Two AI Cats, 3.15 Billion Views, and 400 Yuan a Video

"If a kitten posts the ad, I'll watch it. If a real person posts it, I scroll right past."
That line sits in the comments under 橘猫胖墩墩, and it has drawn a lot of likes. The account stars two orange tabbies who live like flatmates — shopping, cooking, taking selfies, every gesture and expression uncannily human. They aren't real cats. Around 350,000 followers across platforms, sponsored slots booked into November, and the owner tells reporters that revenue passed costs a long time ago.
On Douyin, #AI萌宠 has drawn 159,000 participants and 3.15 billion cumulative views. At that scale it stops being a "fun new trick" and becomes a full production line: models upstream, accounts in the middle, advertisers downstream.
First, get the math straight
The owner of 橘猫胖墩墩 offered a figure few people in this business volunteer: 400 to 500 yuan to generate a single video. That doesn't sound cheap. AI video runs roughly 1 yuan per second, so a clip of a few dozen seconds already burns over a hundred yuan in compute — and the same shot usually needs three or four do-overs. On a bad day, seven or eight.
Time is the harsher cost. Working part-time, one video takes two or three days. What actually eats the hours isn't pressing generate, it's writing the plot: why the cat was late to work today, why it slacked off in the bathroom, why it stopped for a milk tea on the way home. Not one of those details came out of a model.

The revenue side is cleanly tiered. The top account, LT小狗日记, quotes up to 80,000 yuan for a single custom ad, according to 36Kr. A mid-tier Xiaohongshu account, 川岛小猪, charges about 4,500 yuan a video and earns mainly commission from selling through its product showcase. So a 400-yuan video selling for 4,500 is ordinary business. Selling for 80,000 is the rare exception.
Someone will immediately compute a 90% gross margin. That arithmetic skips two things. One is throughput: ten videos a month part-time already counts as industrious. The other is the dud rate: get two out of ten to perform and you're doing well. Plug those numbers in and a few tens of thousands a month is plausible. A few hundred thousand belongs only to the handful of accounts that have already built a persona.
Why cats, and not something else
Pet content starts with an installed base — a 300-billion-yuan pet economy is sitting right there. What AI changed isn't whether people watch cats. It changed three very practical things.
First, no shooting constraints. Real cats don't take direction; getting three usable shots in a day is a fight. An AI cat hits its marks, its persona never breaks, and it never gets sick.
Second, sets cost nothing. Office, subway, wet market, midnight convenience store — all the set-building budget disappears.
Third, you can test in batches. One character, ten plots, and whichever takes off is the one you continue. Real filming can't match that cadence.
But what keeps people watching to the end isn't the cat. It's the resemblance. One account logs a kitten's working day: alarm off in the morning, squeezed onto the subway, coffee at the desk, short videos in the bathroom stall, food street after work, asleep late at night. The comments are all "I cried" and "so comforting." Viewers aren't watching a cat's day. They're watching their own.
Platform data hides another thread
Xianyu published a report in the first half of this year: 9.816 million AI service orders, up 157% year on year, with nearly 5 million people having bought AI services on the platform, up 98%. The two fastest-growing categories were AI coding and site building (+1732%) and AI comic drama (+1425%).
Those numbers say "AI helps ordinary people make money" has stopped being a slogan and become a marketplace. Someone pays 50 yuan for an avatar. Someone pays 500 yuan for a 30-second short. Pet videos are simply the best-exposed cell in that market.
The real barrier in this line of work
The tools are dirt cheap now. DeepSeek writes the script, Jimeng or Kling produces the images and video, CapCut assembles it — a practiced hand turns one out in fifteen minutes. Technology is not the moat. Three other things are.
One, character design. Audiences follow "this cat," not "a cat made by AI." The character needs a stable temperament, verbal tics, a daily trajectory — and it has to survive a switch of model.
Two, emotional design. Comfort content moves slowly. No twists, no payoff beats; the whole thing works by laying an ordinary person's day out flat. That ability has nothing to do with whether you can use AI. It has everything to do with whether you notice people.
Three, serial stamina. A single viral hit can be luck. A series means coming up with five non-repeating daily moments every single week. Creative burnout is the most common way people die in this business.
Homogenisation has already set in. Hundreds of accounts are running the same template, and viewers swipe away by the third one. Nor will platforms keep feeding traffic to one kind of content forever.
Three sentences for anyone thinking of getting in
First, don't treat it as an "AI project." It's a content project. AI only brings the cost down from tens of thousands to a few hundred; topic and character are still human work.
Second, do four weeks first, three videos a week, and see whether you can produce steadily. If you can't last four weeks, no amount of money will help.
Third, put the duds into your math. 400 yuan is the cost of one video, not the cost of one hit.
That cat earns money not because AI generated it, but because someone wrote a story onto it that reminds people of themselves. Tools will keep getting cheaper. Stories won't.
Related articles
Chinese text in video always turns to mush. Dreamina handed the typesetting to code
Dreamina launched Motion Studio on November 10. It no longer generates frames; the model writes code that lays out text, shapes and pacing. Chinese titles stop blurring, but whether the content is right is still on you.
A community spreadsheet of taps that turns off the ads inside Chinese Android builds
A no-root tool that switches off the ads Chinese phone makers ship inside the system UI. Its method is a hand-written tap script per phone model, maintained by volunteers every time a ROM moves. It works, and the maintenance is the point.
GPT Image 2.6 held the token price and raised the bill
GPT Image 2.6 kept every token rate from 2.5 and still costs more per image, because more of the work now happens in output tokens. Eleven days into the ramp, the pricing structure says more than the leaderboard does.
Someone rebuilt IDM from the outside and wrote down what they did not touch
A Rust reimplementation of Internet Download Manager claims 5x the speed and 86 languages. The more interesting document is the paragraph explaining how it was built without copying anything.