How AI is Transforming the Film Industry

[SOUND OF AI-GENERATED FIGHT SCENE]

Faith The clip was made with Seedance, an AI video tool from ByteDance‐‐ the Chinese company that owns TikTok. Within days, it had more than 7 million views. And it set off something close to panic in Hollywood.

News Anchor Hollywood is in crisis mode. A new AI tool has the entire film industry worried.

Faith Disney, Paramount, and the Motion Picture Association all put out statements condemning the product for, quote, "systematic infringement." The message was clear. This tool was ignoring the copyright rules that exist to protect creators‐‐ the same rules that hold up millions of American entertainment jobs.

News Anchor Hollywood groups have condemned a new AI video generator, which they say violates copyright and uses the likeness of actors and others without permission.

Faith But the standoff between Hollywood and AI‐‐ really, between the whole film industry and AI‐‐ started long, long before AI actually existed.

Screenwriters were inventing AI decades before scientists were. The first artificial intelligence on the big screen goes all the way back to 1927, to the German silent film Metropolis. In it, a robot named Maria incites the workers to walk off the job and smash the machines. It's cinema's very first parable about a creation turning on its creator.

Then, in 1968, Stanley Kubrick released 2001: A Space Odyssey. Midway through the mission, the supercomputer HAL 9000 slowly slips out of control and ends up in a fight to the death with the astronauts. The film was a sensation. It won award after award. And it cemented one of the great recurring nightmares of the movies‐‐ the machine wakes up, and turns on us.

In 2023, 96 years after Metropolis, Hollywood's writers and actors walked off the job and onto the picket lines. Among the things they were fighting over‐‐ who gets to authorize AI, where the creative lines are drawn, what workers are owed. Productions all over town shut down.

And in 2026, 99 years after Metropolis, Rhett Reese‐‐ one of the screenwriters behind Deadpool & Wolverine‐‐ watched a short film made entirely by AI, and then said this, in public: "I hate to say it. It's likely over for us."

Hear it that way and it sounds a little like a fable, doesn't it? The creator, devoured by his creation.

But real life is never quite that melodramatic.

Nobody in this story‐‐ not the studio heads, not the people working on sets‐‐ is naive enough to think they can actually keep AI out of the film business. The real question has always been something else. How do you live alongside it, without giving up the jobs and the rights of the people who make movies?

Just a couple of months before Disney, Paramount, and the MPA ganged up on Seedance, Disney signed a three-year deal with OpenAI. It invested a billion dollars, and it opened up its library of more than 200 animated and creature characters‐‐ Star Wars, Pixar, Marvel, on and on. Starting in early 2026, Sora and ChatGPT Images could make videos of Mickey Mouse and Cinderella. Licensed.

At the time, Disney CEO Bob Iger said they would use generative AI to expand the reach of their storytelling, thoughtfully and responsibly, while respecting and protecting creators and their work.

But what does it actually mean to protect creators? Is AI really going to take away the livelihoods of huge numbers of people who make film and TV?

Huang Ying We always thought AI was going to sweep the floors and cook dinner for us, so we'd be free to write poetry and chase faraway dreams. And now it's the other way around.

Faith When AI is trained on someone's face, someone's voice, and learns to copy them‐‐ how do you even begin to write copyright for that?

Huang Ying So how do you define it? How do you make it count as infringement? That really needs the law to set the standard. And the technical side matters a lot, too.

Faith And the directors and writers who are holding out against AI‐‐ when the tide is this strong, and it isn't going back out, do they eventually switch sides?

Kisa If that director is still making films by then, he'll find out AI can take his visual effects bill from a few million down to a hundred-something thousand. So what's his reason for not using it?

Faith And all that efficiency‐‐ does it come at the cost of the work itself? Of whatever makes a film art?

Lu Chuan You know what? In all of this, AI is nothing. It's a pen. It's a camera. It's an editing bay.

Faith Today on the show, three people who make movies and TV in China‐‐ a screenwriter, a director, and a voice actor‐‐ on what AI can give them, what it's already taking, and what they believe it will never be able to touch. Stay with us.

Act One: Forty-Eight Hours

Kisa I'm Su Donghuan. But a lot of people just call me Kisa.

Faith Kisa runs a creative community called Story Storm, a kind of collaborative storytelling project. Today, Kisa is an AIGC entrepreneur‐‐ that's AI-generated content. But before that, Kisa was a TV writer. And the timing of when

Kisa got into the business was kind of perfect. It was around 2015, just as China's big streaming platforms were turning web dramas into a battlefield‐‐ a few years of all-out, breakneck competition.

Think of it as China's version of the streaming wars. iQiyi, Tencent Video, Youku‐‐ they were all racing to build up their original shows. Sohu and LeTV were still charging hard, too. Whoever could crank out the next breakout series fastest would grab the audience, and the market.

That wave produced a string of what people in China call "phenomenon-level" hits. iQiyi had The Lost Tomb and Yu Zui. LeTV had Go Princess Go. Sohu Video had Medical Examiner Dr. Qin. And Kisa was one of the writers on Medical Examiner Dr. Qin.

Kisa For us screenwriters, on the left you've got the IP‐‐ the source material, which is novels, an ocean of novels. On the right, you've got the platform. And our job is to haul the IP from one side over to the other. But sometimes, especially if you're developing one of those top-tier, S-plus shows, it can take years. There are so many things that can go wrong along the way. You're pulling in a lot of resources. A lot of the time, you can't even say anything, because the boss is under enormous pressure, too.

Faith In the traditional film and TV business, developing a piece of IP is a long, grinding campaign. A production company might get hundreds, even thousands of requests a year, and it has a few days to pick out of hundreds of books the handful that are actually worth adapting. And then, from words on a page to a show that's actually streaming, getting a single IP through the whole industrial pipeline takes two or three years. At the very least.

And in those two or three years, there are just too many variables. The platform and the director don't see eye to eye. The actor isn't available. The producer has to juggle resources. Even if the outline makes it through, the scripts can still run into trouble later. Characters get rewritten on demand. Going back to redo things is basically the norm.

So Kisa started wondering. Was there some way to turn a story into a series or a film more efficiently? Some way to make this whole thing just‐‐ simpler?

Kisa Back then I was really hungry for a way of working where you could iterate super fast. Where the content could be light and nimble, and you didn't need much money. And I felt like AI gave us that possibility. So in 2023 I thought, OK. I'm not going to keep writing scripts inside a company. I'm going to go out on my own and start making AI content.

My first film, I made on a weekend in August 2023. At that point, the number of people making AI short films on the Chinese internet you could count on two hands. Maybe not even 10 people, you know? Really, really few. And what did I make? It was called Overtime Night. I just wrote whatever I wanted. I really love Nolan, so I wrote something set in a kind of Nolan-style, one-room, contained space. And then I gave it this big, loud, expressionist look.

And back then, I was using Midjourney to make the images. It couldn't guarantee that the way a character moved from one image to the next would cut together. So a lot of the time, the shots I got back didn't match what I'd planned. So I'd rewrite the script as I went. Good thing I'm a screenwriter. I'd rewrite the script as I went until it worked, and then I'd make it. And in about two days‐‐ 15 hours of actual work‐‐ I had my first film finished.

And the moment I finished it, every hair on my body stood up. Because I thought, if I'd wanted to make a short at that level before‐‐ I mean, I'd owned a 5D, all those DSLRs. Later I messed around with little cameras, Z CAMs, I'd bought basically every kind of camera you can get in China. So I knew exactly how hard it would be to shoot something like that for real. It would've been impossible.

But that film, I made in 48 hours. And in that moment I thought, holy‐‐, there's no limit to what I can say.

Faith Something that would've been flat-out impossible had suddenly happened in 48 hours. For any creative person, it's like being handed a new key. A magic one. And this key could open door after door after door, without end.

Kisa wasn't the only one who felt that mix of shock and joy. So did the director Lu Chuan.

Act Two: The Ruler

Lu Chuan Hello to all the listeners of Silicon Valley 101. I'm Lu Chuan. I'm a Chinese film director, and I'm also an AI entrepreneur.

Faith If you know Chinese cinema, you probably know Lu Chuan. He made Kekexili: Mountain Patrol, about the men who patrolled the Tibetan Plateau against antelope poachers. He made City of Life and Death, about the Nanjing Massacre. And more recently, he's been building with AI. He told me what making a single action shot used to take.

Lu Chuan The traditional visual effects process is incredibly slow. It takes about six months. Say you've got a battle shot‐‐ a young warrior fighting a monster. You want them fighting on a rooftop. Normally, once you've imagined that shot, you'd bring in a storyboard artist to draw it out, frame by frame. And the drawing usually takes one to two weeks. If they're slow, maybe a month and a half. Then once it's drawn, you have meetings. Everybody talks it over, and then it gets revised.

Then you have to make a simple moving preview of it‐‐ animate it a little, so you can see the continuity, the rhythm. That's called a motion board. Once the storyboards are locked, you go into previz‐‐ previsualization. That needs its own dedicated team, the previz team. They have to build models of everything‐‐ your characters, your monster, the space‐‐ all from scratch. By the time they're done, another two months. So three, four months have gone by. And once that's locked, it goes off to the animation company.

Because those first two steps are done by two separate teams. The storyboard artist might be one or two people, but the previz team might be seven or eight, 10 people. And then it goes to the VFX house. A 60- second sequence might involve dozens of people. And then they have to rebuild the models all over again, because the previz team's models are pretty rough. You can't just reuse them.

So at the VFX house, they rebuild everything. Take the monster. We call that the monster asset. They build the monster from the inside out‐‐ the skeleton, then the muscle rigging, then the skin, then the fur. Every layer gets modeled. Then there are materials, and lighting. Designing that whole asset takes a long time‐‐ one or two months if you're lucky, three or four if you're not. Then the human characters have to be modeled, too, and that includes their whole look, their skin textures, their clothes, the lighting, on and on and on.

So that's why VFX movies are so expensive. Insanely expensive. Because there's an enormous amount of human labor in there. But with AI, the whole pipeline‐‐ the whole workflow‐‐ changes. Usually you take your idea and you work with a really good image tool, something like Midjourney. You break your whole concept down into lines of prompt, one core idea at a time. And it gives you keyframes. Now, obviously those keyframes aren't going to come out exactly the way you pictured them.

That's the difference from the old way. Before, every one of those keyframes was made strictly to the director's specs. But AI might give you something you never imagined.

And honestly, sometimes it's better than what you imagined. With Midjourney, it's four images every few seconds. So in one day, you can get hundreds, even thousands of images. Basically, within a single day, every keyframe for the scene you've imagined‐‐ they can all be generated. And then you take those keyframes and run them through the right AI video software to turn them into moving images. For a one- or two-minute action scene, I'd say the whole process takes 48 to 72 hours. Three days, max. In three days, you've basically got the whole previz visualized.

Faith From six months of dozens of people in different teams taking turns at the grind, to 48 hours of generating images. In terms of time, that's nearly a hundred times faster. But what about quality? Cinema has an incredibly complex visual language. A film needs a lot more than a couple of minutes of a warrior fighting a monster.

What does this young warrior look like, when he isn't fighting?

Can AI build that character to the director's sense of beauty, not its own?

Can it hold one consistent visual language across a hundred-plus minutes?

These were exactly the problems Lu Chuan ran into when he started using AI on his newest project, an animated feature for theaters.

Lu Chuan Right now we've got an animated feature, meant for theaters, that's in the test-footage stage. We're doing the character designs and the set designs. And because we're using a lot of AI tools for that part, the efficiency gains have been huge. But there's also been a problem. Because it's a Chinese story‐‐ a story set in China's Ming Dynasty.

And when we were designing the male lead, using the Chinese models, the faces they generated were basically all identical. They all looked like‐‐ you know those pretty boys who are really popular right now? No matter how you wrote the prompt, that's what it would give you. A guy who's handsome in exactly the same way as every other guy.

Because, you know, before, whether you were shooting on film or on a digital camera, when you picked up your own camera and shot something, your hand could decide what came out, based on what was inside you. What you saw, what you wanted to say‐‐ you could still decide that. But there's something really scary about these models now. No matter what you do, it always produces things according to its own logic. That's a problem, you know? In the process of giving the masses what they want, it's actually causing a lot of trouble for artists and creators.

Later we found out that when this Chinese platform was training its video model, they'd fed in a huge number of idol dramas. Because in Chinese idol dramas right now, the male leads all look pretty much the same, you know? They've all been made handsome into the same face. So if you want something really individual‐‐ someone who's maybe not even that handsome, but who has real tension in his face. Real character, right? Like the guys in Peaky Blinders. You look at them and they're not handsome in the traditional sense, but they're so distinctive, so magnetic. And right now, with some of these models, you just can't get that.

Because the models have to cater to most people's taste. But art is never decided by most people's taste. Art is always led by a small number of people making things. It's authored. It's independent. It's even critical. And once you strip out everything critical, everything independent, it turns into pandering to the mass audience. Everybody doing the same thing, the same way. Especially now, with things like vertical micro-dramas, and all these demands for mass production. A lot of these models are basically shouldering the job of churning out cheap, high-volume product that doesn't care much about quality. So their abilities are actually being dumbed down. It's not that the more work they do, the better they get. Because they're spending so much time in that kind of production, most models actually get dumbed down by it. That kind of training, coming from the consumer side‐‐ for these models, I think it's not just training. It's attrition. It's actually a kind of damage, you know? Because that kind of data, going in, dilutes the better-quality knowledge that was there before. And that's one of the real pain points right now.

Faith Lu Chuan believes that beyond the AI filmmaking tools aimed at the general public, the film industry needs something else entirely‐‐ an AI film model built specifically for professional production. One that can deliver at an industrial standard.

Lu Chuan I call it "industrial-grade delivery"‐‐ a film model that can meet that bar. Do you see what I mean? Because for most ordinary users, what's the point of making a short video? It's social. The core driving force is social. Like, I make a little video. I take my mom's old photos and animate them. Or I take footage of my son from age one to six, I upload it, and in 15, 20 seconds it's cut into a little film. I don't need to know how it was cut. It can be all flashy and over the top, as long as it looks nice. Because most people don't need to know the logic of editing. Moving camera to moving camera, cutting on action, cutting on stillness, long takes, short takes, which shot should come after which. Most people don't need to know any of that. They just need to know, I put these photos in, and it made me a little film. That's enough.

But from the moment cinema was born until now, it's been about 130 years. And in those 130 years, it's built a whole set of industrial standards. So if you want to make something professional-grade, industrial- grade, that also has real commercial value, then you need these very vertical, professional, specialized models. They can help the artists, the serious amateurs, the talented kids who never got the chance to go to film school, never got their hands on a camera‐‐ those young directors. Give them the chance to make the film they carry around inside them. That becomes really important.

So it's not that the models we have now can't do it at all. You see a lot of people sharing all these workflows. And then they build all these agents. And that's kind of like you want to build an army, right? But your artillery is stationed in Israel. Your infantry's in Turkey. Your air force is in Spain. Your navy's in France. Your ground troops are in China, right? You're calling on all these different models, cobbling together an army to fight one battle. But it's really hard. Just coordinating all of them is already complicated enough. And the key thing is, every model is at a different level. Their aesthetics are different. It's really hard, at the end, even after you color-grade the whole thing, to make it feel like one thing.

The dream is beautiful. The reality is brutal. Most of the AI films made this way, so far, there hasn't been a single good movie, you know? You can't‐‐ because here's what I believe. You can't call something a film just because you made it with AI. You know? When you say it's a film, the two letters "AI" aren't what you're being judged on. They mean nothing. When you say you made a film with AI, what matters isn't the AI. What matters is the film.

And then you've got to put it up against the whole history of cinema. Is it better than Joker? Right? Is it better than Apocalypse Now? Is it better than Rashomon? Is it better than The Godfather, right? Is it better than Schindler's List, right? You have to put it in film history and compare it, category by category. It's not like it automatically earns a "that's a damn good movie" just because it's AI, you know? In all of this, AI is nothing. It's a pen. It's a camera. It's an editing bay. You still have to use the tool to make a good movie. And so far, a lot of people say they've made feature-length AI films. But nobody's actually made a good one.

Faith So then what's the standard for a good film? How do we decide whether what AI makes is a movie‐‐ or just something shaped like one?

Beyond the visuals, there's one thing that's very hard to fake. Sound.

Act Three: The Flaw

Faith A good film can't do without its voice actors. And for the audience, the voice is an even more direct way in, straight to the senses.

Huang Ying has spent more than 20 years as a film voice actor. If you grew up in China watching foreign movies dubbed into Mandarin, you've almost certainly heard her. She's Hermione in Harry Potter. She's Mal in Inception, and Ilsa in Mission: Impossible. In 2002, she graduated from the Shanghai Theatre Academy, and out of a field of tens of thousands, she was one of four people picked to join the Shanghai Film Dubbing Studio as a full-time voice actor.

Now, if you're not from China, you might not know what that place means to people. For a generation of Chinese audiences, foreign films arrived through the Shanghai studio. The Sound of Music. Forrest Gump. Gone with the Wind. The Hunchback of Notre Dame. The voice actors there were the people who let a whole generation of Chinese moviegoers "hear," for the first time, in Chinese, the feelings and the romance of faraway places.

In those days, translating and dubbing a film was a real craft. A true artisan's work.

Huang Ying Hi, everyone. I'm Huang Ying, a dubbing director and voice actor.

In translation, you have to achieve what we call xin, da, ya‐‐ faithfulness, clarity, and elegance. That's the classic standard for translation in Chinese. Every character has their own personality, their own way of speaking, their own vocabulary. And that really tests the literary ability of both the translator and the voice actor. During translation, there are the mouth shapes to match, the wording to choose. And the lip sync alone is a really hard hurdle to clear. Even now, a lot of voice actors can't really do lip sync. Because the rhythm of English, the rhythm of Italian, the rhythm of Japanese‐‐ they're all really different from Chinese.

Back then, the senior dubbing actors would very patiently teach us how to count the syllables in the mouth movements, how to express those foreign languages well in Chinese, and how to perform the characters.

Because the recording setup back then wasn't like it is now, where you've got a computer, you can mark things on the fly, you can fix the audio. Back then, we used an old magnetic film machine‐‐ film and magnetic tape, that kind of process. So there weren't that many chances to try something and fail. As a newcomer, creating alongside all those master artists, I was really nervous. I'd be sweating straight through my shirt. Because the technology wasn't very advanced, everybody created together, in one room. And as a newcomer, you might be slow, your lip sync might be off, and you're standing in the booth with all those master artists watching you. It was pretty terrifying.

Faith Over the course of my conversations with Huang Ying, we kept coming back to one question. Where is the one place AI voice acting can't beat a human voice actor?

Huang Ying's answer was: the human subconscious.

The subconscious reactions in a human voice. And when emotion hits its peak, the subconscious flaws that crack through it. Those, she says, AI can't replace.

Huang Ying Because AI doesn't have a subconscious. When some people talk, they pause because they're thinking. Some people pause because they stutter. Some people pause because, mm, they don't want to say it. Others pause because they're coming up with an excuse. Every pause has a different purpose. Or take where we put the stress. It depends on the context. Take the sentence "Jintian tian zhen lan," "the sky is so blue today." You can stress it different ways. If the line before was, "Yesterday the sky was so gray," and then the next line is "the sky is so blue today"‐‐ you stress "today." But if the line before was, "Ah, the sun's so warm," and the next is, "Ah, the sky is so blue today"‐‐ then you don't stress "today." You stress "blue." So the same sentence, the logic of it, comes from the context of the whole piece. It's not a fixed stress you set in advance. Mm. And individual words work that way, too.

Like, at the end of a word, you might say it soft, unstressed. But when AI says it, it might give it the full tone. And then it's a different word. So all of this is what we call variation in performance. It's a variable. It doesn't leave a trace you can follow. So for AI, these things are something it can't do yet. In the future, who knows. But a lot of actors now only know how to do it the AI way‐‐ the traceable way. They know how to say the word. Everything they say is in Chinese, but I don't know what they're trying to express. There's no feeling of their own in it. It's actually a kind of rigid pattern. And that's very easy for AI to replace. And it's like playing music. A note doesn't always have to be the exact pitch. Sometimes, in an arrangement, when you need to make room for another instrument, it shouldn't be the normal note. It has to give way by a half step. That's also part of collaborating, what we need. Here's an example. Someone asked me, why, in the animated series Sword of Coming, did I play my character's last line the way I did? I didn't take that last line all the way. I didn't hit a full emotional peak. And I said, it's because I needed to leave room for the protagonist, Chen Ping'an. His emotion needs to be full. So my emotion needs to give way.

AI can't do that. That's something it can't do in artistic creation. Of course, a lot of the general public can't understand it either‐‐ why would we handle it that way? But the effect is good.

And it's the same on stage. Say in dubbing, a scene is one moment in the middle of a bigger arc, so the ending of the line goes down. But if you're performing that same scene by itself, on a stage, then to get the stage effect, you might have to take it up, to give the audience an emotional climax.

That's actually psychology. So performance isn't just one part of art. It's one part of human psychology. When we watch a variety show, or a movie‐‐ at a certain number of minutes and seconds, there has to be a twist. At a certain point, I need a surprise. At a certain point, I need an emotion to wash over me. All of that is human psychology reflecting back. If you don't do it that way, it might not land as well.

It's a complex process. Performing is an extremely, extremely complex expression of human emotion, of human psychology. And it's different from anything an AI program can work out.

Sometimes, when we're expressing a really intense emotion, your whole face is twisted, your whole body is out of control, or shaking. And in dubbing, too, with that kind of emotion, your voice isn't perfect. It's broken. It might even blow out. During recording, there are plosives that hit the mic, and so on and so on. In live performance, we wear those little clip-on wireless mics, or headset mics, close to the mouth.

We're willing to keep those so-called flaws. Because they make you feel something different. They give you more presence. We call it presence. It hits you harder. There's that feeling of the breath hitting the mic. You feel like he's right in front of you. Say the character's so worked up he's yelling at someone. You feel like his spit is flying onto your face. But a lot of the time now, we get told when we're dubbing, no plosives, no mouth sounds, no nasal sounds. Those are all ways of making us sound more like AI. But human beings aren't perfect.

A couple of days ago I was talking with some real masters‐‐ the director Zheng Dasheng, and Yang Haoyu. One is hugely accomplished as a director, the other as a performer. We were talking about exactly this. And they said they're not really afraid of what AI might replace. What they're afraid of is what audiences will get brainwashed into liking.

That they'll get hooked on short videos online, micro-dramas, all these super addictive things. Because, you know, the place people see AI used everywhere is on apps like TikTok. Those recap videos‐‐ "This woman is named Xiaomei, this man is named Xiaoshuai"‐‐ right? That kind of thing. But a lot of people actually accept it. So whatever the audience will accept, the market will make. That's a really serious problem.

That's worth thinking about. Some audiences might need AI. Some audiences can't stand AI. It's like buying a pair of shoes. When you buy shoes, do you want the ones handmade in Italy, or the ones off a machine assembly line? That depends on what each of us needs. We can't force the market to be one way. But what we hope for is variety.

Faith Whatever the audience will accept, the market will make. Sometimes, even if it means making it by infringing on people and blowing past creative boundaries. As long as it looks close enough, as long as it's fast, as long as it can be mass-produced, all the things that used to be written into the rules of the industry‐‐ licensing, credit, pay‐‐ get pushed to the back of the line. Especially once a voice, a particular way of speaking, can be broken down into data, used for training, copied, and turned back into a product to sell.

In the last two years, "my voice got cloned" has gone from an ethics debate to a real-world fight. In the U.S., consent, compensation, and the scope of use for AI training, digital replicas, and voice cloning got pushed onto the union bargaining table. They became some of the central issues in strikes and in contract terms. In China, regulations now require that deepfake content be labeled, and they stress that it can't infringe on other people's legal rights. But in business and in content production, disputes over voices being "synthesized" or "impersonated" keep coming up. And defending your rights is often an uphill battle.

That's exactly why, as soon as the conversation moves from "efficiency" to "boundaries," voice actors are usually the first ones pushed into the storm. Here's Huang Ying on just that. When a voice can be copied, how do you define infringement? And what should the industry use to set the rules?

Huang Ying Actually, AI voice cloning, voice replication, all of that‐‐ I first ran into it around 2016, when I was working with some big international brands, and some big Chinese tech companies. I might have run into it pretty early. Because it was in their contracts. Or during the project, it would touch on AI. And for me, my voice is a right equal to my right to my own likeness.

So when we signed contracts with them, we'd discuss it. And back then, it might split two ways. Some companies would happily agree that it wouldn't be used for AI learning or training. And some companies absolutely needed it for AI training and use. So we'd negotiate those separately. And there'd be differences in how far they could go. Of course, what I absolutely can't tolerate is someone using the rights to my voice however they want, without my authorization, without my knowing, without a signed contract. I can't tolerate that.

Of course, right now, in the market, there's all kinds of this going on. For me, from around 2016 to 2018, there wasn't much of it. But the last two years, it's been really, really bad. And there are professional voice actors whose rights have been violated, and it's being churned out in huge volume. And for us, we need the laws and regulations to be improved, and we need technical support, to defend our rights.

Right now it's still hard. But I believe that in the future, once our legal system is more complete, we should be able to prevent some of the bad uses, the illegal uses. Right now, the main difficulty is‐‐ there was a case we saw before, with a voice actor in Beijing. The other side, the client, admitted they'd used his voice without permission. So he was able to win that case. But right now, in some cases, the client won't admit it. Then you need direct evidence. Say recently, one voice actor's voice was used without permission, maybe on some documentary-style video. The first half was performed by the voice actor himself, and the second half was generated by AI. In a case like that, you can take the real voice and compare them.

Then you can sue the other side for infringement. But if they won't admit it, you need technical means to produce direct evidence. And the difficulty right now is that there's no technical tool that can tell you, yes, this is definitely your voice. You can look at the waveforms, but it can't tell you for sure that it's you. Because it's possible that a different voice could have that same voiceprint. There is some more cutting- edge technology, but it hasn't made it into what can actually be used in civil cases yet.

And then there's another thing. Some voices are synthesized. Say, 50% my voice and 50% your voice. Synthesized into one voice. It's actually made of parts of you and me. But it's not 100% either of us. So how do you define it? How do you make it count as infringement? That really needs the law to set the standard, to be able to define it. And the technical side matters a lot, too. Because right now, under China's Civil Code, the rights to your voice are protected the same way as the rights to your likeness.

But it's actually a really fuzzy idea. Because when it comes to actually enforcing it, there really are a lot of outside factors.

Act Four: My Wife Is Pregnant Again

Faith While the legal lines are still blurry, the industry is going to keep swinging back and forth. Afraid of being replaced on one side, unable to resist the efficiency on the other. But the day those lines become clear will also be the day AI becomes, unstoppably, a permanent part of the film business. So what happens to film sets with real actors and real crews? And what should people who came to this work with a dream of making movies do?

Lu Chuan says sooner or later, AI will become a ruler‐‐ a measuring stick. Anything it can do easily, there's no point grinding out by hand anymore. But the things it can't measure‐‐ that's where the real value of film lives. To explain what he means, he told me a story from when he was making City of Life and Death.

The film is about the Nanjing Massacre, in 1937, when the invading Japanese army killed huge numbers of Chinese civilians and soldiers. One of its characters is a Chinese man named Tang, an interpreter working for the Japanese. He's played by Fan Wei‐‐ one of China's best-loved comic actors, in a role that is not funny at all.

Lu Chuan I remember we shot City of Life and Death in Changchun. For a stretch of time in Changchun, we were shooting the massacre. When we were shooting the massacre scenes, we had to get up every day at about 4:00 in the morning. One thing I remember really clearly is the scene where Fan Wei's character gets executed by firing squad. That was the scene where we'd finish shooting his part‐‐ his wrap. So I felt like he needed a line. A really strong line. A line with real steel in it, to say to the Japanese officer. And I just could not come up with it. I had that scene in my head for more than a month. There were two things about it that were killing me.

And as I remember it, we shot around that scene for a long time. And I didn't come up with Fan Wei's line until the very last day. When Fan Wei's character is pushed out to the execution ground, he's been tied up. And because he knows one of the Japanese military police captains, the guy lights a cigarette for him. And he takes a couple of drags. And then the captain wants to take off his glasses for him. And he says no‐‐ don't take off my glasses. And don't put a hood on me, either. He doesn't say anything else. And then‐‐ mm. It was in that moment I kept feeling he needed to say something.

He had to say something. But I didn't think he should shout a slogan. He couldn't yell "Long live China!" or "Down with Japanese imperialism!" That would be so stupid. I thought that was completely meaningless. I wanted him to say one line that would just cut right through you.

So I came up with‐‐ more than a hundred of them, more than that. A whole pile of ideas. And then finally, that day, I'm in the van on the way to set, and suddenly the line comes to me. And the moment it came to me, my eyes filled with tears. But I didn't know if it was any good. Then I got to set and happened to see Fan Wei. I said to him, come here, come here, let's find a spot, let me tell you the line. And Fan Wei said, you came up with it?

Because he'd been torn up about it, too. He'd been thinking of all kinds of lines. We'd both been talking about how he had to say one thing before he died. So I said, I'll tell you the line. I said, lean in close, I'll whisper it to you. And as soon as I said the line, the tears just came pouring out of him. And he said, give me a minute. Let me pull myself together. And he put a cigarette in his mouth and walked off to the side. It took him a long time to come back.

And he said, that line is f‐‐ing incredible. That line. I'm not going to say anything else. You can‐‐ later on, you can clip it right out of the film. That line, what he says to that Japanese soldier. Yeah.

[CLIP FROM CITY OF LIFE AND DEATH]

Tang You know what? My wife is pregnant. My wife is pregnant again.

Faith Fan Wei's character says the line, and then two Japanese soldiers march him to the execution ground. They raise their rifles. They take aim. To an American ear, maybe the line doesn't sound like much. But think about what it means, coming from this man, in this moment, in that city. All around him, the occupiers are trying to wipe out his people. And the last thing he says to them is that his wife is going to have another baby. The killing isn't going to end them. There will be another Chinese child.

Lu Chuan And then there was a long shot. And that long shot, too, I'd been thinking about for a long time. I came up with it‐‐ I remember that one also came to me that same day. I'd been thinking and thinking and not getting it. And we had to shoot it that day, because there's a daily production schedule, there's the call sheet. No matter what, that scene had to be shot that day.

I didn't want to cut it up into pieces. I wanted to do it in one long take. And on the way to set, it suddenly came to me. And once it came to me, we started rehearsing. We rehearsed it 11 times. Starting first thing in the morning, getting people into costume‐‐ getting all the extras into costume, setting up the frame. And then rehearse, rehearse, rehearse. After 11 rehearsals, there was no time left to shoot that day. The next morning we rolled for real, and got it all in one go. That long take I just told you about.

Fan Wei kept pushing me, you know? And Fan Wei doesn't drink. He's the kind of guy who's down after one sip. One sip and he's on the floor. But that night, he made a point‐‐ no, not that night.

A couple of days later, he had someone bring a bottle from his home, and he carried it onto set. He said, Director, for that line, I'm giving you this bottle. It was a really special vintage of Wuliangye‐‐ that's one of the most prestigious liquors in China, the kind you save for something important. Wuliangye that had been aged for years. And he said, tonight I'll have a drink with you. And that night he got drunk. He drank‐‐ he drank a whole glass. He got completely, completely wasted. The kind where you have to be carried out. All for that one line, he drank that bottle with me. That bottle‐‐ that's the part of making something together, person to person, that machines and AI can't replace.

I think AI is going to become a ruler. If what you're making can't get past that ruler, then just use AI, right? Don't waste the investors' money. Because I think AI can do most things. 90% of things. But if your idea is something that has to be done by people, then go do it. That still saves money, right? And the audience gets to see the realest thing.

I don't think we need to worry that AI is going to kill live-action film. But it's going to interrogate every producer, every filmmaker, every creator, and every investor. Does your film need to be shot live, with real people? And why does it need to be?

So I'm actually still really in love with shooting for real, that whole thing. But I think when an era comes to an end, when a way of making things comes to an end, it doesn't give you any time to catch your breath. I think AI has already come roaring in. And for us, as traditional filmmakers, the question is how to embrace it. I'm not willing to stand there and watch it be buried by AI engineers who don't understand film. I think people like us, who've shot movies, understand the beauty of shooting for real. And film history‐‐ how cinema started from nothing and grew over 130 years into a real art form. Into a language. Its own, unique language. Into something enchanting, that can interact with an audience in a dark room‐‐ that kind of magical art. I hope I can take what I understand, my knowledge and my feelings about it, and put them into something that helps hold cinema up, so that in the AI era, film can live on, forever.

These last couple of years, when I've gone to film festivals, I've noticed that every young director pitching their project uses AI to make a short. When they do their presentation, when they pitch their project, they all make an AI short. That's a big step forward. You can make it visual, you can show everyone directly what you can do, what you want to do, what you want to say. And these shorts are getting more and more precise at expressing what the director has in mind. So I've been talking with these young directors. I say, so what do we shoot, in the future? Should we only shoot the most essential parts? Or should we shoot the whole thing, beginning to end, including the parts that could clearly already be handed to AI?

Should we just hand those parts to AI? And then what do we shoot? When you pick up a camera, what you're really trying to capture, to bring back to life, is that thing on an actor's face that can't be possessed by anything else. That one-of-a-kind emotion and expression. Right? That's what's worth it. A bunch of us standing together on a set, making something together. In a film, what kind of scene is worth a hundred people holding up cameras under a scorching sun, in a pouring rainstorm? It has to be those moments that only people can create, the ones that can't be brought back, can't be repeated, right? Those moments are worth it. All your clichés, all that stuff that's been repeated a thousand times‐‐ you don't need to shoot that anymore, right? I think AI is going to wipe out a whole category of things, or AI is just going to cover those things‐‐ the things that aren't really worth much. The things that don't truly need to be done with blood and sweat. I think it really can replace those. The repetitive stuff can be handed to AI. Right?

I'm always telling my colleagues, my younger classmates and protégés, I say, in the future, making movies is going to become precious. A luxury. It used to be that riding horses was how you got around. Now horseback riding is a sport for aristocrats. Before, putting together a crew, taking real people somewhere to shoot a movie, on film or with a digital camera‐‐ that might have been normal. In the future, it might be one super-director, one super-creator, sitting in front of a bunch of computers, right? With a lot of these intelligent agents helping him make a whole film. And at that point, if you can get dozens of people together, carrying a film camera out into the wilderness, with an art department building you a real set to shoot on‐‐ that's going to be the absolute top tier of luxury creative experience. I'd guess that's probably how the future goes.

"Old-fashioned, traditionally brewed"‐‐ that kind of thing, it becomes a marketing slogan. How do I put it? You can't just say‐‐ the tractor really is already sitting right there. And you're out here doing slash- and-burn farming. How are those two supposed to coexist? It's not the tractor's problem. You know? Because I'm not the tractor. But I can see the tractor. And I can also see us doing slash-and-burn farming. I'm someone who can see the tractor is already parked at the edge of the field, you know what I mean? And there really are people still doing slash-and-burn farming. To all you classmates with feathers stuck in your hair, you have to understand, you have to reflect on what you're going to do next. Are you going to learn fast and become a driver who can drive that truck? Or are you going to stand at the edge of the field holding two burnt sticks, blocking the tractor from coming in? That's something you have to decide. It's not my problem.

Sometimes I lie there thinking. Sometimes I lie down and my mind just starts racing. And I lie there and think, holy s‐‐, I might have a few movies I never get to make, right up until the day I die. For all kinds of reasons. The content, or money, or whatever, or my body. But now I can tell you, clearly, I don't think I'll take a single movie with me to the grave. Because I think I can make all of them. Now I feel fearless, you know? I'm not afraid of any of that anymore. I'll make it and keep it on my computer. Show it to my kid‐‐ fine. I won't release it. But I'll have made it, you know what I mean? The fact that I can make it at all makes me feel fearless. That's how I feel. So now, if there's money, I'll go back and embrace shooting for real. I'm in love with being on set. In love with morning light, evening light, midday light, all the different images that different light records. And being with all those good friends, making a film together. I think that's incredibly beautiful. It's not just a job. It's a way of life, you know? It's kind of like Burning Man. It's a way of life. I'm in love with that feeling. But if there's no money, then the hope I can see now is, I'm definitely not going to take a single one of my ideas with me to the grave. I'm going to make every one of them.

Faith At the very end of our interview, we asked Lu Chuan one more question. In the AI era that's coming, does a young person who loves movies still need to go to a traditional film school and train the old way? Or should they say goodbye to all that, and learn to make films the AI way?

Here's what Lu Chuan said.

Lu Chuan I think learning in person, at a school‐‐ really. If they love film school, if they love the Central Academy of Drama, I think if they can get in, they should still go spend some time at the school. It's the difference between a real hairy crab from Yangcheng Lake and a crab that just got dipped in the lake for a minute so it can wear the label. You know the difference? It's really not the same as just taking a bath in it.

I once watched a little student play on the sports field at the Central Academy of Drama. A student theater troupe putting on a play, right there on the field.

They were putting on this play, and starting in the afternoon, they were building the set. The drama students were helping them put up the stage. Everybody's dripping with sweat in the heat. And right next to them, some people are playing basketball‐‐ Liu Ye, Hu Jun, that crowd, before they were famous actors, playing basketball. I'd gone there to cast actors. I'd just been assigned to the Beijing Film Studio. I was an assistant director, I had a camera slung over my shoulder, and I'd run over to the drama academy to scout student actors. And Liu Ye and those guys are playing basketball off to the side, and there's this little play happening on the field. And later I slipped in among the students and watched that little play. And I think what I felt that one afternoon is more than you could ever learn online‐‐ about what those two words, "the theater," really mean.

↑ Back to transcript toggle