Drivers and a stay-at-home mom made an AI history drama claiming 12 billion views

What's going on

A TV series whose visuals were generated entirely by AI is making waves in China. It's called "持械入宋", or "Armed Into Song" in English, loosely "Armed into the Song era". The plot: five modern elite soldiers travel back in time with their firearms, landing exactly five years before the "Jingkang Incident". That's the name Chinese history gives to the fall of the Northern Song dynasty, when its capital was captured.

According to a November 8 report by the China News Service, ten seasons totaling about 1,000 episodes premiered on the short-video platform Hongguo (红果短剧) between late October and early November. The producers say the series had passed 12 billion views across all platforms by November 7. That figure comes from the creators themselves and hasn't been independently verified. The show belongs to the micro-drama genre: very short episodes, often just one to two minutes long, made for smartphones. In China, that's a huge market.

A team with no film experience

What's really drawing attention is the team. Its core is six people living in different provinces. According to business outlet 36Kr (as cited by BigGo Finance), none of them has any film or TV training. Before this, they worked as ride-hailing drivers, small-truck delivery drivers, in wholesale, or were mothers looking after their kids at home. With around 30 additional helpers, they produced the first season in just 20 days.

The team is led by 54-year-old Zhang Hansheng. He only got into AI filmmaking in March 2025, starting with an online course that cost 1,000 yuan, about $150. He then became the eleventh student at a studio. That's where the group's name comes from: "Laoshiyi Theater", roughly "Old Eleven's Theater".

Zhang is more cautious about money and success than many reports. Media outlets have talked about 10 million yuan in revenue. That's a media estimate and it's wrong, Zhang says; he hasn't received anywhere near that much. The series cost around 1 million yuan, almost all of it for compute. Staff costs were minimal because each member works as a one-person business. The rights to the source novel were free. And the team never paid for reach:

"Once the series is finished and has passed review, it goes out on several platforms at the same time, and that's it."

The first season did only moderately well. After that, each season grew more popular. The fourth topped Hongguo's chart of new AI dramas, and on November 4 a Weibo hashtag about the series showed up among the trending topics. On the platform, a single season hit a peak of 82.06 million on its so-called heat score. Across all seasons, the series has been saved to favorites 11.41 million times.

How an episode gets made

Each member makes their episodes alone, from start to finish. They read the novel, break it down into a script and write the prompts, the text instructions for the AI. Then they generate the videos and edit them. The finished pieces go into WeChat groups and cloud storage, where the others review them. There are hardly any meetings, since some members work at night and sleep during the day.

The team generates the videos mainly with Seedance 2.5 and uses Doubao for text. Both models come from ByteDance. Director He Xiushan singles out one new feature as especially useful: version 2.5 can draw on more than ten reference images instead of the previous nine. That helps with scenes involving lots of characters and with large battles. To keep characters and locations looking consistent across hundreds of episodes, the team maintains an asset library: a shared collection of images of characters, sets and props that everyone can access. But you can't lock in personality and emotion with a consistent face, Zhang says.

It's still a lot of work. A regular shot takes about three generations on average before it's usable; a complex fight scene often takes more than ten. A clip with no continuity errors, normal acting and normal camerawork only scores about 60 out of 100 internally. "That's just really flat," He says. In season six, the character Qin Hongying dies in battle. For the mourning scene, He wrote a song himself and cried while working on it. Zhang: "If you don't cry yourself, the audience certainly won't."

The team found the source material in the rights library of the novel platform Fanqie. At the time, the novel had fewer than 100,000 clicks. It was written by an English teacher who goes by "Chagen'er". Zhang liked that the time travel works without a "system", the genre term for magical aids that make heroes almost all-powerful. Here, by contrast, the rifles jam, the ammunition runs out, and the soldiers have to deal with the Song era's household registration, money, grain and military system.

Why historical stories

The main source, smzdm.com, explains why historical dramas are such a good fit for AI. Viewers remember characters and events more than faces. That makes a typical AI weakness less noticeable: faces that lose their shape from shot to shot. Sieges, imperial court scenes and massive armies cost a fortune in a real shoot. With AI, "one prompt builds a whole city". But you need to know your stuff. Clothing, objects, forms of address and official titles have to be right, or viewers will immediately call it out in the danmaku, the comments that scroll right across the video.

The team checked the colors of officials' robes, jewelry, shoes, ID tokens and travel papers, as well as seating order at ceremonies. Some viewers explain Song-era history in the danmaku. Fans have even built a platform that sorts official titles and battle maps into three tiers: historical fact, invention and speculation. State media praised the series too. The magazine Ban Yue Tan ran a commentary titled "This micro-drama is different", and the Economic Daily and Guangming.com struck a similar tone. Their message was always the same: AI can lower costs, but that doesn't mean standards should drop.

Criticism, complaints and tighter oversight

There's criticism, too. Viewers complain about "AI faces" and lots of filler across the 1,000 episodes, and accuse the series of distorting history. On Weibo, the hashtag "#AI历史剧火了也遭举报#" briefly trended, roughly "AI history dramas are a hit and still get reported". Many fans fear that complaints will get episodes blocked or deleted. Some would rather wait until seasons are complete and then binge them.

Then there's politics. On November 5, China's broadcasting regulator, the NRTA, reaffirmed its line: it wants to promote micro-dramas with real actors and regulate AI micro-dramas. Platforms must ensure that every episode is fully reviewed by humans before it's released ("凡播必审、全量人审"). A figure from market researcher QuestMobile shows how big the market is: Hongguo has 168 million daily users, more than the four major streaming services iQiyi, Tencent Video, Youku and Mango TV combined.

The team has changed how it works, too. It used to run five production groups in parallel, turning out about 20 series a month. Zhang himself calls some of them "assembly-line series", meant to get members earning money first. Now it's fewer than ten a month, and the team only wants to make high-quality series. Anyone paid by output only gets paid for the minutes that actually make it into the series after review. The city of Shijiazhuang has offered an office in Jingxing district, rent-free for the first two years. But the team wants to keep working online.

The bigger picture

The story shows how far video models like Seedance have come: a small team with no training can use them to mass-produce a series that finds an audience of millions. That said, the 12 billion views figure comes from the creators themselves, and view counts for micro-dramas are hard to compare. Zhang himself warns against reading the story as proof that "anyone can do this". AI mainly levels access to the models, he says. A feel for storytelling, an understanding of characters and good editing don't come with it automatically.

What happens next also depends on regulation. If every episode has to be fully reviewed by humans, that slows down exactly the fast AI production that makes series like this possible in the first place. The main source's advice to viewers is down-to-earth: if a historical scene moves you, check the facts behind it yourself.