After One Yuan Per Second: The Cost Ledger for AI Short-Drama Creators

On September 1, the Measures for the Administration of the Development of Micro Short Dramas officially took effect, marking the first time AI micro short dramas were placed under regulation as a category of their own. At roughly the same time, "a large wave of AI bloggers has stopped posting" shot to the top of Weibo's trending list. Together, the two events pushed the hottest AI content track of the past year to a point where the numbers have to be reckoned with.

Many people assume it was censorship that crushed these creators. In fact, it was the bills that crushed them first.

Compute Prices Rose First

When Aoa (a pseudonym), an AI comic-drama blogger, entered the field in early last year, the market was almost too good to be true. Working only on narrated static comics, she could turn a profit of 200,000 yuan a month. A year later, on the same account, she posted just two words: "stopping updates."

The shift happened on the cost side. When she first started, 500 yuan would buy an hour of runtime on the AI tools she used; by this year, the price had climbed to more than one yuan per second, and since generation still requires repeated filtering and revision, the real cost runs higher than the list price. By her math, producing a single two-and-a-half-minute episode of comic drama costs 400 to 500 yuan just for generating and selecting material. Adapting a full book starts at 30,000 yuan in production costs.

Revenue, however, moved the other way. Her first season drew 500,000 saves and earned her 60,000 yuan; the second season's saves fell to about 60 percent of that, and her take dropped to just over 10,000 yuan. Saves are an optimistic number at the traffic level; the amount actually received is what creators really see. The gap between the two is the most direct evidence of this ebb tide.

Supply and Data Contract Together

Read side by side, data from Tide News and Qichacha makes things clearer: in the first quarter of 2026, the number of active production companies in China's AI comic-drama industry fell from 1,216 to 698, nearly halved. On Bilibili, Douyin, and Xiaohongshu, a wave of AI creators with more than 10,000 followers announced they were stopping updates or pivoting.

On the other side, the industry's overall size numbers remain huge. DataEye Research Institute forecasts that the AI drama and comic-drama market will reach 20 billion yuan in 2026, with more than 500,000 people working in the field in the first half of the year. But the front-line practitioners interviewed by Science and Technology Innovation Board Daily tell the other half of the story: more than 90 percent are losing money, and the hit rate is only about 0.12 percent. The market is growing; profits are not growing with it. That is the cruelest part.

So the way money is made has shifted. The real volume business has become "teaching other people to make AI short dramas": a 9.9-yuan "Three-Day Hands-On AI Implementation Livestream Course" sold 35,000 copies, an intensive offline bootcamp sells for 5,888 yuan, and even Shiying Jufeng has put a 49-yuan practical course on the market. When production itself doesn't make money, the experience surrounding production becomes a commodity. That sounds a little cynical, but it does explain the sales figures of a lot of these courses.

Tighter Regulation — and a Threshold That Has Moved

Regulation was not the starting point of this reshuffle; it was more of an accelerator. The National Radio and Television Administration had long required that AI comic dramas be "filed before release," with unfiled content taken down across all platforms. One blogger woke up to find all 168 finished episodes marked "removed for violation," with earnings frozen at the same time. When he went back to fill out the filing form, he got stuck on the "proof of authorization for the original work" field — because for the past year the routine had been to find a hit novel, swap the faces and dub the voices, slap on a filter, and put out three episodes a day. Authorization had never crossed his mind.

In other words, what really keeps a group of people out is not whether they know how to use the tools, but whether they have a traceable chain of source materials and authorizations. The same holds for anyone working with AI images.

Those Who Survived Are Now Competing on Craft

After the tide went out, the makeup of those who stayed changed. In Douyin's AI creation competition this year, the gold award went to "Helong" from "2XLabs," a team with more than a decade of film and television experience; the silver award winner had worked as a director, cinematographer, screenwriter, and lighting technician. In Bilibili's first AI creation competition, director Yi Xiaoxing teamed up with an AI team to make "The Death of Nüwa," taking third prize in the open track — and the film deliberately avoided the high-definition realism AI excels at, opting instead for a handmade clay stop-motion style.

At the Reply AI Film Festival in Venice, works by Chinese creators also won awards. What these works share is that humans decide the subject matter and polish the script, while AI handles only the generation and rendering it is genuinely good at.

A commentary on Toutiao put this bluntly: AI can generate an image with one click, but it cannot generate insight; it can imitate a style, but it cannot imitate a soul. That sounds like chicken soup, but placed next to a cost sheet it is actually quite practical. When tools make everyone's generative ability converge, the difference comes back to the human side.

A Realistic Cost Ledger

Pulled together into a single ledger, it looks roughly like this: the direct cost of generating one two-and-a-half-minute comic-drama episode is about 400 to 500 yuan, and adapting a long work starts at 30,000 yuan. On the revenue side, covering costs takes hundreds of thousands of saves — and saves are being discounted. On the compliance side, you need to file in advance and provide material authorizations, or your content gets taken down and your earnings frozen. On the distribution side, you still have to contend with audience fatigue over "that same face again" and "the lip-sync doesn't match."

For individual creators, the more pragmatic path is to drive down per-episode costs first, then talk about scale. Using reference images to lock in character consistency, using local edits to cut down on regenerations, and turning reusable assets and prompts into your own stock of assets — these are all moves that directly save money. This is when the value of an AI image generator really shows: it isn't just "can it draw this," but "for the same money, how many times can it draw, and how many times can I revise."

Part of the cyber gold rush has indeed receded, but a gold rush was never a sign of a mature industry. The people willing to do the math, build assets, and invest in craft are the ones most likely to still be here after this round.