AI comic drama production costs drop by 90%, money flows to computing power and platform revenue sharing

One model is to pay for shows out of your own pocket and bet on the platform's revenue share. The other is to take contract work, charging a production fee by the minute, with the finished show belonging to the client.

Money comes in through only these two openings; money goes out on labor, compute, and paid traffic for promotion.

Strangely, though, production fees have fallen by 90%, yet the people making shows don't feel any more flush. Where exactly did the saved money go?

……

According to industry quotes, in 2024 making an AI comic drama cost 15,000 yuan per minute; by 2025 it had dropped to 100,000 yuan per show, or 1,000 yuan per minute.

The price cuts haven't stopped. Huang Haonan, founder of Jiangyou Culture, said in October:

Batch-produced shows are down to just 400 yuan per minute; a 100-minute work costs less than 40,000 yuan in total. Premium shows are more expensive, still around 1,000 yuan per minute.

The tiers differ, and so do the accounts. Using batch-show quotes to describe premium shows, or the reverse, easily overstates things; but the money saved hasn't all stayed in the hands of the people making shows—some of it is now paid out somewhere else.

At the start of this year, in batch-produced shows labor, compute, and miscellaneous costs were still split 6 to 3 to 1; now labor is only 30% to 40%, and compute has become the biggest chunk. One practitioner calculated that the compute cost per episode rose from 17 yuan early on to 80 to 100 yuan.

Logically, as technology matures, compute should get cheaper and cheaper.

In April 2026, some cloud services actually raised prices instead. Notices went out in March, taking effect in April; compute cards rose by up to 34%, storage by 30%, and the input price of some Tencent Hunyuan models rose by 463%.

These increases aren't on the same basis and can't be added together, but whether it's compute cards, storage, or model APIs, whenever any one of them goes up, the people making shows have to redo their accounts.

Small teams especially have little bargaining power; when a supplier changes its price list, they can only accept it.

I came across a production retrospective: one work ended up using 95 storyboard images, selected from 1,700 generated images—roughly 18 to 1.

Characters and scenes have to stay consistent; the same face looks fine in one image, then changes in the next. To keep it from going off-model, generating dozens of times over is not unusual.

In the end only one image makes it into the show; the dozen or so discarded images before it still incur compute fees. Discarded images aren't cheap—it's just that no one cuts them into the final show.

Put the two things together, and the cost accounting looks different.

In the past, making animation, the money went mainly to people. Artists, colorists, in-betweeners, editors—every item was wages. Wages are relatively fixed; produce a bit more, and unit costs can be spread lower.

Now labor has been saved, and part of the money that used to go to wages has turned into money spent buying compute. The compute bill follows output: generate once, pay once.

Both look like costs, but their nature differs: wages are relatively fixed, while compute is billed by usage. At large scale you can get a cheaper unit price; small teams don't have that option.

You can save on keeping people, but buying compute means watching the price list. The barrier to production is lower, yet dependence on upstream compute is actually deeper.

Now look at the accounts of an individual creator.

One creator interviewed calculated: compute plus discarded images costs 1 yuan per second, editing 0.7 yuan per second, and at most 5 minutes of finished video per day.

For a daily update of 5 minutes, compute plus editing costs about 510 yuan a day. Yet a work that gets 60,000 plays earns only 45 yuan. However cheap production gets, the math still doesn't hold up.

Of course, one person doesn't represent the whole industry, but this account at least shows that lower costs don't mean the money can be earned back.

Douyin Group's short drama copyright center stipulates that the free revenue share for Hongguo and Fanqie depends on how long users actually watch. The algorithm isn't complicated—four numbers multiplied together:

new valid viewing time in the month, the per-unit price of time, the content-type coefficient, and the copyright coefficient.

If users click in and leave without watching a few seconds, the people making shows earn nothing either. This set of rules applies only to these two platforms; Douyin's main app has a different algorithm.

You see, what the producer submits is a show, but what the platform is really buying is the time users are willing to spend on that show.

Second, how much is that time worth, and who decides? The platform holds three knobs: the per-unit price of time, the content-type coefficient, and the copyright coefficient.

First, content type:

In February 2026, the coefficient for simulated humans was 60, 3D was 50, 2D was 40, emoji packs were 10, dynamic narration was 5, and static narration was 1.

In April, simulated humans fell from 60 to 40, 3D from 50 to 40, and narrated comics were no longer split into dynamic and static, both cut to 1. The other tiers didn't move.

In just two months, simulated humans dropped by a third, 3D by 20%, and narration by as much as 80%.

In July, the platform adjusted again. Simulated humans, 3D, and 2D stayed at 40, narrated comics went from 1 to 10, and shows with over 50 million Hongguo popularity got cash rewards of 10,000 to 200,000 yuan.

Coefficients can go down and can go up; whether up or down, the platform decides.

How much the people making shows can earn depends not only on whether viewers like it, but also on which type of content the platform is willing to pay more for; the copyright coefficient is even more direct—full copyright 2, non-full copyright 1. For the same viewing time, income can differ by half.

As for the per-unit price of viewing time, the platform doesn't disclose it.

The people making shows can't even calculate accurately how much they'll earn per hour a viewer watches, let alone bargain; the platform only has to adjust a coefficient, and income may change accordingly. The show is still the same show, the viewers haven't changed, yet the settlement price has.

Third, as shows pile up, what happens to the price?

According to DataEye Research Institute, in the first half of 2026, 221,900 new AI dramas and comic dramas launched on Douyin, and only 1,055 surpassed 100 million plays, accounting for 0.48%.

With more shows, the platform isn't short of content—it's short of people willing to spend time watching.

Practitioners report that the people making dramas are also getting a smaller and smaller share of the money. Around 2024, 10,000 plays could still earn about 100 yuan; later it fell to 50 yuan, then 30 yuan, and in 2026 many projects are left with only 5 to 10 yuan.

Some creators have posted their back-end data: 180 million plays, settled at 180,000 yuan. That averages 10 yuan per 10,000 plays; this is only revenue back-calculated from play counts, not the platform settling accounts directly by play count.

What's more troublesome is that a film being made doesn't mean anyone watches it.

To be seen, you still have to spend your own money buying traffic. There's a saying in the industry: of 100 yuan in revenue, about 70 yuan goes to promotion, and the producer ends up with less than 10 yuan.

Ad delivery monitoring data shows that on the Douyin side, AI short dramas are being promoted at 150 million to 170 million yuan a day. The more money is spent, the more it's still the audience's attention being fought over.

Taking contract work isn't easy either. According to practitioners' quotes, at the hottest point it was 1,500 yuan a minute; now it's been pressed down to 600 to 800 yuan, and the final payment isn't necessarily settled on time.

So this is where the business gets interesting.

Producers buy compute, and the price is set upstream; they sell content, and revenue is calculated by platform rules; to get more people to see it, they still have to spend money bidding for traffic.

The platform doesn't need to pay for how much the producer spent; it only looks at how many viewers the content can retain. Everyone wants to grab viewers, and traffic is still expensive.

Revenue sharing is the platform giving money to producers; paid traffic is producers giving money to the platform. The money goes around in a circle, and the producer may not have much left.

...

So on this chain, who made money? I divide the accounts of this industry into four books: the content book, the platform book, the IP book, and the compute book.

For contract work, an ordinary comic drama is 500 to 800 yuan a minute, premium CG is quoted at 2,000 yuan, and S-grade runs into the tens of thousands. As long as you can steadily take orders and deliver on time, the income is relatively easy to calculate.

If you put in your own money to make dramas, you have to bet on platform revenue sharing.

Tencent disclosed in June 2026 that a single comic drama had a revenue share of more than 20 million yuan, with a cost of about 400,000 yuan. However, the title and settlement details were not published.

It is a sample used by the platform to attract business, showing that a ceiling exists, not where the median is; on the other side, someone made 30 episodes for 13,000 yuan and got only 5 yuan per 10,000 plays.

In the same industry, one end is hit shows, the other end is the long tail; 0.48% is not the probability of profitability, it is the proportion of plays exceeding 100 million. But at least you can see how many works actually break through.

The content book is difficult precisely because neither end is under your control. Hit shows can make big money, but most works don't have that luck.

Now look at the platform.

Hongguo's plate is very big. Third-party data shows that as of July 2026, daily active users were 168 million, more than doubling year on year; 2025 revenue exceeded 15 billion yuan, a figure not officially confirmed.

But the platform isn't making money lying down either.

Kunlun Wanwei's 2026 half-year report shows that in the first half, the AI short drama segment's revenue was 1.535 billion yuan, up 163% year on year; costs were 499 million yuan, up 406%, and the segment's gross margin fell to 67.49%.

In the same period, the company's selling expenses were 3.340 billion yuan, and operating cash flow was negative 467 million yuan. These two items are overall company data and cannot all be attributed to AI short dramas.

The platform also has to spend money to grab users; it's just that compared with producers, it has more power to decide at what price content is settled.

In October, WeChat's Lüpaopao short drama offered up to 100% revenue incentives for high-quality new dramas. For the first time, producers got a second quote besides Hongguo, although this quote was bought with subsidies.

Having one more buyer is a good thing; whether the quote can last still depends on how long the subsidies can burn and whether the producers have another good drama in hand.

Money isn't only being added by WeChat: Volcano Engine's premium AI drama incentive is up to 200,000 yuan per title, requiring all use of Seedance and more than half at 2.5; Hongguo also put up a 3 million yuan guarantee, with a top prize of 1 million yuan.

The conditions are being raised together—model, popularity, duration—and if one item doesn't meet the standard, you drop a tier. Subsidies are buying premium shows, not providing a safety net for assembly lines.

The third book is the IP book.

Good stories don't become worthless just because AI can draw; Yuewen's 2026 interim results show that in the first half, copyright operation revenue was 1.614 billion yuan, up 41.9% year on year; short drama and AI comic drama revenue exceeded 430 million yuan, up 2.3 times year on year.

However, net profit attributable to shareholders of the parent company was only 135 million yuan, down 84.1% year on year.

IP business revenue is growing; the ticket is real, but the ticket does not equal profit.

Finally, the compute book.

This book was calculated earlier. Model companies and cloud vendors don't have to bet on which drama will be a hit; whether you shoot a feel-good drama or an art film, they don't care—as long as you are generating, they are collecting money.

Of course, they also have equipment, R&D, and operating costs; not all the money coming in can become profit.

With the four books laid out, the difference becomes clear; the platform holds the revenue-sharing rules and the traffic entrance, the IP side holds the stories, the compute side provides the production tools, but the content producers have to bear the risk that no one watches the work.

This is somewhat like web movies and mini-program short dramas back then: once the threshold drops, more people enter, and traffic instead becomes more expensive.

So do ordinary people still have a chance?

Take a term that has been worn out by people selling courses: one-person company. At the Yuewen Creation Conference, CEO Hou Xiaonan said that more than 5,000 individual creators and studios had signed contracts, and the top ten OPC creators all had annual incomes above 1 million yuan.

Huang Haonan, founder of Jiangyou Culture, said one person can make three in a month, while institutions are actually shrinking.

What they mean by one person is that you first have to understand content. To actually make a watchable comic drama, practitioners say it usually still takes 3 to 5 people and 7 to 14 days to complete. The director, the people responsible for repeatedly generating and screening images, and the editor each handle one part.

Huang Haonan said this is still a content market; if the script is no good, it flops. So what a one-person company can deliver belongs to the few people with judgment. "Zero foundation, 100,000 yuan a month" is just a sales pitch.

Small teams are not relaxed either. More and more people know how to do it, but not that many can steadily take orders, deliver on time, and keep making money.

On September 1, the Measures for the Administration of the Development of Micro Short Dramas took effect, and the filing and labeling of AI dramas also entered the regulatory framework. Compliance is a fixed cost, and the smaller the team, the less it can bear.

So what is truly valuable in this industry is the ability to judge subject matter, control costs, deliver steadily, and get works in front of audiences.

Back to the beginning:

For people making dramas, the production threshold has dropped a lot, but the truly hard hurdle to cross is distribution.

Data sources:

[1]. DataEye Research Institute, Douyin Group Short Drama Copyright Center, Kunlun Wanwei 2026 semi-annual report, China Literature Group 2026 interim results, China Literature Creation Conference, Tencent Video 2026 annual release, National Radio and Television Administration, incentive plans publicly disclosed by Volcano Engine, Hongguo and WeChat Open Platform, publicly disclosed cloud vendor price adjustment information, platform public data and practitioner interviews; data as of October 21, 2026

This article comes from the WeChat public account "Wang Zhiyuan" (ID: Z201440), author: Wang Zhiyuan, published by 36Kr with authorization.