Surviving 12 Years in the Cracks: How He Finally Built China's Most Active AI Product | WAVES

By Jiaxiang Shi

Edited by Jing Liu

Many assume AI is a young person's game, which makes it hard to imagine a hit AI product with tens of millions of monthly active users coming from a founder in his forties. That founder is Jiang Duan, and the product is Fotor. Fotor isn't based in Beijing, Shanghai, Guangzhou, or Shenzhen — nor in Silicon Valley. It sits in a quiet corner of southwest China: Chengdu.

According to AI Product Rankings, among more than 5,000 AI applications worldwide, Fotor has long ranked near the top of China's AI app export list by user count, and was once the only domestic AI product with over ten million monthly active users.

Fotor is not a "native AI" company riding the current wave. Before breaking out, few even knew it was Chinese. Its origins trace back twelve years to the mobile internet era. At first, it was simply an image editing app targeting the same market as Meitu.

Twelve years later, generative AI reset the playing field for everyone. Fotor, which had been toiling overseas all along, saw Midjourney explode and trained its own text-to-image tool in just two months — filling the gaps in Midjourney's user experience.

When Fotor finally surfaced, it had truly made the leap from zero to one: user scale multiplied sevenfold, and it achieved profitability at scale. Duan says the most agonizing part of entrepreneurship is how slow it feels — most people can't see the end and choose to leave.

Simple luck can't explain everything Fotor did, because forced moves and missed opportunities defined its entire journey —

Once, Meitu raised dozens of times more capital than Fotor, so Duan decided to avoid direct confrontation, abandon the domestic market, pivot overseas, and focus on the PC platform where customer acquisition costs were lower;

Seven years ago, as its first round of funding ran dry, Fotor was rejected by nearly every investor it approached. It had no choice but to end its free tier and enter the paid phase early;

When the second wave of image transformation arrived, Fotor belatedly followed Canva by adding graphic layout features, but the traffic dividend had already passed. Given headcount and resource constraints, this business line was suspended around 2020;

Even three months before its user base exploded, the company was being pressured by investors to buy back shares, nearly wiping out all its cash on hand.

The contrarian path taken a decade ago caused Fotor to miss the mobile internet boom, but it also made Fotor one of China's rare AI companies today — profitable, at scale, and sustainable.

This is an atypical startup story. All it can teach us are the most朴素的创业箴言: stay at the table, and when opportunity comes, seize it at all costs.

"WAVES" is a column by Anyong. Here, we bring you the stories and spirit of a new generation of entrepreneurs and investors.

The Victory of Contrarianism

When tens of millions of users flocked to Midjourney but were forced to use it through Discord, Fotor founder Jiang Duan had just one thought: something this good should work by simply typing on a website — why make it so complicated?

Then Duan realized the moment that would change Fotor's fate had arrived. He paused all company projects and told the team this was a massive opportunity. "It's not just image generation — the entire image space is about to be transformed," he said. They had to move immediately in this direction.

This wasn't an easy choice. It was August 2022. Not long before, Duan had traveled to Beijing to meet investors. The pandemic had emptied the train stations of all life.

Investor confidence had been similarly hollowed out — Fotor's investors wanted an early buyback, even though the contract wasn't due until end of 2023. In exchange, they'd accept a discount. Duan did everything he could, giving investors nearly all the company's cash to almost fully repay them.

The investors who retreated before dawn never lived to see Fotor's "ChatGPT moment."

At this point, Fotor's books were nearly at zero. They had to fight with their backs against the wall — no retreat possible.

Duan restructured the team. Building on open-source Stable Diffusion, they used Fotor's long-accumulated high-quality image data and deployed every GPU the company owned. In two months, they launched a Midjourney-like text-to-image feature on their website.

When hordes of C-end users flooded in from Google keyword searches, Duan knew his bet was right — Fotor's traffic multiplied sevenfold.

Contrary to outside assumptions, Fotor isn't "young" like Midjourney. This was its tenth year.

At birth, Fotor positioned itself as "lightweight Photoshop." In 2012, it raised $3 million from Legend Capital and pushed into mobile.

Back then, mobile internet operated on infinite games — burn cash for traffic. Fotor didn't have much capital. Duan knew he couldn't compete with rivals like Meitu, which had completed multiple funding rounds.

So Duan turned his gaze to the uncharted territory overseas. "It was like feeling an elephant in the dark," he says — no reference points, almost entirely self-taught. The only advantage: the team had overseas backgrounds, so no language barrier. They also shifted product focus from the traffic-driven mobile end to the less competitive, SEO-dependent PC platform.

There was no shortage of opposition within the team. Investors also wanted them to grab a slice of the domestic market. But Duan held firm. He admits that abandoning China's still-rich mobile internet and choosing overseas and PC were both forced moves. "Our situation at the time left us no other choice." The endless waiting without seeing an end caused many team members to leave.

In 2017, Fotor went from free to paid — a reluctant move after its first round of funding ran out. This was an era of abundant capital, yet not a single venture firm would back them. "We'll have to generate our own blood," Duan told the team.

At the time, Fotor's monthly active users had already crossed one million. Overseas paid conversion rates were higher than the team expected. Seeing the possibility of profitability, Fotor also secured 25 million RMB in Series B funding from GF Securities.

And precisely because of more than a decade of investment in image editing, we see the result today: when the once-maligned choices of PC and overseas became new trends, this "non-mainstream" Chengdu company had already been running hard in the lane for ten years.

To this day, Fotor has invested the most energy in English-language markets — specifically, the United States, United Kingdom, Canada, Australia, and New Zealand. "Although they only account for 40% of overseas user traffic, they contribute 80% of our current revenue," Duan reveals.

Back to the Comfort Zone

After the traffic surge, investors swarmed in. But the team ultimately decided to reject over 100 prospective backers. "When we had investment before, we didn't do that well. When we had no investment, we made it work. We still feel this money shouldn't be taken," Duan says.

The confidence to refuse came from Fotor having accumulated substantial cash, with monthly profits still coming in. Another reason for rejection: "Fotor hasn't found a direction worth investing in at scale," Duan says. Headcount only grew from around 80 at year-end to roughly 120.

In fact, a worthy direction had once existed. In 2023, as the "Midjourney" story continued, Fotor followed the "industry chain" into video generation. Initially, they poured the most energy into training video generation models.

The turning point was Sora's emergence. Even as "vaporware," plus the explosion of more AI video generation companies like Kuaishou's Kling, made Duan realize that without sufficient data and resources, their odds of competing weren't strong — Duan was reminded of the helplessness of facing a blue ocean yet being forced to battle "giants."

Duan's choice was again to temporarily retreat. He proposed that they didn't need to self-develop every technology, and directed the team to put greater emphasis on workflow construction — with the goal of building an "AI CapCut."

In March this year, Fotor launched Clipfly, a one-stop AI long-video platform it had been preparing for over a year, integrating AI video generation, AI video enhancement, video editing, and other functions. Users can add music, transitions, effects, video enhancement, and subtitles in Clipfly, stitching together AI's "gacha" outputs into complete videos — a process closer to how humans actually make videos.

But Clipfly's performance in the highest technical-barrier generation segment can only be described as mediocre. For generation, they simply used basic open-source solutions without further investment.

Whether from conservatism or habituated passivity, after "dipping a toe in video generation," Fotor returned to its comfort zone.

Duan says he believes more that AI-only products don't necessarily succeed, but successful products definitely have AI. "If you only focus on models, there will always be stronger technical players than you, and even giants will pour effort in and launch a product that defeats you."

Inside Fotor, a dedicated team researches cutting-edge AI applications. Each application becomes a feature, and features get added to Fotor, bringing in users through search.

Users who come to Fotor through keyword searches like "AI Image Generator" or "AI Art Generator" can, after text-to-image generation, continue using image processing (PS) and graphic design (poster) functions. When daily needs are also met, they stay — Fotor has accumulated dozens of features over ten years.

So when investors ask about user personas, Duan often doesn't know how to answer, because "our users often have no persona at all. We can serve professional users, or girls who want whitening and beauty filters." Such a product could only be born in a blank market.

This partly stems from Duan's study of ByteDance. If ByteDance was the app factory of the mobile internet era, what they're trying to do is become an AI app factory — features bring traffic, which eventually grows into standalone AI products.

This is a safer product path. Duan also doesn't shy away from acknowledging this involves some degree of wrapping. "For things that genuinely require GPUs and resources, we can't do them right now."

How to define Fotor now? Duan says it's a product company combining AIGC with traditional image tools.

Image source: IC Photo