An AI-made Chinese short drama pulled 4 billion views in two months, and its studio calls the commercial return roughly ten times the compute spend. Founders read that as a cheap customer-acquisition channel. Run the three ledgers first — production, media buying, content. Most people only ever count the first.
Several e-commerce founders asked us the same thing this month: short drama is exploding — even arthouse directors and novelists are hooked. The viral story: a small team with no film background used AI to produce a drama that hit 4 billion views in two months and returned roughly $1.4M. The whispers: AI comic drama at $113 a minute, a full AI title at $14k. Small budget, one blockbuster — why not.
Live-action short drama runs $70k–$140k per title at comparable heat. AI titles: $14k–$28k. AI comic drama: $113–$170 per minute, about a fifth of live-action. Directors, crews, locations, shooting execution — that money is really saved.
In this business production is only 10–30% of total cost. The remaining 70–90% is media buying: paying, batch by batch, to bring viewers in. The head of the micro-drama association put it on record via Beijing News: media buying can lose money — but not buying means no viewers at all, so you buy at a loss.
Media buying usually eats 70–90% of recharge revenue. A title that crossed $14M in recharges spent $8.5M–$11M on ads; after production and splits, what lands is a few hundred thousand dollars. That is the winner's ledger.
Supply makes it worse. In the first half of this year, Douyin's charts logged 220,000+ new AI drama and comic-drama titles; roughly 2,900 reached break-even — 1.3%, one in 77. In Q1 alone the industry launched 128,000 micro-dramas, 95%+ of them AI-made. And Honggu, the biggest distribution platform, has been trending for "eating user time," having stopped new-user coin subsidies on November 1. Platform subsidies are the platform's money — never book them as your revenue.
At this ledger people still think in savings and reach: the story is a formula — heiress feuds, security-guard comebacks, cleaner comebacks; the traffic codes are off the shelf. Copy a popular character or script and you are buried in a river of near-identical products.
This is not a scare figure: last September, 521 titles carried the word "comeback." In front of your cleaning lady stand five hundred identical aunties. Even if you pay for reach — same face, same plot, why would viewers remember you?
One drama did run out of that river — the 4-billion-view one. Chi Xie Ru Song (持械入宋), top-ten on Honggu's AI drama chart: ten seasons, about a thousand episodes in two months, compute spend above $140k, commercial return around $1.4M. Sit with that number: $140k of compute already pushes halfway out of "low cost."
It won on craft, not a hot script. Ordinary shots were regenerated about three times; complex battle shots ten, twenty. They researched the color of official uniforms and how period travel permits were written. The hero's gun runs out of ammunition; the gasoline runs out; knowing history's direction, they still couldn't change it. A team that had never made film before ground the details to that level — that is what 4 billion views cost.
So the real function of AI here: an amplifier. An amplifier can only amplify the signal you already have — it cannot invent one. A persona that holds up, a story you can tell in one evening, content that connects to your product: AI multiplies it by ten. Have none of those, and it amplifies blank. SK-II ran "rewrite destiny" for ten years into brand equity; Disney grew theme parks and licensing out of film IP. Brand life works this way: what was already there, amplified again and again.
Po Shi — a dry-shampoo brand doing about $280M a year — cut its social content to one scene: a woman fresh off the treadmill, hair collapsing; does the spray work or not? The picture speaks. Content runs on testing: the model's hair, the first three seconds, when the product enters — one variable at a time, until the winning cut reached an ROI of 8–10.
No. Production cost is genuinely lower; acquisition cost is not. What you save is shooting-execution money. The money to acquire customers hasn't dropped by a cent, and the logic hasn't changed at all.
Two conditions before you try. One: you can state the person and the story — who they are, what happens, how it links to your product — in one evening, without copying. Two: the budget reserves a line for media buying and a head for content; don't sink everything into production. Both — test it. One — hold, and keep the money. Short on budget but committed: go to a team with actual hits on a fixed-KPI custom deal — you bring product and budget, and lock the targets. One wash-care brand did exactly that: 350M views, 10M+ new users, the brand funded production.
The wealth story is real. The 4 billion is real. Only — "low cost" was never in that ledger.
Before you budget a drama, map the signal: the person, the story, the product link. That audit is the part we run first.
Start a conversation →