Meitu now has 19.83 million paying users, and that says more than any benchmark

On November 4, Meitu made Baijing Chuhai's "2026 influential Chinese brands going global" list. The award means little: yearly, and the list is long. The numbers attached to it are the interesting part. By September 2026 Meitu had 19.83 million paying subscribers, a net gain of about 1.39 million in a single quarter, the best of the past four. Paid growth overseas has also outpaced mainland China for several quarters running.
A company founded in 2008 as a photo-retouching tool is still accelerating its paid base through the two hardest years AI image generation has seen. That deserves more thought than any leaderboard.
Four in ten seats on the top 100
Chinese developers hold roughly 40 of the seats on a16z's Top 100 Consumer Gen AI Apps. Meitu alone placed five: Meitu, BeautyCam, BeautyPlus, Airbrush, Wink. ByteDance's CapCut reaches 740 million monthly users, ranks inside the global top 30 as the highest-traffic Chinese AI export, and booked about $220 million in the latest month. Hypic and Jimeng sit on the same list, both in the tens of millions of MAU.
Meta's free editor Edits is worth a look. Over 48 million monthly users, a serious audience, yet neither scale nor revenue has moved CapCut.
Less visible names matter too. PixVerse at No. 40 comes from Beijing's AIsphere, founded by a former ByteDance visual tech lead who built Douyin and TikTok's visual stack from zero. The company is valued above $1 billion and reportedly plans a Hong Kong IPO as early as this year. AI Mirror at No. 54 belongs to Polyverse, whose founding team includes a former ByteDance strategic partnerships director; it started in casual game publishing overseas before moving into AIGC.
Chengdu's Haiyi Zhichuang is another sample. Its SeaArt platform has over 65 million registered users and 30 million monthly visits across 200-plus countries and regions, and closed a Series B above 100 million yuan. More than 2 million AI creation SKUs sit on the platform, covering models, workflows and template apps, and over 90% come from users. The team shipped 48 releases in a year, one every eight days on average, with R&D at 30%-plus of revenue. After launching ready-to-use creation apps in late 2024, MAU jumped from a few million to 12 million.
Haiyi CEO Ma Fei gets quoted for one line: "one year in AI, ten years in the human world." It sounds like a slogan, but eight releases a month is exactly that pace.
Same cohort, two price lists
If you look for what these exports share, "best model" is unlikely to be it. Most have no foundation model of their own. SeaArt even argued the point internally and decided against betting everything on a base model, putting resources into content and community instead. What they got right was turning image generation into ready buttons: pick a template, swap assets, apply a workflow. Users never need to know what CFG means.
Put another way, they sell the process. The model is one segment of it.

Bring the same thinking home and the other half of the bill appears. Jimeng has rewritten its credit rules several times in recent months. One short-video creator ran the full math: a 15-second clip cost about 0.65 yuan before the changes; after several rounds of repricing and paid queue-jumping, it can reach 5.03 yuan. A 2-minute video went from 5.2 yuan to 40.2 yuan, close to eight times. Membership perks shrank too: 15,000 credits a month cut to 6,160.
The reason is simple. On day one of Seedance 2.0's full release, queues hit 8 to 10 hours, and some people waited 24. Video generation burns money: a few seconds can eat as much GPU time as a chat model answering a thousand hard questions.
ByteDance moved the opposite way on the enterprise side. Volcengine Ark put Seedance 2.0 Mini up on June 15 at 0.023 yuan per thousand tokens for image-to-video and 0.014 for video-to-video, roughly 0.5 yuan per second at 720P, half the standard edition. Raise consumer prices to filter for willingness to pay, cut enterprise prices to buy call volume. Two hands, two jobs.
The price hike lands on the bottom of the chain
The real cost falls downstream. DataEye found 98.7% of AI short-drama projects failed to recoup costs in the first half. Douyin put 221,900 new AI dramas live in the same period; 1,055 passed 100 million views, under five in a thousand. Production houses taking work fell from 1,216 to 698 in a quarter. Hongguo paid 30 to 100 yuan per 10,000 views last year; now it pays 5 to 10.
So you get an awkward loop. Tools push per-episode cost from 10,000 or 20,000 yuan down to just over a thousand, then creators spend another 5,000 yuan on traffic so anyone sees it. Traffic bidding pushes one price up, compute pushes the other, and both sit in the same ecosystem. The savings get clawed back a bit at a time.
Back to Meitu's 1.39 million new subscribers. They are a different crowd from Jimeng's. One group buys certainty, the picture gets finished, and pays monthly for templates, assets and team collaboration. The other buys more rolls of the dice, and when each roll gets dearer, they drift away.
The export cohort keeps growing on the first kind of demand. That demand is unglamorous, and it renews every month.
Related articles
The Jensen Discount: What Nvidia Loses When Its Founder Walks Out
Nvidia lost roughly $600 billion in market value when its founder stepped down. The earnings did not change. The multiple did, and that says where the value really lived.
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.