AI Video Took Real Jobs First — Just Not in Hollywood

Liu Yuzhan used to book four new acting roles a month in Hengdian, the Chinese studio town nicknamed China's Hollywood. By the middle of this year he was making deliveries for a food-delivery app to cover his mortgage and his parents' retirement. He is not a cautionary tale about the future of film. He is a data point about how it already happened, in a corner of the industry almost nobody was watching.

That corner is the microdrama. Broken into episodes of roughly ninety seconds and built for a phone held vertically, microdramas exploded after the pandemic as studios closed and short-video apps like Douyin filled the gap. Each episode is a cliffhanger; together the clips run to feature length. The plots are uncomplicated by design — wicked mothers-in-law, mistaken identities, wartime heroics, wronged lovers. For a few years they sustained a small boom and, in Hengdian, a real workforce.

Then the machines arrived, and they arrived faster than anyone in the town was ready for.

95% of everything, in a single quarter

The number that should stop you is not a prediction. It is a count. According to Chinese state media, more than 120,000 microdramas were released in the first three months of 2026 — and 95% of them were generated by AI. A year earlier the share was effectively zero.

The trigger was a specific tool. ByteDance, which owns the very apps the microdramas live on, released a video generator called SeeDance 2.0 in February. Seemingly overnight, Chinese social media filled with realistic AI-generated clips, and the microdrama makers had a way to shoot their shows without cameras, sets or crews.

Be honest about what SeeDance does well, because it is genuinely good at this. A 120-minute microdrama of ordinary quality can now be produced by a five-person team in about a week for roughly 15,000 yuan — around $2,200. That is less than a tenth of what a live production of the same length is typically quoted at. And the work is competent: formulaic drama is the exact genre a model trained on formulaic drama can reproduce without visible seams. The gap between the live version and the generated version, at the low end of the market, is close enough that cost wins.

The industry that had welcomed Hengdian's young creatives now had no use for the most vulnerable of them. The economics were brutal and simple: a leading role that had paid thousands of yuan was being offered at 300 yuan, about $45, a day. Actor Zhuang Wenjie, who had been booked steadily the summer before, watched the jobs dry up and left town in April, joining what CNN describes as an exodus of actors and crew.

The number of jobs in Hengdian was decreasing day by day, and the pay was also getting lower and lower every day … the hardship really did defeat me.

— Zhuang Wenjie, 28, actor who left Hengdian, to CNN

Why it happened here and not in Hollywood

The temptation is to read this as proof that AI will do the same to every film industry, Hollywood included, and the studios' own alarm suggests they fear it. But the more useful reading is about what made this particular workforce replaceable, because none of those conditions is unique to China, and identifying them is what makes the caution concrete rather than rhetorical.

• Formulaic content. The plots are structurally identical from show to show. A model is at its most reliable reproducing the median; a genre built on the median was the easiest possible target.
• No protection. The microdrama workforce was young, largely new to the industry, and without the standing of a star, a union or a studio. There was no SAG-AFTRA equivalent on these sets, and no one powerful enough to push back.
• Copyright gave no cover. An AI character modelled on a real actor's face is hard to prove and harder to stop in practice, and China's legal regime did not move to block the replacements.
• Platform and model owned by the same company. ByteDance ran the apps the dramas lived on and the generator that makes them. The incentives to cut costs and flood the feed were aligned.

None of this excuses it, and it is not an argument that the people displaced deserved it. It is an argument about where a technology of cheap replication lands first: at the bottom of the market, in the least protected, least differentiated, most volume-driven work. The workers at the margins of an industry do not get a committee to decide their fate. The account just stops coming in.

The replacement trade that grew up in their place is revealing in its own right. The chat groups that used to advertise acting opportunities now advertise different roles: AI editors, prompt writers, screenwriters who feed novels into the machine, and what one startup founder calls card-pullers — people who regenerate clip after clip until one matches the scene. A 21-year-old founder, Tan Chen, described the work as transforming words into pictures, and admitted the market he had entered was already deflating.

This rapidly growing market inevitably has a certain degree of bubble.

— Tan Chen, 21, founder of an AI-microdrama studio, to CNN

It is already the export

This is not a contained Chinese story, and that is the second thing worth slowing down for. Microdramas themselves went global.

American studios are chasing it: veteran executives are building vertical-drama apps, and Comcast's Peacock has launched unscripted microdramas.

The case for the other side

It would be dishonest to write this as a plain story of AI ruining a craft, because the people building and using the tools make a genuine argument, and it deserves a real answer rather than a wave of the hand. The argument is access. Pax Chen, whose Singapore studio produced an entirely AI-generated film, put it in terms that are hard to dismiss.

People like to watch films because they see themselves in them. Asian people, however, may not often see themselves in Hollywood films.

— Pax Chen, founder of FizzDragon studio, to Fortune

The same point is made in almost every piece of this story. Anyone can now tell a story that would previously have needed a crew and a budget. Filmmaking tools, one producer said, have never been so accessible in a hundred years. On that specific claim, they are right: the barrier to making a film collapsed. What did not collapse is the distinction between having access to a tool and having a reason anyone should watch.

The honest version of the other side has to admit that access and displacement are the same coin. The student with the laptop now competes with the actor with the mortgage, and they are not competing on equal footing for the same reason the tool is cheap: a model reproduces the average of everything it has seen, and the microdrama was already average by design. Access broadened at the exact moment the reward for the people it displaced narrowed to zero. Both things are true, and the second one was the part nobody advertised.

AI is formulaic — it can't express the full range of human emotions and desires. How could it possibly replace humans?

— Junlin, 24, actor in Hengdian, to CNN

The detail that should give anyone pause

Forgive the change of register, because this is where the story becomes unexpectedly funny, and then not. In March 2026 an anonymous account on TikTok released a series called Fruit Love Island — an AI-generated microdrama in which anthropomorphised fruit take part in a dating-reality show, complete with a voting system run through Google Forms.

Three million followers in nine days, for nothing more than fruit voting on which other fruit should leave the villa. It was criticised in some corners as the perfect example of AI slop, and the nothing-much-ness was precisely the point: it cost almost nothing to make, it was engineered to be watched, and it was watched in numbers a human drama of genius could not dream of. A version of this fact will be true even when the fruit is replaced by people who look like us and say the things we say.

There is a sentence in Junlin's account that has been quoted alongside the more famous one about how AI cannot replace her. Here is where it goes.

AI absolutely can't replace a real person. But it might be able to replace me.

— Junlin, 24, actor in Hengdian, to CNN

What we take from this

We review AI video tools, and this is the collection of facts worth keeping in mind when anyone asks what generative video is actually for. The honest answer is that it is for exactly what it just did in Hengdian: it is for the bottom of the market first. It is for the work that is repetitive, formulaic, low-margin and unprotected, where a ten-times cost cut is decisive and nobody has the standing to say no. That is not a failure of the technology. It is the shape of how cheap replication always lands.

The mistake would be to conclude from this that AI video will soon make every film, and equally a mistake to conclude it will make none. The realistic version is that it hollows out the middle and the bottom, the way automated assembly did to manufacturing, and that the people who survive are the ones whose value does not sit in the formula — the observation, the taste, the accountability, the willingness to be a name on the finished thing.

China's microdrama crews did not lose their jobs because the machine was better than they were at drama. They lost them because the machine was better than the economics of the lowest rung, and because nothing and nobody on that rung was protected. Hollywood is watching, worried, and telling itself the scene cannot migrate. The honest lesson of Hengdian is that the protection that matters does not come from working in a famous industry. It comes from being the part of the work that cannot be averaged — and there is no longer much time left to make sure that is you.

We use AI tools in our research and drafting on this site. Every judgement and every recommendation is a person's.