Loading today’s stories
Product1 distinct publisher3 min readPublished
DataEye counts 89 of the top 100 animated dramas on Douyin as AI productions, and one animation studio published thousands of episodes in May at roughly a tenth of human cost. Those are supply numbers.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Take Zeelin's May run as the reference price. At $90 to $120 a minute, that month's output cost somewhere between $720,000 and $960,000 [2], and if the claim that this is about a tenth of the human figure holds, the version shot with people ran $900 to $1,200 per finished minute [1]. That is the entire mechanism: a line item fell by roughly an order of magnitude, with no regulator moving and no platform mandating anything, and the production method simply followed the cost.
The labour arithmetic is starker than the money. Five people over about three months is on the order of 450 person-days for one three-to-five-minute drama; one employee for a day or two is one or two [6]. Whatever else generative video is doing, it is removing between 225 and 450 person-days of paid work per title in a category that measures its output in tens of thousands of titles.
And tens of thousands is where the reading needs discipline. 128,000 releases in a quarter is about 1,400 a day [4], and "more than three times the entire previous year" puts that year below roughly 42,700 [3]. Every one of those figures counts things that were made, not things that were watched. The material carries no completion rates and no revenue per episode. Teams will describe this as audiences demanding more; what the evidence supports is that the marginal cost of a minute fell far enough that nobody has to forecast demand before producing. Note also that the two share figures are not the same measurement: one is a chart of animated titles on a single platform [1], the other is a trade association's estimate across all short dramas [3], and "AI-generated" is nowhere defined, so an AI-assisted edit and a fully synthetic cast land in the same bucket. Digital Trends dates the platform and studio numbers only as "May," while the 128,000 figure is tied to the first quarter of 2026 [11].
Separate the pitch from the practice. Amazon's AI Studio, per Reuters, builds internal tools for pre-production planning and post-production editing and says it involves writers, directors and actors at every stage [9]. In the short-drama market the same cost logic has already reached the performance layer: one actor-producer went from three shoots a week to, in his words, "everything just went" after Seedance 2.0 shipped [7], and performers are being asked to distill their images and voices into the systems before the work stops [8].
Anyone weighing this in another category should run the actual math: multiply the per-unit saving by annual volume, then subtract what consent costs. The only published rate anywhere in this material is the children's programme host's deal: one year's salary and veto rights over five years [8]. A saving that survives that subtraction points to a real pipeline. A saving that only works because nobody has priced the likeness yet means the volume has been bought before its value is known, in a category where the only number anyone is publishing is how much got made.
Ranked by verification strength, evidence, and original report placement.
According to DataEye, 89 of the top 100 animated dramas on Douyin, ByteDance's Chinese short-video platform, were AI productions in May.
The Financial Times reports that about 128,000 short dramas were released in the first quarter of 2026, more than three times the number released during the entire previous year.
About 95% of those releases were AI-generated, according to the China Netcasting Services Association.
Zeelin, an AI animation company, reportedly published about 8,000 one-minute short-drama episodes in May, with an average cost of around $90 to $120 per minute, about 10% of the cost when humans were used.
Professor Shen Yang of Tsinghua University told the FT that a three-to-five-minute drama that required five people and roughly three months to produce in 2024 can now take a single employee one or two days.
The Financial Times reports that video-generation tools including ByteDance's Seedance 2.0 are accelerating AI adoption in short dramas, letting companies produce higher-quality video faster while dramatically reducing production costs.
Distinct publishers with included, body-backed reporting in this cluster.
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Alibaba's $10.2bn placement funds an AI bill running 6.4 times its quarterly profit1 distinct publisher
leadership
TikTok Shop Is A Discovery Channel With A Payroll, And Amazon Still Takes The Order1 distinct publisher
leadership
What an AI filmmaker actually owns is the show bible, not the footage1 distinct publisher
build
Hugging Face's $13B process puts most teams' model pipeline under a single owner2 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One relay, four originals we cannot see
DataEye's 89-of-100, the Netcasting association's 95%, Zeelin's per-minute price and the Financial Times' 128,000 all arrive through a single Digital Trends write-up of other outlets' reporting. The attributions are specific and the numbers are checkable arithmetic, which is why this scores above the floor. But Digital Trends adds its own 'reportedly' to the Zeelin figures, leaves the two May datasets without a year, and no second publisher in our coverage touches any of it.
The production side has already switched
This is the dimension the reporting actually nails down. Three independent-looking counts point the same way — a near-sweep of Douyin's animated drama chart, a 95% AI share of a 128,000-title quarter, and one studio shipping 8,000 episodes in a month — and the practitioner accounts of vanished shooting work are what that switch looks like from below. Adoption of the tools by producers is not in question here; adoption by paying audiences is simply never measured.
Supply counted, demand assumed
The piece promises that AI performers 'are already competing for screen time' and that audiences may not know the difference — a claim about viewers, argued entirely with production statistics. Nothing here reports what the 128,000 titles were watched or paid for, and the human cost of the shift rests on two individuals, one of them unnamed. The direction of travel is well evidenced; the leap from a cheap finished minute to a changed entertainment market is the overstatement.
The toolmaker owns the scoreboard
ByteDance sells Seedance 2.0 and owns Douyin, the platform whose animated-drama chart supplies the story's most quotable statistic — a fact visible in the reporting and never remarked on in it. The 95% share comes from an industry association rather than an auditor, Zeelin's cost advantage is a number that flatters Zeelin, and Amazon's programme is characterised by the executive who runs it. Against that, the two displaced-worker accounts cut against the vendor interest, and Digital Trends has no stake beyond the traffic.
Direction firm, magnitudes soft
That AI production has taken over Chinese short drama is supported from three angles and by practitioner testimony, so the thrust holds. The precise numbers are another matter: one relaying outlet, a hedge on the studio figures, no year on two of them, and every derived total — the implied human baseline, the monthly spend, the person-day ratio — inheriting whatever error sits in the originals.