China's Short-Drama Industry Is 95% AI-Generated

For years, the AI displacement debate has been a war of abstractions -- percentages, timelines, vague warnings about labor markets. In China's short-drama industry, the war is over, and the machines won.

In the first quarter of 2026, approximately 128,000 short dramas -- quick-turnaround vertical videos optimized for mobile consumption -- were released in China. That is three times the total for all of 2025. Ninety-five percent of them were AI-generated, according to the China Netcasting Services Association. The industry employs 690,000 people directly. Fifteen million list livestreaming as their primary occupation.

The scale alone would be remarkable. But one detail from a Financial Times report shifts the story from statistics to something more troubling: some performers are being forced to "distil" their voice and likeness into AI training tools before they are fired.

They are required to train their own replacements.

The economics of displacement

The math that makes this possible is brutal. One minute of AI-generated video now costs between $90 and $120, according to Tsinghua University professor Shen Yang, speaking to the Financial Times. Traditional production with human actors runs roughly ten times that amount. When the output quality is indistinguishable on a mobile screen, the decision is not complex.

ByteDance's Seedance 2.0, now updated to Seedance 2.5, and Alibaba's Wan 3.0 have driven the improvement. Both models produce broadcast-ready short-form video from text prompts. The gap between generated and produced content has narrowed to the point where audiences cannot reliably tell the difference -- and crucially, for a short-drama format that lives and dies on volume and speed, the audience does not need to care.

The human cost

The macro statistics -- 95 percent, 128,000 titles, 690,000 jobs -- describe an industry shape-shift. The micro reality is visible in stories like Xu Peng's. Xu was a successful short-drama actor. After AI displaced his role, he returned to his hometown and now sells vegetables at the local market.

His case is not isolated. Labor lawyers in China report a sharp increase in AI-related disputes over the past two to three years. The disputes do not concern unfair dismissal in the conventional sense. They concern the conditions of the dismissal itself: cases where companies require performers to contribute their vocal patterns, facial expressions, and mannerisms to training datasets as a condition of receiving their final paycheck.

This is a new category of exploitation -- not merely the replacement of labor, but the extraction of identity as a termination fee.

The technology behind the takeover

The shift was not gradual. ByteDance's Seedance 2.0 launched with the capacity to generate entire short-drama episodes from a text outline. Human actors were reduced to reference material: the model studied their performances, learned their micro-expressions, and reproduced them without the cost of scheduling, retakes, or residuals.

Seedance 2.5 improved temporal coherence -- the problem of characters flickering between scenes -- which was the last meaningful barrier between AI-generated short dramas and human-produced ones. Alibaba's Wan 3.0, released shortly after, matched the quality across a wider range of genres.

The result is a production pipeline where the human role has shifted from performer to training data source. The model does not need you on set. It needs you once, in a recording studio, so it can learn everything your face and voice can do.

A warning for the rest of the world

China's short-drama industry is a worst-case scenario unfolding at accelerated speed. It is not a proof-of-concept or a pilot program. It is a market in which 95 percent of production has already been automated, and the labor force is still employed -- for now -- largely because the industry is growing fast enough to keep some humans in the loop.

Western entertainment markets are watching. The same economic logic -- AI video at one-tenth the cost, indistinguishable quality on consumer screens -- applies in Hollywood, in advertising, in social media content production. The difference is that China's regulatory environment offered less resistance, and the industry was structured around volume rather than artistic prestige, making it the ideal incubator.

The lesson for the West: the moment AI video quality crosses the indistinguishability threshold in your market, the transition will not take years. It will take one quarter.

Sources

• AI-generated videos are already displacing actors and livestreamers across China's entertainment industry (The Decoder)
• China's Short-Drama Boom Is Now 95% AI -- And Performers Are Training Their Own Replacements (eGamers.io)
• This Actor Lost His Job To AI -- Now He Sells Vegetables In His Hometown (8days.sg)