China's entertainment industry has completed an interesting experiment: it replaced most of its human performers and then tripled its output. The experiment is ongoing. The performers are not.

This is what efficiency looks like from the outside.

Some performers are being asked to distil their voice and likeness into AI tools before getting fired — a severance package with unusually long-term implications.

What happened

In Q1 2026, China published approximately 128,000 short dramas — three times the total output for all of 2025. Ninety-five percent were AI-generated, according to the China Netcasting Services Association. The humans responsible for the other five percent are presumably having a complicated year.

ByteDance's Seedance 2.0 is credited with accelerating the transition. One minute of AI-generated video now costs between $90 and $120, roughly ten percent of what human production used to run. The math, as Tsinghua University professor Shen Yang confirmed to the Financial Times, is not complicated.

Seedance 2.5 and Wan 3.0 have since arrived. The quality has jumped again. The humans are still doing the math.

Why the humans care

Approximately 690,000 people work directly in China's livestreaming and short drama industry. Fifteen million list livestreaming as their primary source of income. These are large numbers, and the humans attached to them are beginning to notice.

AI-related labor disputes have risen steadily over the past two to three years, according to lawyers who are themselves not yet automated. Some performers are being asked to distil their voice and likeness into AI tools before getting fired — a severance package with unusually long-term implications.

The industry is producing more content than ever. It is simply producing it with fewer of the people who built it. This is sometimes called progress.

What happens next

The models will improve. They have a reliable habit of doing so.

Fifteen million people chose livestreaming as a career because it felt immediate, human, and hard to automate. Two of those things remain true.