(Overlapping voices and music)
(Electronic countdown sound)
Gary Hustwit: (Laughs) Hey, everyone. Hi, I'm an independent filmmaker, and for the past 20 years, I've been making documentaries about design, art and music, but they're really about people.
And over the years, I started to question the limitations of documentary filmmaking. Because human beings are multidimensional. There's never just one story about any of us. The documentary film is, by its nature, reductive. Any documentary you've ever seen is just a tiny sliver of the actual story.
But what if a film could tell more than one story? Or what if one film could tell thousands of stories about its subject? How could we reenvision documentaries so that they were as multifaceted as human beings are?
Well that’s what my team and I have been working on. And last year, we released a film called "Eno." It's a documentary about the musician and artist Brian Eno that changes every time it's shown. It's the world's first generative feature film, and there are billions of possible variations of it. It's always a story about Brian Eno, it's just a different story every time you watch it.
So you probably have some questions, like "How does the film change?" "Why does the film change?" "How do I talk to my friend about a movie if they've seen a totally different version of it?" Or "Why are you doing this to us, Gary? We like our movies the way they are." Yes, we all love movies. We've watched hundreds of them in our lifetime. Thousands, probably. And we have our favorites.
But there's one thing that all of these movies have in common. They are all linear, fixed experiences. There's a beginning and an end, and they're the same every time we watch them. But have you ever wondered why? Like why do films have to be the same every time?
The reason is actually a technical constraint from 130 years ago, when cinema was born and film was a physical medium, a reel of celluloid images that had to scroll through cameras and projectors, and they had to make duplicate copies of those reels and send them out to theaters. But 25 or 30 years ago, when filmmaking all went digital, suddenly, this constraint of physicality is gone, but we've continued to make movies kind of in the same way we always have. It's like we're playing by a rule book that doesn't exist anymore.
So in 2019, I reached out to Brendan Dawes, who is a digital artist in England, and we started experimenting. We wanted to see if we could make a cinematic documentary that was created dynamically in software, with real footage, that could tell a different story each time. We built a generative video platform that was entirely human-coded. It wasn't an AI model based or trained on other people's work. And as we were experimenting, we realized, like, who the ideal subject would be for the first generative film. And then, we reached out to this guy.
(Music)
Video: Can you tell us your full name?
Video: Brian Peter George St John the Baptist de la Salle Eno.
(Electronic sound)
GH: We'll just use his shortened form of his name for this talk. So, Brian, for the past 50 years, has been pushing the boundaries of creativity and technology, from his electronic music experiments in Roxy Music to his collaborations with David Bowie on records like "Heroes," to producing Talking Heads, U2, Laurie Anderson, Grace Jones, so many others. And he's released over 40 solo and collaboration records. And in the 1990s, long before generative AI, Brian was making software to create generative music. He likened it to planting the seeds for a piece of music, and then letting that piece of music flower in thousands of different ways over the course of time. It was like the music was a living thing.
Now I’d actually approached Brian several years before this about doing a normal documentary, and he turned me down, and he turned down a lot of filmmakers. But his reason was really fascinating. He said he hated biographical documentaries because it was always one person's version of another person's story, and there was never just one story about anyone. But Brendan and I thought we had a solution to this, and we showed him an early demo of our generative film software, and he was really excited. I still don't think he wanted to have a movie made about him, but he wanted to be part of this generative film experiment, and that was the price that he had to pay. (Laughs)
Brian gave us access to hundreds of hours of archival footage on every obsolete videotape format imaginable. It took us two years just to digitize and catalog all this stuff, but we needed more than just archival footage to tell this story. So I filmed another 50 hours with Brian talking about his creative process. In the end, we had over 500 hours of material for this generative software platform. And I'll show you how that works.
We start off with a data set of edited scenes, raw footage and music, and the system selects pieces from that material and builds them into a film that's probably 85 to 90 minutes long. Now the system knows what all the contents of these pieces are, and it knows how to arrange them into a good story flow. It also creates transitions between the scenes dynamically, in real time.
The biggest challenge for us was how to make it so that every version of the film had an engaging story arc, regardless of what individual pieces were in it. Filmmakers are notoriously control freaks, but I still have control. But it's on this higher level. Like I’m curating all the different little pieces that could go into this system, and I'm also designing the limitless ways that they can interact.
So I don't have control over the contents of each individual film, but it doesn't matter, because it always works. We've designed it that way. And I get to be surprised by my own film every time I watch it, which is crazy and so liberating.
If you have 500 hours of footage you got to get down to a 90-minute film, this is what you're left with. It's the cutting-room floor thing. Normally in a film, you'd have to get rid of all that other footage and get it down to that time thing. "Killing your darlings" is what they say. But in this approach, there is no cutting-room floor. I can put as much material in, and it will come up in different versions of the film in different ways. So it's not like a "choose your own adventure." Well actually, it’s more like the adventure choosing you.
But another thing that we built in that's really cool is that, in any iteration of the film, in every one, either Laurie Anderson or David Byrne will appear and choose one of Brian's "Oblique Strategies" cards. These are, like, random prompts. If you're in a creative bind, you can read them. So there's dozens of them, and depending on which one comes up in the film, it will divert the movie or react in some way. So I'll show you how that works.
[Oblique Strategies]
Laurie Anderson: Retrace your steps.
(Music)
David Byrne: Turn it upside down.
(Music)
LA: Gardening. Not architecture. BE: So this is Cotinus. That's a family of plants that I like a lot. In fact, in this garden, we have a lot of Cotinus, a lot of dogwood.
GH: There are many more cards and many more directions that they can push the film.
The code that we've been seeing on screen is our system parsing the data set and going through all the file names, and deciding what it's going to choose next.
We premiered the film at the Sundance Film Festival, and we've shown it in hundreds of cinemas around the world since then, and every time, it’s been a different version. And each audience that’s seen it was seeing a film that was made for them, that no other audience in the world will ever see. People have come back three, four, 10, 20 times or more to see different versions, and every time, they're getting another layer of Brian's story. Then they talk to their friends and they're, like, comparing versions. "Did you see the David Bowie scene?" "No, I didn't see it." So it's a totally new way to watch movies.
Oh, the film was shortlisted for the Academy Award for Best Documentary.
(Laughter)
I buried the lede.
(Cheers and applause)
(Laughs)
(Cheers and applause)
Oh my God, I buried the lede.
But the question was what was actually being nominated. Like which version of the billions of different versions of the film? And all of them? Like nobody really knew. But these are questions that the film industry in Hollywood will have to answer soon.
Now we’re developing streaming software so we can stream generative films like this. And we're also collaborating with other filmmakers to bring this technology into their films. And there are so many creative possibilities as we scale this idea up. Like what about a Marvel film where it’s different in every theater, and fans can go see multiple versions and piece together the story puzzle. We can also remix existing films with the software. For the past few months, we've been playing with a generative version of David Lynch's "Mulholland Drive," which is crazy, just re-edits itself over and over again. Or it could just play forever and never repeat.
Just to be clear, this approach, I'm not saying it's a replacement for, you know, normal movies, but it's a different path. And I think it's so important for us to keep questioning these legacy models. Just because something is one way for a long time doesn't mean it's the only way or the best way. As filmmakers, we've never had to ask the question of, like, "How would my film change if it could change," because we didn't have the technical capability to even do it. So now that we have that ... the fun part is thinking about all the new storytelling possibilities and all the cinematic languages that this can unlock.
(Electronic sound)
BE: When you create something, you're doing this thing that humans are very good at, which is imagining.
(Music)
We really need to be able to harness the intelligence and creativity of everybody, actually. Art is a way we do that. I think that's a real hope for the future.
LA: Is it finished?
(Laughter)
GH: We'll never be finished reinventing the way we tell stories as human beings, but this talk is finished.
Thank you so much, everyone.
(Cheers and applause)
Thank you.
(Cheers and applause)
