AI Is Infinite, Life Is Finite | A Conversation with KK and 山音, Frontline AI Creators
Summary
- Once AI has lowered the capital barrier to filmmaking, the scarce capability shifts from generating assets to taste, judgment, and final decision-making. 山音 calls this the “furrowed-brow instinct” (“眉头一皱的能量”): when AI writes 1,000 words, the human needs to know which 3 passages are wrong and which line unexpectedly works. KK stresses that the director remains the final arbiter of the finished film, and that AI leaves no excuses: responsibility for prompts, editing, music, and pacing all comes back to the individual.
- The editing timeline is becoming the control plane for AI filmmaking, collapsing scripts, assets, generation, rough cuts, and revisions from a linear workflow into a live loop. 《A-Ming’s Day》 went from a one-line idea to a locked cut in 3 days; 《Mangus》 took nearly two months because of KK’s other work, but can now start from a roughly 500-character outline and complete Agent-assisted expansion, character and scene assets, video generation, and editing inside CapCut Video Studio. “Text is text, the finished film is the finished film”: whenever a cut does not connect, it gets revised immediately.
- Creative supply can scale without limit, but life experience cannot, so the content bottleneck may arrive before the compute bottleneck. 山音 believes material that truly moves people must come from having “been there,” and more than a decade of accumulated experience could be consumed quickly at high production velocity. AI can produce many stories in a day, but it cannot replace firsthand experience of family, work, music, or life in the real world. Creators still need to replenish their experience rather than continuously extract from old reserves.
- The economics of an AI director look more like a low-capital, high-personal-leverage creator business than a traditional capital-intensive film project. Compared with films that routinely require tens of millions of yuan and cannot start shooting without financing, AI projects are already affordable to individuals, who can choose between taking commercial work and making auteur pieces. The other side of that freedom is full-stack pressure: income and financial security do not arrive automatically, while a traditional director can blame actors, cinematography, production, or budget; an AI director is largely left to own the result alone.
- Model progress is moving the workflow from image to video toward reference to video, with asset reuse and finished-film-level revisions driving the next efficiency leap. Nano Banana improves static consistency across characters and scenes, while Seedance makes characters, props, settings, and art assets repeatedly reusable. 山音 says that a 15-day production cycle was once an almost unimaginable speed; today, 15 days can produce a 5-minute film, and the same 5-minute film can now be finished in 3 days. The commercial value increasingly lies in being able to revise after the cut: the product, wardrobe, or even the male lead can be replaced wholesale, while music can be generated directly to match the existing edit’s rhythm.
- For Chinese creators, this is not an automatic lead but a chance to return to the starting line alongside global peers. KK believes the new production system weakens Hollywood’s influence through its established industrial workflow, while Chinese creators also bring long exposure to both Eastern and Western content. But both acknowledge that the industry remains in an early period of disorder: “no one is stronger than anyone else.” For international reach, KK reduces dialogue and exposition, while 山音 deliberately creates a hybrid cultural atmosphere that feels “behind a layer of fog.”
- The next investable wave may not be cheaper video, but new media that emerges once generation becomes faster than viewing. When a model can generate a 5-second video in 3 seconds, 山音 is more interested in real-time games, livestreamed games, and interactive forms that were previously unimaginable, saying the games moment “may already be on the eve of arrival.” Given a hypothetical $1M, KK would support 大扎秀 in completing a feature-length Journey to the West film, while 山音 would rather find a new film-game direction “that nobody has ever made.”
Deep dive
1. Two signature works began as creative impulses shelved for years by traditional costs
KK heard the Jiubao band song 《十丈通嘴》 in 2014 and 2015, then read the manga of the same name two years later. He was captivated by the image of a sister transformed into a giant beast hunting down her younger brother, but what he really wanted to film was the deeper story “that nobody else had told.” The host describes 《Mangus》 as starting from a Mongolian legend and rewriting misfortune and curses as family devotion and mutual support. As an advertising professional, KK had no high-budget route to bring this kind of work to the screen; AI finally provided an executable outlet.
KK did not outsource the story’s origin to AI. He first set the complete structure, core idea, characters, and roughly 500-character outline, then held multiple rounds of dialogue with the Video Studio Agent to develop the shooting script. AI entered at the text stage, but only after the story architecture, central theme, and character setup had been established.
山音’s 《A-Ming’s Day》 took the opposite route: there was no written script of any kind, only the small concept that changing an outfit changes a personality. He let AI improvise, then worked backward from visual ideas such as a wolf skin representing a wolf-like employee and Snow White’s dress representing a gentle customer-service agent, allowing the text and story to grow from those images.
2. A personal film no longer needs to revolve around a script; it can grow from an editing track
山音 was formally trained as a screenwriter, but he did not lock the script first for 《A-Ming’s Day》. The traditional process involved repeated adjustments across script, storyboard, on-set blocking, and on-set writing, often because the budget was insufficient, the actors could not deliver, or a scene could not be executed. In personal filmmaking, audiences ultimately see only one edit timeline, and the creator’s job is to hold on to the core thread in their head.
His view is that creation can start from the editing timeline itself: fill whatever the track lacks with blocking, text, art direction, or shots, rather than completing each department’s documents in sequence. 《A-Ming’s Day》 went from a one-line idea to a locked cut in 3 days. To make it happen, he shut himself off completely and told his partners, “Don’t contact me for 3 days.”
《Mangus》 took close to two months from start to finish, though KK notes that this was not pure production time because he was also doing other work. By the middle of the process, generation and rough cutting had become one loop: a storyboard that theoretically cuts together does not mean the finished film actually does. “Text is text, the finished film is the finished film.” Whenever the pacing or story was missing something, he returned to the same canvas and added shots in real time.
3. Ordinary people get an entry ticket, while the professional gap shifts to taste and trade-offs
KK believes self-taught creators have a real chance to make good work, citing MX Shadow’s 《Zombie Singapore》 as an example. But reaching that level still requires a large viewing history, accumulated taste, and a sense of rhythm. How much you have watched determines your taste—and whether you know when to hold back and when to let go.
AI can flag what to consider in structure, characters, and narrative, but it cannot make the final call for you. 山音 says people are not especially good at producing something from zero in a vacuum, but they are good at selecting, revising, and advancing from references AI provides. Even if only part of the output is usable, that part can become the basis for pushing toward a complete work.
The key metric is not how many years someone has studied directing, but how strong their “furrowed-brow instinct” is. If the 500th or 800th word of an AI outline makes you think something is wrong, that is precisely a sign that your own version is taking shape. As long as you do not accept the output wholesale, you still have the potential to make a good work—and you will not become “AI’s puppet.”
4. What truly moves people is the creator’s first jolt, not a polished formula
Both guests mentioned 《Paper Phone》 and 《The Exploded Hair Salon in a County Town》. When 山音 first saw the latter in his feed, he did not even notice who made it; he only thought, “This person is incredible.” After recognizing the creator as a familiar friend, he developed a competitive urge and watched it repeatedly. Since appearing in March, 《Paper Phone》 has remained the work in the warmth genre that moved him most over the following 6 months.
They observed that many later sentimental works copied the form of 《Paper Phone》 but lacked the same purity of original motivation. KK says you can immediately see which projects followed because an earlier example had gone viral and attracted traffic. Without “the thing that genuinely moved you in the first place,” it is difficult for the final work to have the same impact on anyone else.
山音’s rule for his studio is: “The less you think, the better you can make it.” He does not mean thinking less about execution; he means not calculating traffic, audience reaction, or master techniques before shooting. When a work stops moving forward, ask which shot, line, or emotion is something you “have to say or you’ll burst,” then build the drama from there.
5. 《Building Blocks》shows how lived experience rewrites a creator’s understanding of a subject
Around 2018, 山音 wrote a novel about Alzheimer’s disease. No one in his family had the condition at the time, so he relied entirely on imagination and fell into a heavy, stereotyped narrative about caring for the elderly. When his grandfather became critically ill in 2020 and could no longer recognize family members during his final months, 山音 gained a kind of firsthand experience he had never had before.
When he remade 《Building Blocks》, he stopped assuming that an elderly person remains continuously lucid and crushed by grief. Instead, he imagined the person first falling into confusion—no longer knowing the year, the people around them, or their own age, only sensing vaguely that something was wrong—until memory and age suddenly returned together and the grief arrived all at once. The first half of the film is therefore deliberately less sad, attempting to reproduce the subjective experience of facing the unknown.
KK counts 《Building Blocks》 among his favorite works and says that among the realistic or animated films on the same subject he has seen, “none surpasses it.” What truly sets the work apart is not an AI technique, but the source 山音 later acknowledged: “That period during the final months of my grandfather’s life.”
6. AI compresses the production cycle—and consumes finite life experience faster
The host notes that KK may have made 2 or 3 pieces a day in his early period, only for his creative cycles to lengthen later. A traditional director might make one film every 3 or 4 years and not worry about running out of inspiration until age 60; an AI director could rapidly exhaust more than a decade of accumulated experience. 山音 accepts the need to slow down because things you have truly seen, heard, and experienced cannot be restocked at generation speed.
He summarizes the source of creation as “being there”: the family setting, the music setting, and all kinds of real-life settings. Only as the firsthand experiencer can someone turn experience into visual language. AI can produce many stories in a day, but those stories may not have moved the director first; after another layer of transformation and information loss by the director, they are even less likely to move the audience.
Neither believes there is a single answer reconciling output with accumulation. Traditional cinema has Spielberg- or Eastwood-style high-output directors as well as Cameron-style long gestation. AI creators can spend 6 months refining one work or isolate themselves for 3 days and finish at high intensity after catching an idea. The key is finding a pace that matches the urge to express.
7. AI is best as a tireless creative sparring partner with no small talk
Even with no ideas at all, AI can provide an entry point. Recalling 《A Dazzling Leap》, a 24-hour work made at the Hainan Hackathon, 山音 said his method was to have AI generate 20 one-line story ideas, then select the direction that genuinely sparked something. A small anti-war premise was expanded into a complete work from there.
For introverted creators, AI removes much of the social cost of discussing a script with real people. A 2-hour human conversation might spend only 10 minutes reaching the soul of the script; with AI, creators can skip “Have you eaten lately?” and go straight to whether an idea works, where its flaws are, or what would happen if it were written another way. The system can respond consistently and indefinitely.
AI’s opinions are “not necessarily right, and not necessarily good.” Their value lies precisely in provoking a reaction. It may call a formulaic passage a highlight, while a human is struck by an unconscious line at the beginning or end. Capturing and extending that feeling of “this point works in such an unexpected way” still depends on human sensitivity.
8. AI lowers the organizational barrier but raises the director’s full-stack bar
The most important trait shared by traditional and AI directors is knowing what the finished product should be and serving as its final quality arbiter. The difference is that a traditional director must mobilize a large division of labor across actors, cinematography, art, and production. On a set where every second burns budget, the director also needs intense communication, organizational ability, and sometimes the ability to lose their temper.
AI suits creators who do not want to deal with people constantly, but it requires the director to understand every step of the traditional production process. Camera angle, focal length, lighting, performance, editing, and music all have to be explained to AI. KK sees this as the major dividing line between professional and nonprofessional AI creators. The closer someone is to the traditional-set director who is exhausted from managing everything, the stronger they may be after making the transition.
山音’s blunt assessment is: “When you do AI badly, there is no excuse.” On a traditional set, you can still blame the actors, cinematographer, producer, resources, or budget. When an AI work is bad, the problem comes back to the prompt, the edit, and the taste. As the barrier falls, AI also resembles “the tide going out and revealing who has been swimming naked.”
9. Low costs create career freedom, not low pressure
KK believes income depends on how hard a person works and how much they want to earn. 山音 has relatively modest material needs, takes on few advertising and commercial projects, and spends most of his time thinking about the next work. What he values is not maximizing income, but being able to “go make money immediately when I want to, and choose to make work that people will remember.”
A single film can easily cost tens of millions of yuan, and without financing it cannot begin production. The investment in an AI film is now affordable to an individual, allowing even a nonprofessional creator to start directly. The traditional director’s predicament of waiting for the next job and remaining in a prolonged “floating state” can therefore be partly rewritten by personal production capacity.
Freedom does not mean ease. In the past, a director could ask the art department for a proposal, send an executive director to the set, or hand some decisions to other departments. Now every detail has to be controlled personally. KK says project pressure may even exceed that of traditional work because “you have to do everything yourself.” More of the fate is in your hands, but so is more of the responsibility.
10. From frame-by-frame lottery tickets to asset reuse, model upgrades rewrite the unit of production
Early image-to-video workflows required a complete static storyboard and every keyframe up front, followed by converting the images one by one into video and editing them together. Characters and settings could not remain consistent. When KK made 《Mangus: The Age Without a Master》, roughly 95% of the shots were generated with Midjourney, yet the characters’ faces could still be “worlds apart,” forcing him to rely heavily on Photoshop for corrections.
Nano Banana marked a clear turning point. A single character or background image could begin generating different angles of the same character or setting, improving static consistency. It did not eliminate all control problems, but it reduced the cost of repeatedly retouching images by hand to connect adjacent shots.
After Seedance, the workflow moved further toward reference to video. Creators only need to prepare the characters, settings, props, and art assets, then generate video from text-based storyboards and reuse those assets repeatedly. There is no need to pre-build keyframes for every shot.
山音 made his first AI film in February last year and spent roughly half a month on it, which he says was an almost unimaginable speed at the time. Today, 15 days can produce a 5-minute film, while the same 5-minute film can now be completed in 3 days. More importantly, he no longer has to make character consistency—the kind of issue a traditional director would not normally have to center the job around—the core of his workflow.
11. The moat of an integrated tool is the closed revision loop, not a single model call
Both guests say CapCut’s first differentiator is its model coverage, including overseas models unavailable in some domestic products. It also supports Chinese-language interaction and offers more varied aesthetic options. Video Studio connects the front-end Agent, infinite canvas, image and video generation, asset management, and editing into an almost end-to-end workflow.
山音 especially values its origins in editing tools. Professional software once made editing feel like a high barrier; CapCut made him realize that “editing can actually be this simple.” Unlike canvas products that begin with a script and then move through separate departments, an editing-native tool means every generative capability ultimately serves an existing timeline.
While in Singapore, KK learned that a finished cut could still be modified as a whole. If an ad was already edited and the client wanted to change the product, the wardrobe, or the male lead to a female lead, there was no need to regenerate and re-edit every segment. Music could also read the cut’s content, rhythm, and emotion and generate a score directly. He had thought his process was already highly efficient, but “actually found that it could be improved further.”
12. AI gives “dragon-slaying skills” an executable outlet for the first time
Answering a question about his broken film-school dream as a freshman at Shanghai Theatre Academy, KK described what it felt like after entering the school. He had thought the rare talents he had struggled to find were scarce, only to discover that “they were everywhere here.” He felt like he had “gone through endless hardships only to become one of the 100,000 heavenly soldiers sent to defeat Sun Wukong.” When every classmate could write scripts, had ideas, or already had stage experience, he began to wonder whether China really needed so many screenwriters and directors.
With no professional outlet, he came to see what he had learned as “dragon-slaying skills” and for a time stopped enjoying films, plays, and art. After encountering AI and accidentally completing his first work, no outside force required him to continue, yet he kept telling himself, “Try making one more.” After living through the Lola era and the various uncontrollable early stages, he realized he did not hate film; reality had simply made that love unbearably painful.
After Seedance became popular around the Spring Festival, friends who had left film and television years earlier for games or education came back asking how to use it. KK sent them a few prompts, and they responded, “It’s that simple.” His advice for getting started is therefore not to first read DeepSeek From Beginner to Mastery or take 200 classes, but simply: “The most important thing AI needs you to do is start.”
13. On junk content and good ideas that used to be suppressed
KK spent 10 years in the 4A advertising industry, where the familiar process was to make the first draft as creative as possible, only for the client to ask for it to be toned down as “too far out.” For a good idea to reach the public, the creator, the ability to sell the pitch, multiple layers of client decision-making, and media promotion all had to align. The same brilliant idea could remain forever in the reject pile if the creator lacked social or sales skills.
He worked with many top-tier celebrities. The final pieces were often fashionable, visually polished, and popular with fans, yet had “not much creativity.” The client was buying the celebrity’s influence and only wanted viewers to make a purchase, not to think. The industry calls this kind of technically competent but inexpressive production “dry work,” a term that carries a sense of emptiness and disparagement among creators.
At the end of 2024, the Yili AI advertising competition became a turning point. After working on mass-market consumer projects including Ambrosial and Qiaolezi since 2013 and 2014, KK received the brief, completed the work quickly, and won the national championship. AI gave this “frustrated advertising man” a way to release his surplus creative energy for the first time, and prompted him to consider leaving his job to take on advertising projects independently.
Asked about the claim that AI has increased the volume of junk content, 山音 responded that genuinely good work can also be seen faster precisely because AI exists. Good ideas previously blocked by layers of approval now have a chance to surface. KK acknowledges that those 10 years built his taste, values, and know-how, but also imposed an opportunity cost. If AI had arrived 10 years earlier, his development might have been completely different.
14. Initiative, personal style, and the ability to ask questions become the new educational dividing lines
KK believes the AI era especially favors people with initiative—those who try things immediately rather than rejecting them first. Recent graduates have no burden from the old system and “nothing to lose,” so they face pure opportunity. The least suited are people who procrastinate continuously and keep searching for flaws in the tools as an excuse, because after every model upgrade they will find a new excuse.
山音 believes people who only do “dry work” are more exposed to disruption. Actors, cinematographers, and other specialists also need to transform and develop professional styles that AI cannot easily replicate. He uses Stephen Chow as an example: that highly personal form of performance would not be replaced by AI even today.
Their shared direction for educating the next generation is to turn people who know how to answer questions into people who know how to ask and solve them. Scoring 100 on an exam or 750 on the gaokao does not mean someone can handle a real-world problem, much less tell AI what they can do that cannot be replaced.
KK compares this to learning how to make wishes. Faced with a system that is almost omniscient and omnipotent and can fulfill a huge number of wishes, many people do not even know what they want. 山音 asks his team to pose 10 questions every day and open a chat window, even if the question is only what to eat that day. Curiosity takes shape through the act of asking again and again.
15. Global competition offers a common starting line, but China’s advantages still need to become concrete work
The host cited international results including KK’s 《Mangus: The Age Without a Master》 being selected for the 30th Rhode Island International Film Festival and 《Tango》, made by 山音 and his teammates, winning gold at the Cannes International Film Festival’s Sci-Fi Film Summit. Their discussion focused less on the awards themselves than on whether AI makes it easier for a new generation of Chinese directors to reach global audiences.
KK believes the answer is yes. AI is a new form of productive capacity distinct from the traditional industrial system, and creators in both East and West are writing the film and television workflow at the same time. Hollywood’s old production machinery once shaped content direction and value expression; the new system gives creators a chance to escape some of that path dependence. “At least in this respect, we are all standing at the same starting line.”
Both reject declaring a winner too early. Counting from Seedance 2.0, the technology is “not even 6 months old,” and no truly global breakout figure has yet emerged. China’s potential advantages include population, technology, official and platform support, and familiarity with Hollywood, European, and local cultures. Its range of subject matter may therefore be broader.
KK pursues universal readability through fewer lines of dialogue, less narration, and less background exposition: “As long as you are human,” you can understand the emotion. 山音 instead pursues a hybrid culture “hidden behind a layer of fog,” so different audiences feel that a work resembles a Chinese story without being entirely one, creating a shared point of entry that is both recognizable and ambiguous.
16. “AI director” will eventually disappear; what remains is taste, judgment, and model selection
KK believes the label “AI director” will inevitably be discarded. Once AI becomes a standard household and personal tool like the computer, people will not ask which model was used or who it was consulted with. They will only judge whether the story is good enough, whether the images are good enough, and whether the story is told compellingly.
The most irreducible part of directing is taste and decision-making—the “furrowed-brow moment” 山音 describes. Give 20 people the same ideas and different directors will choose different ones. 山音 even offers a “somewhat bold view”: traditional directors mainly exist to coordinate a crew, while AI directors correspond to smaller teams and greater decision-making power. They are therefore “closer to the image, closer to the work” than traditional directors.
The tool layer will not become a one-model market either. 山音 says Stable Diffusion 2.5 is, in some respects, more “floaty,” making it suitable for imaginative actions that other models would have to split into multiple shots. Fixed-camera shots, handheld work, and simple movements may call for more restrained models. Commercial delivery may also still require approval of Image-to-Video keyframes, allowing clients to confirm each stage before video generation.
17. Once generation outpaces viewing, the real new market may lie beyond film
When a model can generate a 5-second video in 3 seconds, content has a first opportunity to be “generated faster than it can be watched.” 山音 believes its significance is currently more inspirational than commercial for video, but it opens the door to real-time interaction, livestreamed games, interactive film-games, and unknown formats. After the moments of large language models, Nano Banana, and Seedance, “the games moment” may already be on the eve of arrival.
Faced with a $1M investment question, KK says a pure AI video creator might not need that much capital. He would rather back 大扎秀, who spent 1 year making 《The Battle of Che Chi Kingdom》 and is using AI to continue the Journey to the West series. KK hopes funding could help him make a feature-length film. 山音 does not want the “interactive film” format that already existed 10 years ago; he wants to find a form of image-making and gaming that nobody has made or even imagined.
山音’s question is: assuming technology stays at its 2026 level, what would creators make if given another 10 years? He hopes Chinese filmmakers will use this “wind of equalization” to define the language of cinema for the next 20 years and beyond, using new technology to reinterpret “where this people came from.” The names that will one day be written into film history “may already be around us.”