AI Girl Groups Replace Commercial KTV: 2026 Is the First Year of AI Talent Shows, and the Soft Spot Behind a 41 Billion Market
2026 is being called the first year of AI talent shows. AI girl groups such as HeyDream, Twelve Girls, and Fang Taozi have gone viral in quick succession. China's AI virtual idol industry chain has surpassed 41 billion yuan in scale, with a core market of 18.26 billion yuan and a year-on-year growth rate of 37.4%. The South Korean AI girl group BABYQUEEN suspended activities on the grounds that member Mity was "pregnant," sparking skepticism; the operator subsequently apologized, exposing the soft spot of AI idols: they lack a "human touch" and struggle to make fans empathize. In 2026, the Measures for the Supervision and Administration of Live-Streaming E-Commerce have brought AI-generated character images and videos under regulation, and live streams using AI avatars must be continuously labeled.
Topics
Body · AI translation
The commercial KTVs that once made easy money have finally gone cold in 2026.
In September 2026, Hangzhou was the first to crack down hard. A batch of leading commercial KTVs—Skyline, IN11, Manhattan, Dongfanghui and others—were sealed off one after another, some had their licenses revoked, and authorities even retroactively audited three years of transaction records.
In fact, before the commercial KTV flame went out, many top tippers had already quietly moved their spending elsewhere.
Since the start of this year, HeyDream, Twelve Concubines, Fang Taozi and AI talent-show girl groups have gone viral in quick succession, and in October the authorities explicitly named "virtual digital human companionship" as a new business format to be cultivated.
Previously this flame was confined to niche fan circles. What truly brought it into the mainstream was a recent piece of news: a Korean AI girl group called BABYQUEEN issued an announcement saying that member Mity was pregnant and would suspend all stage performances, variety shows and group schedules. The announcement was even written in a more official tone than a real agency's—"after full discussion," "we earnestly ask fans for their understanding and blessings."
Mity's own account then posted a nearly 500-word long letter, every word and sentence modeled on the template of a real idol leaving a group, and an hour later the team's Chinese account released a comeback teaser for a new song.
Netizens, however, were not buying it, questioning, "Is she giving birth to a robot?" The incident quickly fermented and trended, and the operator soon came out to apologize. It turned out that behind it was a fan project run by a few girls who love K-pop. Because raising several AI characters at once was too tiring, they wanted to take Mity offline to lighten the load, so they chose the absurd reason of "pregnancy."
From big tech to grassroots players, all have their eyes on this tens-of-billions cake
2026 is also being called the inaugural year of AI talent shows.
It started with variety shows: "12-Star Training Competition," "Starlight 108," "AI Idol Debut Season"... a slew of AI talent-show projects launched in quick succession. Behind this, big tech's business instincts were sharp, and MCNs and entertainment companies rushed in one after another.
Data from CCIDnet shows that in 2026 the overall scale of China's AI virtual idol industry chain exceeded 41 billion yuan, of which the core market was 18.26 billion yuan, up 37.4% year on year, and users aged 35 and above already accounted for 22.3% of virtual idol users.
Capital, meanwhile, was not idle either. The number of MCN agencies also surged in tandem, with more than 500 new institutions squeezing into the virtual idol field within a year.
Generating images is basic operation for AI tools on the market. Write a prompt with Jimeng, Ketu or MidJourney and an image is generated quickly; then drop the image into platforms like Kling, Runway and Pika, add beat-synced cuts, cheering sound effects and fan-comment phrasing, and a video is ready.
Soon a bunch of AI girl group accounts suddenly popped up on Douyin, Xiaohongshu and Bilibili. Five-member formations, Korean-style stage names, nine-grid ID photos, practice-room selfies, member interactions... the full set copying the ritual sense of the Korean entertainment industry.
Generating images is only the first step. Next comes roughly the same rhythm as with real people: setting up team members, with Chinese, Korean and English names all in place, imitating the preheating rhythm of a rookie group.
AI idols' selling point is exactly the reverse: no private life, no scandalous downfall—an ideal asset for brands. For example, an AI short-drama heroine like Fang Taozi can command 258,000 yuan for a 60-second ad.
A real idol practices for months for one stage performance, gets out of breath after three songs in a row at a concert, and looks green in the face by the last stop of a tour. An AI girl group is different: once the model is rendered, in theory it can dance 180 times an hour, without a day off all year.
Yet the players miscalculated one step. Starting with BABYQUEEN's pregnancy departure, AI's soft underbelly has become increasingly clear:
Being able to dance doesn't mean being able to become popular; not getting tired is instead its most fatal weakness.
Dancing 180 times an hour, behind it lies the common disease of commercial KTVs
A real idol's appeal has never been entirely in the face and the voice. What fans chase is the trace of a person growing up—for example, the clumsy fancams from trainee days. These things are moving because they carry the warmth only real people have.
Obviously AI girl groups don't have this ability. Its every movement comes from training sets and prompts. No matter how well it dances, fans know it can loop infinitely and doesn't need rest. The audience can't find that sense of immersion of accompanying someone along the way—so perfect that no one can empathize.
Especially in the idol industry, the more perfect and never-scandalous it is, the less fans can develop emotions.
Lacking "human flavor," not getting tired is instead its most fatal flaw.
Most virtual girl groups now are either motion-capture driven, with a real person behind them dancing in a suit, or purely algorithm-generated, with everything from the person to the background fake.
But the problems with AI-made content are obvious: smooth positioning eats into the sharpness of movements; when body proportions don't match the model, clipping and de-rendering occur; and especially with movements like rapid head turns, sudden stops and chest rises, the body outline often gets smoothed flat. As soon as such problems appear, the top tipper is immediately pulled out of the illusion.
After all, virtual humans have no real experience. If they say in first person, "I've tried these colored contacts and the experience was great," at best it flops and gets taken down; at worst it violates advertising law and portrait rights.
So many projects get built fast and fall even faster:
At first a generated video blows up and followers grow to tens of thousands; then they find daily updates can't cover computing costs, the top tippers won't pay up, and commercial orders don't come in—in the end they can only quietly exit.
From capital's perspective, the former is obviously still more "attractive." Because pure AI group livestreams don't need real people on camera, can hang online 24 hours to collect gifts, and carry no risks of contract termination or streamers' personal scandals bringing them down—the logic sounds very sexy.
But in reality, if girl groups and boy groups truly want to last, especially to take away commercial KTV's users, it is by no means as simple as replacing people with AI.
What laypeople see—modeling, voice timbre, image generation—is only the basic threshold. What really burns money is the subsequent operations and maintenance, such as real-time rendering computing power and motion-capture equipment, fan group operations and so on. Cultivating a truly valuable AI girl group is actually not much easier than with real people.
Whether short dramas or girl groups, the essence is still the content industry, and competition is brutal. Dozens of girl groups debut every year, each with five to seven members, plus those returning from overseas—the market simply cannot absorb them all. Before, only a few groups competed for one stage; now hundreds of groups compete for the same stage.
The audience is only so big, platform resources are increasingly concentrated, the top groups take 80% of exposure, and the rest can't even make a splash.
Take the blogger "Tao Agoujun," who went viral with unique creativity, turning ordinary outfit and travel subjects into imaginative creations like comic-girl freeze-frames and phone touchscreen outfit changes. Netizens say, "One look and you know she shot it."
Girl groups are still a good business. Top players are still eating meat, and their money-making potential should not be underestimated. Data shows the fan economy is about 1.23 trillion yuan, and the combined merchandise revenue of the four major agencies in 2025 is estimated at about $680 million. For AI girl groups to enter the game, they must produce good works and find more solid monetization methods to truly get a slice of the cake.
Now the authorities are also stepping in. The 2026 "Measures for the Supervision and Administration of Livestream E-commerce" have already brought AI-generated character images and videos under regulation: livestreaming with AI figures must be continuously labeled and must not impersonate real people. Once content causes trouble, it is not AI that takes the blame—the livestream room operator bears full responsibility, and penalties are even harsher than for real people.
Those who do not respect the market will sooner or later be hit hard by the market. Just as BABYQUEEN's flop has sounded the alarm:
Images are from the internet; please contact us for deletion if there is any infringement.
This article is from the WeChat public account "Jincuodao" (ID: ijincuodao), author: Jiang Yuan, published by 36Kr with authorization.