By mid-June, Wu Xinhong had flown more than 230,000 kilometers in 2026, spending nearly 300 hours in the air. His travels took him across Asia, North America, Europe, and South America, with Brazil and Argentina as his farthest destinations.
The stops on his flight map trace the markets Meitu has expanded into over the past three years through its artificial intelligence products.
Meitu’s 2025 financial report showed that its overseas monthly active users returned to 100 million for the first time in years. Among its products, video editing tool Wink and image creation agent RoboNeo grew rapidly in Southeast Asia, Mexico, Brazil, and other markets, frequently topping iOS download charts.
Profits have risen alongside global expansion. In 2025, Meitu’s revenue reached RMB 3.858 billion (USD 569.2 million), while net profit rose 64.7% year-on-year to RMB 965 million (USD 142.4 million). Its artificial intelligence-powered imaging and design products have become its main revenue engine, with their share of revenue rising to 76.6% from 35% a year earlier.
A few years ago, Meitu was still hovering around the breakeven line. As AI surged, Wu told 36Kr that disruption had brought a sense of crisis, saying he felt as if he was “walking on thin ice.” Wu is Meitu’s founder, chairman, and CEO.
Over the past two years, Meitu has also made a series of organizational changes to respond to the AI era. Wu set up a group of small studios within the company, which has more than 2,000 employees, offering up to RMB 10 million (USD 1.5 million) in funding for individual teams to encourage rapid internal incubation of AI products.
The company also spent several months building a shared imaging platform and a shared growth platform. The goal was to reuse pipelines for technical engineering, initial product launches, paid traffic acquisition, and other functions across different products, shortening the product development and validation cycle.
“In the AI era, you have to move fast to have a chance of winning,” Wu told 36Kr.
To that end, Meitu has created a fairly strict product assessment system. From project approval and R&D to market validation and launch, the timeline is capped at one month. The standard for validating product-market fit (PMF) is that a product must reach USD 100,000 in annual recurring revenue (ARR) within six months of launch.
Wu has also proposed that legacy products with large user bases, such as the Meitu app, be barred from heavily directing traffic to new products. The idea is to test whether new products can grow organically.
Among the four new AI products released at the 2026 Meitu Multimedia Festival, portrait retouching tool Picchi and image workflow platform MeituHub were described as products that grew organically.
Picchi, for instance, came from Meitu’s insight into high-frequency user behavior on its namesake app. “Every time users retouch a photo, they are trying to make it look like their ideal selves. But it is troublesome to repeat the same retouching steps every time. Could users upload previously edited photos for AI to look at, so AI can automatically learn their retouching methods?” said Meitu chief product officer Chen Jianyi.
The other two products, which are more niche, came from Wu’s personal interests and his bet on non-consensus areas. They are music visual generation tool Mvland and concept video creation tool Artflo.
“Most successful products begin in scenarios where there is no competition,” Wu told 36Kr.
36Kr spoke with Wu and Chen about insights into emerging markets, competition, and a product methodology driven by passion.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: Has competition in imaging become more intense this year?
Wu Xinhong (WX): For a long time, imaging tools did not get much attention from venture capital investors, major companies, or entrepreneurs because they were hard to monetize.
Now imaging products are showing commercialization effects. Imaging covers a wide range of productivity scenarios, including images, videos, 3D, and more. Many industries have imaging needs. But imaging is also a highly fragmented sector, and as more players enter, it is hard for one company to dominate.
Right now, we are racing against time.
36Kr: Why were imaging products hard to monetize before?
WX: Before subscription models became widespread, many tool products could only rely on advertising. The problem with the advertising model is, first, that its monetization efficiency is not high, especially for brand advertising, because you need to build a professional team to serve clients. Second, ad placement and user experience are at odds with each other.
36Kr: Many years ago, you already wanted to make a product around music videos (MVs)?
WX: I studied art and learned to paint from elementary school, so I have always been interested in artistic things. What I have long thought about is how to use products to combine my interests with my long-term accumulation in imaging and visual work.
Besides Mvland, Artflo, which was also released this year, came from my personal interests too. It is hard to imagine another Chinese company making a product that turns inspiration into finished work, because on the surface, there does not seem to be a market.
36Kr: Can it make money?
WX: We do not know. But I just really like it.
36Kr: So Meitu is not a PMF- or data-first company?
WX: I hope that within Meitu’s product matrix, there is a private plot for me where I can do things I like. That is a bit of my own selfishness. Turning interests into a career makes things more interesting. I can do things that purely make money, too, but they would lack some passion.
We now have two ways of building products. One is bottom-up innovation. We identify a need or trend through data, do some validation, and then gradually scale the product. The other is a top-down strategic push.
The two are not contradictory. This helps create a healthy product ecosystem. Younger colleagues have opportunities to validate their ideas, while I can pursue my own interests.
Chen Jianyi (CJ): In fact, PMF-driven and passion-driven approaches are not contradictory. Mvland has performed the best among all new products over the past year. Within two or three months of internal testing, its ARR reached USD 100,000. It is now close to USD 500,000.
36Kr: Among the products and features released this time, which came from passion, and which were bottom-up?
WX: It is actually hard to separate them. But in most cases, bottom-up products have stronger staying power because they grow naturally rather than being forced.
For example, Kaipai, which has seen rapid revenue growth, is a bottom-up product. At first, we did not intend to make a talking-head video app. It branched out of BeautyCam’s teleprompter feature. At the time, we simply noticed that a lot of users were using that feature.
36Kr: When you pursue a product you are passionate about, but that has not been validated by data, does anyone on the team oppose it at the project approval stage?
WX: In the end, we let product data speak. If after multiple tests the product data is still poor, then we can only end the business. So we basically do not spend too much time arguing for or against something. Instead, we validate it quickly.
Meitu is a medium-sized company with about 2,000 people, so it has some scale advantages. We can put a lot of manpower into building shared product platforms and establish a reusable product development pipeline, which shortens the time needed to build and iterate products.
36Kr: How much can that shorten the timeline?
WX: For RoboNeo’s newly launched Agent Team, we completed it in one month.
In theory, an agent is a technically and engineering-intensive product. Because we have an imaging product platform, a growth platform, and other infrastructure, we can efficiently reuse mature experience in technical engineering, cold starts, marketing, and growth for new products or new features, enabling fast PMF validation.
In this era, the time needed to validate PMF should theoretically be as short as possible. No one knows how technology and the market will change one month from now. So we keep the product validation cycle within one month.
36Kr: What is your standard for validating PMF?
CJ: Within six months of launch, ARR must exceed USD 100,000. That is the baseline. After all, there are too many products in Silicon Valley with an ARR of USD 1 million.
36Kr: I heard you are about to fly to Brazil.
WX: Yes. We are taking the team there to experience the market and speak with local users and creators. We also plan to visit some Chinese companies that have done well overseas and learn from their experience.
36Kr: What are Meitu’s priority markets now?
WX: The priority market will always be China. China has the world’s largest user base and the most unified language environment, and Meitu also has a local advantage.
The Chinese market is very competitive, but that has its benefits. It trains the team’s combat capabilities and helps us compete overseas. After breaking out of local competition, when we go to other countries, we find that the intensity of competition is far lower than in China.
We began going overseas in 2013. Today, Meitu products’ market share in some Asian countries, especially in East Asia and Southeast Asia, is already approaching the level it has in China.
36Kr: Which markets do you most want to develop right now?
WX: Latin America. Brazil, Mexico, and other countries also have large absolute populations. Brazil, for instance, has a population of 210 million. People there have strong demand for social sharing, which gives rise to a series of imaging creation needs. For some products, such as RoboNeo and Airbrush, Brazil has the most users, surpassing China.
In the future, we also place great importance on the African market. The stereotype is that African users are peripheral and do not seem to have strong demand for imaging creation. But in fact, much of the user feedback we receive comes from Africa. So when there is an opportunity, I would also like to take the team there to understand the African market.
36Kr: Did you decide from day one that you wanted to pursue markets such as Latin America and Africa?
WX: No. We often launch the product first, then discover new markets and new growth opportunities based on user feedback.
Because imaging products, or internet products in general, have relatively low marginal costs, they can serve users around the world in a lightweight and fast way. So after we prepare language packs for different markets, we wait for users to download, use, and give feedback.
Gradually, we build a heat map that tells us which countries and regions the product is popular in. Then we keep looking for the reasons. If we can find a good fit between a product and a market, we try to amplify it.
So many things are not planned. They grow naturally. Once they grow, we water them and take better care of them so the products can become stronger.
36Kr: If markets are often understood only after the fact, what judgments can be made in advance?
CJ: Emotional judgment can certainly be made in advance. Mvland is Xinhong’s “small private interest,” so to speak.
But PMF cannot be validated in advance at the early stage. It has to rely on a subsequent value measurement system for commercial delivery. Take Mvland as an example. We conducted a lot of research with music studios, clarifying song output, MV production costs, and how many songs had MVs.
Many music studios said that if they were given a low-cost production method that allowed every song to have an MV, they could distribute content on platforms such as Douyin and YouTube, earn advertising revenue, or do more packaging for singer IP. These are business values that target users can specifically quantify, allowing us to reverse-engineer Mvland’s PMF.
36Kr: How do you find early-stage users? Are popular creators your core early-stage users?
CJ: We try not to include creators with huge followings as early-stage users. They are too capable across the board and are not typical users, so bringing them in would “contaminate” early-stage user behavior.
Their role is to help us reach more users in the early stage and support brand promotion.
36Kr: Which demands have you disproven so far?
CJ: AI-powered virtual try-on. Try-on has two scenarios. One is a commercial scenario for e-commerce platforms. Technology is currently the biggest bottleneck, and we are still tackling problems such as garment consistency and material fidelity.
The other is a consumer-facing scenario, helping users use AI for outfit styling. But we found that we had imagined users too ideally. People will not take photos of every outfit in their closet and upload them one by one just to do AI try-ons. They also enjoy the real-world process of trying on clothes and styling themselves.
36Kr: Rather than expecting AI to replace human labor, Meitu offers creative services for users who do not want to learn AI, allowing designers to take orders online. Could this hinder the formation of an AI-first mindset among users?
CJ: We borrowed from the premium service model used by Didi. We operate an in-house service team, provide professional training to designers, and deliver standardized services. At the moment, there are roughly several dozen people providing AI-supported creative services.
36Kr: Roughly what percentage of users need such services?
CJ: Looking at the world as a whole, at least 90%.
AI adoption is actually astonishingly low. Take the Chinese market as an example. Labor costs are very low, so users are unwilling to learn how to use some tools themselves. Instead, they directly purchase human services.
For example, a fruit plantation owner is thinking about growing good fruit, not learning how to edit videos or do short-video marketing. They would rather spend a few hundred RMB each month to hire a team to make three videos a day.
36Kr: From the Meitu app to today’s agents, what is the biggest change in product interaction interfaces?
CJ: Products like the Meitu app mostly start with editing, with conversation as support. Agentic products start with conversational features, then combine them with editors, asset libraries, and so on.
36Kr: Meitu is more of an application-layer company. Why does it still choose to train its own large model?
CJ: MiracleVision has made many optimizations for portraits. We have tested most models on the market, and none of them could be implemented for portraits or meet our requirements.
36Kr: Cost, quality, and generation speed form an impossible triangle for image generation models. As a downstream application company, will you have to sacrifice one of the three?
WX: Quality is the soul of imaging products. Even if we want to enter a promising scenario, if the product effect is not good enough, we will strategically give it up for the time being. Similarly, if we cannot judge whether we can deliver high-quality generation results, we will not enter the scenario.
Once we decide to enter a scenario, we will fight to the end and try to produce the best results.
36Kr: It is not uncommon for products to rely on low prices and subsidies to acquire users. Would you sacrifice some gross margin for growth?
CJ: Some startups are now burning money aggressively on compute subsidies. It is a lot like using investors’ money to help model providers inflate transaction volume, pushing ARR higher so they can raise more money. It is like walking a tightrope. Computing costs could fall in the future, and that risk is high.
It is like how the fastest way for an e-commerce company to increase gross merchandise value is to subsidize iPhones and gold bars. That playbook is only suitable for companies with tens of billions of RMB in ammunition behind them.
We will offer some subsidies for user acquisition, but we are more inclined to maintain a certain level of profit. Continuous subsidies cannot create product advantages. Product advantages are where our pricing power comes from.
36Kr: Meitu’s products start charging from day one. How do you decide the timing?
CJ: In the AI era, you have to charge from day one because tokens are expensive. If you make something free first and then start charging, many users will criticize you. Users who come in early because it is free may not be your real users anyway.
Doubao can be free first and charge later because ByteDance has money.
36Kr: Picchi and Kaipai were both incubated from features within legacy products such as the Meitu app and BeautyCam. Are they more vulnerable to disruption?
CJ: People who hold that view are mostly over 30. Why do young people today dislike buying luxury goods and prefer new streetwear brands? Because they like using that method to define how they are different from the previous generation.
Likewise in imaging, we also hope to create “fashion items” like Picchi.
The Meitu app is an integrated platform. Once we discover that a workflow under a specific scenario on the platform is used particularly often, we turn it into an independent product.
36Kr: You have mentioned that legacy products with large user bases, such as the Meitu app, will not direct traffic to new products. Why?
WX: We need to ensure that new products have the ability to grow organically rather than relying on legacy products to feed them traffic. If a new product is not effective and users do not stay, no matter how much traffic you give it, the traffic will leak away.
In the future, our main competitors will be startups. Their resources will certainly not be abundant, so they will desperately think of ways to grow. Therefore, we need to simulate a startup environment as much as possible.
36Kr: Why will your main competitors be startups rather than major companies?
WX: Major companies’ resources will flow toward more foundational chips and models. We are competing in the application layer. Every company has its own focus.
Even if major companies really enter the field, will the giants take everything? My guess is no. QuestMobile data shows that in China’s video creation tool market share ranking, CapCut is first, our Wink is second, and our Kaipai is fifth. That means we have broken through layers of blockades.
Moreover, markets and user needs are constantly changing. The Meitu team is good at thinking deeply in vertical niches and continuously polishing product forms that meet user expectations.
36Kr: Do you agree that AI has brought all players back to the same starting line?
WX: Not really.
Scale effects are very important. To be honest, for the same thing, investing 800 people and investing only two or three people are completely different. Behind a product, there is more engineering than algorithms.
Engineering requires real investment. It cannot be solved just by vibe coding and being smart enough. That is why we built shared product platforms, trying to turn the dirty and labor-intensive parts of engineering into reusable pipelines.
36Kr: Meitu is a company that has succeeded many times, and each of its products is in a highly competitive space. What have you learned?
WX: Most successful products begin in scenarios where there is no competition. For the four new products released this time, there are hardly any competitors in the market. We are only competing with ourselves.
Also, Meitu is best at running marathons. We have competed with Tencent’s Pitu and Baidu Motu many times before. If you persist, product quality will speak for itself.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Zhou Xinyu for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.78 = USD 1 based on estimates as of July 16, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.