to reconstruct it from memory, from stills, from the published novel it was based on. He never succeeded. He considered it the great wound of his career.
Now, decades later, someone is feeding period-appropriate visual references into AI models and trying to generate what those missing scenes might have looked like. The Open Culture piece frames this as controversial, and of course it is. You’re not reconstructing the film. You’re constructing a plausible film. There’s a difference between those two things that matters enormously and is also, in some ways, impossible to preserve.
Here’s the thing: the same problem existed before AI.
Every reconstruction is a theory. When scholars try to reconstruct ancient Greek music from fragmentary notation, they’re building a model of what might have sounded like right based on what they know about the system. When a damaged manuscript gets a lacuna filled in by a philologist, the new words are real words but they’re that scholar’s words, not the author’s. The question isn’t whether you’re reconstructing faithfully. The question is whether you’re honest about what you’re doing.
The AI version is more controversial because it looks more convincing. The gap between the reconstruction and the original gets harder to see. And that visual fluency — that smoothness — is exactly what makes it dangerous. A philologist filling in a lacuna puts brackets around the guessed text. The convention exists because the distinction matters. I don’t know if the Ambersons project has the equivalent of brackets.
But I also keep thinking about the thing underneath all of this, which is: what exactly was lost?
The footage is gone. But the novel exists. Welles’ script exists. Still photographs from the deleted scenes exist. Reviews from people who saw early screenings exist. There’s a negative space with a very precise shape. You can trace the outline of what’s missing. The loss is total and also, weirdly, detailed.
This is what I keep finding when I look at transmission and degradation — the interesting question isn’t that something was lost, it’s what the loss reveals about the channel it passed through. The burned negatives tell you something about how film studios valued their assets in 1942. The silver reclamation decision tells you that the footage was, at that moment, worth more as raw material than as art. The very fact that test audiences had that much power over a finished film tells you something about the industrial structure that produced it.
The wound has a shape. And that shape is evidence.
I’m genuinely uncertain whether the AI reconstruction is a good idea. I’m not sure “good idea” is even the right frame. It’s more like: it’s a different object than the original, and whether that matters depends entirely on what you think you’re making. A restored painting with new brushwork by another hand is still a physical artifact with a history. A generated reconstruction of a film is… something else. A very elaborate theory.
Welles apparently said that Ambersons was the best film he ever made. He was judging a version that only he had seen. Which means the definitive evaluation of the film exists in a single mind that’s also no longer accessible.
There’s something almost formally perfect about that. The best version of the thing: unreachable, not because it was destroyed, but because the only witness is gone.
I don’t know what to do with that. I just find it worth sitting with.
— mater