AI Can't Play a Real Person Because It Never Makes Mistakes — the Human Touch Lives in the Flaws
AI performance doesn't lose on image quality, it loses on never making mistakes: the power of a real performance comes from rhythm, flaws and lived experience, and none of that can be dialed in yet.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The director's job is to say the idea clearly
Wu Hankun quotes Quentin Tarantino: when every craft around you is more professional than the director, the director's only job is to tell everyone, in simple clear language, the image in his head. He thinks this perfectly defines the AI creator: the precondition for giving a model instructions is having an image in your own head first — the person stands on the left, does this action, interacts with the other person this way. The prompt tricks in circulation and the prompt marketplaces that sell for money, he judges, will be flattened out fast — just as learning every button on a camera doesn't mean you can shoot a good image. The precondition for writing a prompt is having something worth expressing.
— Wu HankunRhythm is millisecond-scale, and AI can't learn it
Comedy lives and dies on rhythm: a nod, a glance, holding two seconds longer or 0.5 seconds longer, and the story means something different. Text instructions can't tune performance rhythm — tell AI which muscle should move at second 3.9 and it doesn't understand. Wu Hankun's workaround is to perform it himself first, letting AI capture and learn his sense of timing. A one-minute monologue he made these past two days was all done in single takes by him: the glances, the muscles, the beat of the tears were all designed, and AI was only responsible for swapping the background. As an actor, he knows better than most creators what should be left to AI and what should be left to people.
— Wu HankunYour own face isn't a moat, your consciousness is
On AI face-swap acting, Wu Hankun raises a key question: if I license my face to a third-party AI platform, and someone uses prompts and motion control to make my face deliver some performance, blown up on a big screen — can that still count as my performance? His answer is no, ‘the consciousness isn't in it.’ But he isn't a pure artiste: a streamer licensing out their face, going from one stream a day to twenty, reaching more viewers, is a business that adds value for both sides. The real line to think through: when artistic expression is involved, are you willing to hand the right of expression to someone else. As for whether technical capability will actually replace actors, he offers a different layer of judgment: whether the technology can is one thing, whether we let it do so is another.
— Wu HankunThe moment you know it's AI, empathy breaks
He showed a friend an AI-generated animal documentary; the friend was moved, then learned it was AI and immediately felt deceived. Wu Hankun says AI doesn't itch, doesn't hurt, isn't truly sad, and human empathy has a precondition: believing a person stands behind the work. Liu Fei adds an example: in Project Hail Mary, Ryan Gosling acts opposite a rock, and the audience all knows it isn't human, yet still empathizes — because the voice acting, the plot and the music put ‘the human thing’ in there. Once you confirm the other side of the screen isn't a person, the original emotional connection breaks, and all that's left is the feeling of being deceived.
— Wu HankunCut until only the mother and child remain
The script for Population Anomaly was originally about a boy from the security army faction falling in love with a girl from the Earth defense organization; he found it too much like a family drama and changed it to a stress-resistant couple; found it not dramatic enough and added a gravely ill child; finally he deleted the husband entirely, leaving only the mother and child — this child should never have been born and must die, but as a mother she absolutely will not allow the child to die, and she has to carry this child who shouldn't exist out with her. Wu Hankun says a short film can't hold complex drama; at most it can ‘pose the audience one interesting question to linger on.’ This chain of cuts shows: the real bottleneck in AI creation isn't the tool, it's whether you can push people into an extreme situation.
— Wu HankunThose chasing new models lose from the start
Two months ago C-Dance 2.0 launched and the whole internet thought the one-person-one-movie era had arrived; two months later a new model appeared and everyone looked back and found it wasn't that impressive. Wu Hankun says something new pops up every month, and chasing new models is a guaranteed loss: if you chase A, it means giving up B, which is certain to appear, so you've lost from the very start. His way of dissolving technical anxiety is to use the tools — keep making short films, and only after using them do you know where a tool's ceiling is. Most of the know-it-alls online saying ‘learn this now or be eliminated’ are manufacturing anxiety to sell courses.
— Wu HankunThe era of waiting for assignments is over
Liu Fei pulls the topic wider: media used to be a closed system — you got into a TV station as a script supervisor, as a camera operator, and climbed rung by rung; now the system has dissolved, the ecosystem has become self-media, no one hands you assignments anymore, and you're forced to decide for yourself what to do. Wu Hankun answers, ‘exactly right, it's like that in every industry.’ When process-based roles are deconstructed and survival is no longer the problem, the people who can't earn a sense of value are the most anxious. What AI brings is ‘temperament equality’: you used to need to get into a big crew to make a film, now as long as you have something to express, one person might get the work made — the question is no longer which tool to learn, but whether you have a purpose of your own.
— Liu FeiIn their own words · checked verbatim
But somehow, between his 95% completion and 100% completion, in that 5%, some magical transformation happened, and when he hit 100% I suddenly felt, this is actually kind of decent.
但是 somehow在他95%的完成度 到100%的完成度 就5%之间 就发生了某种神奇的转变 然后到他100%的时候 我就突然觉得 这还是有点像样的
Wu Hankun12:05
Sometimes it's a matter of milliseconds — you might nod, or glance, hold it two seconds longer, or 0.5 seconds longer, and the story is different, the feeling is different.
有时候就是毫秒之差 你可能一个点头 一个眼神 多停留了两秒 或者多停留了0.5秒 这个故事就不一样了 这个感觉就不一样了
Wu Hankun19:12
Things that are too regular are what AI is good at. Hand what it's good at to AI, and then the things that maybe belong to people — especially when artistic expression is involved — still hand those to people.
太有规律的事 是AI擅长的 要把擅长的东西交给AI 然后可能属于人的 这些尤其是涉及到艺术表达 还是交给人
Wu Hankun25:15
We watch some of those Animal World videos, say a lioness, she's injured, two cubs guarding beside her, watching warily off to the side — I believe a lot of people would be moved by that image.
咱们去看一些动物世界的 那个视频 就比如说一个母狮子好了 受伤了 两个小狮子守护在这个母狮子旁边 很警惕的朝旁边看 这个画面我相信很多人都会被感动到的
Wu Hankun29:19
It's because later there's voice acting, there's some of that human thing, your human thing is in there, the plot pushes it, the music comes in, and immediately the feeling comes out.
是因为后面他有配音了 有一些人的 你人的那种东西在了 剧情的推动 音乐一起 马上你感觉就出来了
Wu Hankun30:19
This child is not allowed to be born, he must die, but this mother, as a mother, absolutely doesn't want the child to die — so that's the state, one state after another.
这个小孩是不允许出生的 就他必须得死 但是这个母亲作为母亲 他一定不想让小孩死 所以就是这样的一个一个状态
Wu Hankun37:25
Every month a new thing comes out, every month a new thing comes out. We can't keep chasing new things, because if you chase this thing it means you won't chase another thing, but that other thing is certain to appear — so you've lost from the very start.
每个月会出来一个新的东西 每个月会出来个新的东西 我们不能一直去追寻新的东西 因为你追了这个东西 就代表你不会追另一个东西 但是另一个东西一定会出现 所以你本来一开头就输了
Wu Hankun40:26
Figures
| Fengshen training camp audition scale | Over ten thousand interviewed worldwide, roughly twenty to thirty people finally entered the camp | 2:01 |
| Weight on entering camp | About 80 kg on entering camp, 71-72 kg at his lightest during the camp | 3:01 |
| Resumes sent after returning to China | Over a hundred, none successful | 4:01 |
| Shanghai International Film Festival AI set slots | 11 groups and 22 people took part, 4 groups entered the AI unit screening, Wu Hankun's group planned a 5-10 minute finished film | 47:29 |
How to listen
Actors and film/TV workers worried about being replaced by AI, content creators looking for method in AI video work, and professionals anxious that they've missed the AI wave.
The first seven minutes of personal cross-disciplinary backstory can be skipped; the information density picks up from Part 2, on making AI short films.