A phone camera still smooths a pimple in a tenth of a second. But the app behind that tap has quietly become something else: a paid studio where small businesses generate product posters, creators cut talking-head videos, and designers draft ads without opening Photoshop.
Meitu (01357.HK) no longer pitches itself as a beauty-tool maker. It calls itself an "AI application" company built on images and video, and the numbers behind that shift are now public.
The model underneath the filters
In June 2023, Meitu released MiracleVision (奇想智能 / 奇想大模型), its self-developed visual large model (大模型). Where most image models optimize for raw realism, Meitu trained MiracleVision on aesthetics — the company's line is that a generated image should look designed, not just plausible. The model powers text-to-image, image-to-image, and image-to-video features across the product line, and it had already reached version 4.0 by December 2023, adding vector graphics, text effects, smart layout, and video generation.
The point was never to ship a chatbot. It was to embed generation inside tools people already open daily — MeituXiuXiu (美图秀秀), BeautyCam (美颜相机), Wink — and then charge for the output.
Subscriptions, not ads
Meitu reported its 2025 results in March 2026, and the headline was the subscription base. Paid subscription users reached 16.91 million at the end of 2025, up 34.1% year on year, with a subscription penetration of 6.1%.
Translate the top line: total revenue was about ¥3.86 billion (≈ US$544M / HK$4.24B), up 28.8%, and adjusted net profit attributable to owners was ¥965 million (≈ US$136M / HK$1.06B), up 64.7%. The engine is the image-and-design business — revenue of ¥2.95 billion (≈ US$416M / HK$3.24B), up 41.6%, now 76.6% of the whole company.
Founder and CEO Wu Xinhong (吴欣鸿) has described the strategy as avoiding a head-on fight with general-purpose large models and instead acting as a "model container" — plugging capable models into narrowly useful, paid workflows.
Two subscription tiers
Meitu splits users into life-scene and productivity buckets, and the productivity side is growing faster.
- Life-scene apps (MeituXiuXiu, BeautyCam, Wink) carried about 14.75 million paid users, up 30.3%. Wink alone passed 40 million global monthly actives.
- Productivity tools reached 2.16 million paid subscribers, up 67.4%, at a 9% penetration rate — the richer, faster-growing slice.
Inside that bucket: MeituXingtu (美图设计室) automates e-commerce artwork — product posters, detail pages — and has worked with Alibaba, JD, and Amazon storefronts; Kaipai (开拍) is an AI workflow for polished talking-head videos aimed at health, education, beauty, insurance, and real-estate verticals, with monthly actives nearly doubling in 2025; Vmake targets fitness and wellness in the US with annual recurring revenue around US$3 million.
The agent layer
In July 2025 Meitu shipped RoboNeo, an AI assistant that edits images, extends frames, and builds posters and brand designs from a prompt. It topped China's App Store category chart on launch and later led category charts in 15 countries including Brazil, Portugal, and Mexico. The move signals the next step: from single-purpose filters to an "AI team" the user directs in natural language.
Paying to keep the model working
None of this is cheap to run. Meitu's R&D spend in 2025 was ¥945 million, up 3.8%, and the company has begun opening its models to developers — a MeituCLI tool and a first batch of MeituAISkills were wired into the OpenClaw ecosystem. That developer push matters because the moat is not one model but the daily workflow: the more a shop or creator depends on Meitu for finished assets, the less likely a free general-purpose tool pries them away.
Why this matters outside China
Meitu's overseas monthly actives crossed 100 million in 2025, up 6.3%, and its global total reached 276 million. MeituXiuXiu has held the No. 1 spot in China's image-beautification category for ten straight years, and in 2025 it took App Store overall No. 1 in 52 countries and category No. 1 in 119. The overseas story is less about China-specific features and more about a subscription habit — European, North American, and East Asian users accounted for most new paid subs in the second half of 2025.
Honest limitations
The financial figures above come from Meitu's own 2025 annual results and founder commentary, which are authoritative for "what the company reports" but not independent proof of product quality. The "model container" label is a strategy statement, not a benchmark. MiracleVision's aesthetics-led training helps it win beauty and design use cases but does not make it a general competitor to frontier image models from OpenAI or Google. A 6.1% penetration rate means roughly 94% of users still do not pay, so the subscription ceiling — not the floor — is the open question. And like every consumer AI app, retention depends on features staying ahead of free alternatives.
What readers can do now
- If you run an e-commerce store, test MeituXingtu (美图设计室) on a real product-poster batch and compare turnaround against a designer or a general image model.
- If you make short videos, try Kaipai (开拍) for a talking-head clip and note where AI scripting and editing save time versus where you still reshoot.
- If you track the China AI market, watch whether Meitu's productivity-tier penetration (9% and climbing) keeps outpacing the life-scene tier — that gap is the real signal of whether "AI studio" subscriptions scale.