Chinese AI short dramas drop from 5,000 yuan a minute to a few hundred in six months

Zhu Zhili set up an AI film studio in Shenzhen two years ago. He also runs the AI-generated content department at China Wit Media, a production company in the same city ^[17][18]. He told Reuters that officials from big cities and small towns now call him every day, asking him to move his technology or his company to them ^[17].

The offers behind those calls are specific. Shanghai began giving computing capacity and hosted AI models to micro-drama producers in May ^[4]. Beijing's Huairou district, a centre of the Chinese film industry, hands out vouchers that cut computing bills for AI short dramas ^[6]. Shenzhen offers cheaper rent and technical support to AI film talent, and some cities add living allowances ^[7][2]. Reuters compared the approach to the one China used to build its electric vehicle, solar and robotics industries ^[3].

The few hundred yuan a minute that studios are being pitched reflects both better generation tools and public money. The tools have improved: ByteDance's Seedance 2.0 went viral this year, and Kuaishou's Kling helped make Chinese companies major players in generative video, according to Digital Trends ^[20]. Pan's AI wedding scene cost about 2.3 percent of the conventional shoot, a gap of roughly 43 times ^[21]. He told Reuters that local subsidies and rent waivers brought his computing costs down further ^[9]. CCTV's per-minute average mixes both effects.

Of the new AI shows on Douyin in the first half, about 0.48 percent reached 100 million views ^[22]. A view count measures reach, and 100 million is a high bar, but more than 99 percent of the new shows fell short of it ^[22]. Viewers have complained about plagiarism and repetitive content ^[14]. Some read the box-office success of Niu Lai, a crudely animated film made without AI, as a reaction against the trend ^[19].

On conventional production, the evidence is one price comparison. A gap that size on one surreal scene is pressure on anyone bidding that kind of scene the old way. The reporting does not show conventional producers cutting prices or losing contracts.

A studio head or buyer weighing an AI bid can sort the job on two questions. The first is whether the saving holds at the unsubsidised compute price. The second is whether audiences have already accepted AI output in that format. When both hold, the conventional crew is competing against the AI price and is likely to lose on cost. A saving that holds in a format audiences have not accepted just adds to the long Douyin tail. In an accepted format whose saving depends on vouchers, the saving lasts only as long as the local subsidy does, and the fair comparison uses full compute cost. Where neither holds, the job adds to the overcapacity that Digital Trends reported the industry may already be heading toward, the problem other heavily subsidised Chinese sectors have hit ^[12].