Essay

221,900 AI Dramas in Six Months — and Fewer Than 1.3% Broke Even

AI collapsed the cost of making a short drama. It didn't collapse the cost of getting watched. The bottleneck moved from production to attention — and the flood is just starting.

In the first half of 2026, about 221,900 new AI-generated drama and animated-drama titles were released on Douyin, China’s version of TikTok. That is DataEye’s count, reported by The Paper. Read it again: two hundred twenty-one thousand nine hundred new series in six months. Roughly 1,200 a day.

And fewer than 1.3% of them broke even.

That second number is The Paper’s calculation, using 50 million plays as a break-even line. It is a proxy, not a profit measure — the outlet doesn’t say who set the line or on which costs, and it covers Douyin only. But the shape of the story is unmistakable: AI made it nearly free to produce a short drama, and almost nobody is watching most of them.

What collapsed, and what didn’t

The cost collapse is real. The Paper puts a premium AI series of 100 minutes at about 100,000 yuan, against 400,000 to 500,000 yuan or more for a series shot with real actors, and around a million for a premium one. A working producer who tried it himself — Ludovic Bostral, writing in October — found a minute of finished AI video costing $26, then $18, $18, and $14 across four episodes at providers’ public prices. Hunan TV, a Chinese provincial channel, aired a drama generated with ByteDance’s Seedance model, with no cameras at all: six months from idea to broadcast, three of them for production.

What did not collapse is everything else. Bostral’s October episodes still cost real time in the places AI is weakest: keeping characters consistent from shot to shot, and sound. In his first episode, the animation model walked the cat on three paws for five full seconds. A model left without a sample voice to follow started speaking Mandarin. His tool list changed five or six times in two weeks. Hours of human work are not in the per-minute figures — and no audience was measured.

And the audience is the whole game. Over the Chinese New Year 2026 period, series shot with real actors made about 1/50 of the volume of AI series — and drew 25 times their views, according to the China Netcasting Services Association, the industry body for online audiovisual services. Production values still appear to count. Cheap supply did not create cheap demand.

The bottleneck moved

This is the part that matters for everyone who isn’t a producer. When making a show costs almost nothing, making a show stops being the hard part. The hard part becomes getting anyone to watch it.

The industry already knew this in miniature. Inside the big apps, the chain is a loop: spot a story that works, adapt it, produce it in days, release it, read the numbers, drop what fails, rebuild what almost worked. ReelShort’s parent company tracks its reading app’s most-read stories for adaptable material; a 60-episode season takes about a week to shoot; three days of data decide whether a series lives. AI just made the loop spin faster and cheaper — which means the selection step, the part where viewers vote with their minutes, matters more than ever.

For viewers, the flood has a specific consequence: the number of shows competing for your first three free episodes is now effectively infinite, while your attention is not. The scarce resource is no longer content. It is curation — knowing what is worth your coins, what was actually filmed, and what is a synthetic imitation.

That is visible in our own library. Of the 279 dramas we currently cover, 49 are labeled AI-generated, 28 live-action, 18 mixed, and 184 unlabeled — unlabeled because the platform doesn’t say and we don’t guess. On ReelShort, whose drama pages directly answer “Is this AI-generated?”, the picture is clear: 46 of 78 titles are AI-generated and 27 are live-action with a credited cast. On DramaBox, which offers no equivalent disclosure, all 33 titles we track are unlabeled. The disclosure gap is itself a finding: the platform that tells you what you’re watching is the exception, not the rule.

What survives a flood

History has a pattern for this. When a production cost collapses — photography, desktop publishing, streaming video — the result is never “everything gets better.” It is “almost everything gets ignored, and the few things that break through do it on something other than production cost.” The winners will be decided by story, by stars (real or synthetic), by marketing, and by the recommendation systems that sit between 221,900 titles and one viewer.

There is also a second-order effect worth watching. As AI supply floods in, the genres that were expensive to film — fantasy, cultivation, supernatural fiction, science fiction — become cheap to attempt. A story that previously made economic sense only as a web novel can now be adapted into video. That is genuinely new: not more of the same, but kinds of shows that could not exist at filmed prices. Whether audiences want them is a separate question, and the 1.3% figure suggests the answer is usually no — but “usually” leaves room for a lot of shows.

The flood is not going to recede. The tools get cheaper every quarter, and the cost of a minute of finished video is still falling. The question for the next year is not whether AI will make more short dramas. It is whether anyone — platforms, critics, guides, or the recommendation algorithms themselves — gets good enough at telling you which ones deserve your evening. Right now, the machines are much better at making shows than at picking them. That gap is where the whole next phase of this industry will be decided.

Sources

Quick answers

How many AI-generated dramas are being made?

About 221,900 new AI-generated and animated-drama titles were released on Douyin in the first half of 2026 — roughly 1,200 a day, per DataEye's count reported by The Paper.

Do any of them make money?

Fewer than 1.3% broke even, per The Paper's calculation using 50 million plays as a break-even proxy. AI collapsed the cost of production, but it did not collapse the cost of getting watched.

Is AI production really cheaper than filming?

For production, yes: about 100,000 yuan for a 100-minute premium AI series versus 400,000 to 500,000 yuan or more shot with real actors. But the savings stop there — character consistency, sound, and hours of human fixing work are not in the per-minute figures, and no audience was measured.

What happens when the cost of making a show collapses?

The bottleneck moves from production to attention. Over Chinese New Year 2026, real-actor series were 1/50 of AI series' volume but drew 25 times their views. The scarce resource is no longer content but curation — and the recommendation algorithms sitting between 221,900 titles and one viewer.