Wu Gaoming of Jingying Technology: AI Video, the Ultimate Solution is a Screenwriter-Centered System

AI has replaced directors, actors, and post-production staff, but it cannot replace screenwriters.

This is the assessment of Wu Gaoming, co-founder of Jingying Technology.

In the first quarter of this year, 128,000 new short dramas were launched in China, with AI-generated content accounting for over 95%. Monthly active users reached 718 million, with an average of 129 minutes of viewing time per user per day, surpassing long-form video for the first time. Global in-app purchase revenue for short dramas reached $2.98 billion last year, more than doubling year-on-year.

On the content production side, the entire production chain is being taken over by AI, but Wu Gaoming believes that screenwriters are the last and most irreplaceable link in this chain.

Because the source of creativity, intuition about people's hearts, experience of the world... these things are not structured and placed on the Internet, so the model cannot learn them.

Founded in 2021, Jingying Technology is one of the world's earliest companies to successfully establish a closed loop for paid AI short dramas. Its subsidiary, Reel.AI, targets overseas markets. According to public reports, in November 2025, the platform's production, "The Billionaire's Return," surpassed live-action short dramas to top the overseas short drama charts.

AI-generated short dramas have achieved this for the first time.

But Wu Gaoming doesn't define Jingying Technology as an "AI short drama company." The backend video generation, dubbing, and editing are handled by AI, while the frontend storytelling and aesthetics are left to the creators. The platform provides computing power, tools, and distribution channels; creators only need to focus on one thing: telling a good story.

In June of this year, Jingying Technology completed tens of millions of US dollars in Series A and Series A+ financing. Investors included Lollapalooza Capital (owned by Wang Huiwen's family), Ant Group, and Yin Yu, former vice president of Tencent.

The next stage of competition may determine the winners and losers on the creators' side.

The ultimate form of AI video is a script-centric system.

The order in which AI replaces each stage of content production is from the end to the beginning.

Models are trained on data; the further back in the process, the more data they have, and the faster they learn. Later stages have massive amounts of video footage, and these are the first to fail; the actors' performances and the director's style and techniques also have discernible patterns.

Wu Gaoming has a saying: film itself is an "art of deception." Actors use their performances to draw the audience into their roles, and directors use cinematic language to create emotions. AI is no different from humans in these "deceptive" aspects; if it's realistic enough, the audience will buy it. Reel.AI's user data confirms this: users neither refuse to watch content because it's AI-generated, nor do they blindly follow it simply because it's AI. "Ultimately, as long as user preferences are satisfied, they don't care about the production methods behind it," Wu Gaoming said.

However, in the short term, AI is unlikely to replace screenwriters.

"Many high-quality creative ideas haven't been structured and data-driven onto the internet," Wu Gaoming said. "Why do people think a story is good? Why do they have the urge to create a certain story? These are people's tacit knowledge."

Models naturally tend to converge to a central distribution, and after extensive training, the output approximates the "most common appearance." This is how "AI faces" came about—the model converges all faces to near the mean.

However, the content industry does not reward averages.

It heavily rewards sharp, off-the-beaten-path content, and anything that captures the creator's unique experience. A short drama set in an Antarctic research station, telling the story of an American scientist and a Russian scientist stranded in the icy wilderness, was once produced on Jingying Technology's platform. This kind of subject matter wouldn't appear on traditional film and television schedules, and top-tier producers wouldn't fund it. But it struck a chord with a genuine user need.

What captures this need is not a model, but a screenwriter's intuition about "what is worth telling".

As AI takes over the back-end production processes one by one, the focus of the entire content industry is shifting towards the front end—who decides what stories to tell, to whom to tell them, and how to tell them is becoming the most valuable aspect. Wu Gaoming believes this is the logic behind AI video moving towards a "writer-centric" model. After the director-centric model is dissolved by AI, the weight of creative judgment increases rather than decreases.

Find the right people and build an industry-leading system.

For Jingying Technology, after the technical threshold was lowered, the real challenge became finding the right people.

The team previously worked on one of the top three free novel platforms in China, and that experience instilled in them a methodology: don't finish the content before testing it; just do a small part.

In 2024, almost no one believed in AI short dramas. Wu Gaoming needed creators who had both film and television backgrounds and genuinely believed that AI could create good content. Such people were extremely rare. He searched online for creators with film and television backgrounds and sent out mass questionnaires. The questionnaire contained a key question: evaluate an episode of an AI short drama that Jingying Technology could already generate, noting that the visuals were rough and full of plot holes.

What he wants to see is the reaction after pointing out the flaws—whether they will find a way to circumvent the shortcomings of AI, and whether they believe that this can ultimately succeed.

More than 1,000 people filled out the questionnaire. In the end, 10 were selected.

One of them was named Cui Yi. He later created the platform's first AI short drama to achieve commercial success overseas.

Jingying Technology does not produce its own content to date; almost all creators come from external sources. The revenue-sharing model is a guaranteed minimum plus profit sharing: a guaranteed minimum is received after the outline passes evaluation, a guaranteed minimum is received upon completion of the script, and a profit sharing is provided after the content is released. Money is received at each stage, and the platform bears the risk.

Through his experience, Wu Gaoming discovered that two types of people are particularly well-suited to this system: those with experience in the traditional film and television industry but who haven't reached the top, and those who studied film and television production overseas but find themselves unable to integrate into the domestic industry. AI has given those who were previously overlooked a new opportunity to shine.

Once the right people were found, the efficiency gap widened rapidly. A traditional live-action short drama costs between $150,000 and $300,000, and only the boss can make the final decision; a single round of testing takes at least three months. Jingying Technology, on the other hand, only releases the first 8 to 10 episodes to the market to observe conversion rates, user reviews, and payment signals. Once the data is available, it's clear whether it's worth continuing. "Others take at least three months to conduct a single experiment," Wu Gaoming said, "while we might only need a week."

AI here resembles a flexible supply chain—extremely low-cost trial and error at the front end, and once a breakthrough is achieved, back-end capacity is immediately maximized. Jingying Technology also has an internal "content evaluator" that can predict market reactions before an idea even reaches the user end.

The company now generates millions of dollars in revenue each month. In January 2025, Five Brothers became the world's first AI-generated short drama to compete with live-action short dramas on the same charts; in November of the same year, The Billionaire's Return topped the charts.

The final piece of the puzzle in the closed loop

Early on, Jingying Technology made a misjudgment in its content decision-making due to data noise in the payment process.

A group of users who appeared to be paying were actually engaging in fraudulent activities. The team attributed the problem to content quality but couldn't find the cause after a long review. Later, they discovered that the problem lay in the payment signal.

Jingying Technology chose Stripe. Initially, it only accounted for about 30% of total revenue, while the company also handled other channels and built its own routers. However, as the data accumulated, it became clearer: Stripe's user renewal rate was significantly higher, with a long-term ROI difference of about 20%.

"This routing result was not manipulated," Wu Gaoming said. Stripe's transaction volume share eventually climbed naturally to nearly 70%.

The underlying logic is that Stripe's risk control system, Radar, filters high-risk behaviors during the registration phase. Melina Lee, General Manager of Enterprise Clients for Stripe Greater China, who has collaborated with Jingying Technology for many years, shared that for AI applications, once users start using the application, it incurs real-money token consumption for enterprises, and the cost is very high. Radar can detect these potential token theft, malicious registration for free trials, and other behaviors in advance, helping enterprises to anticipate potential risks.

At the payment level, Smart Retries automatically retry failed payments at the optimal time, winning back many subscribers who churned due to card issues rather than active cancellation. Liu Sicheng, Head of Corporate Sales for Stripe Greater China, told Wall Street Insights that Stripe maximizes the chances of users completing subsequent renewals. For example, everyone agrees that there's a balance on the card for the first payment; however, the card might be expired or have no funds for the second payment. Each service provider handles this "trial and error" and the second payment differently. Because Stripe has a large dataset and a large consumer base, it can more accurately capture the optimal payment time to ensure successful renewals. This translates to higher customer value for merchants. Smart Retries maximizes the success rate of payments during the renewal process. "This capability also relies on Stripe's large payment model, and our 'corpus' is the payment data itself."

For Jingying Technology, clean payment data leads to accurate content judgment. Whether a show gets produced or given resources depends on user signals. If the signals are dirty, the judgment will be wrong.

end

In 2026, Wang Minjie, former Chief Application Scientist at AWS, joined Jingying Technology as Chief Scientist. Founder Zhu Jiang believes that Coding Agents have proven that agents can profoundly reshape an industry. "The content industry is next."

The AI content industry is still in its very early stages, and it's too early to draw conclusions about what the companies that ultimately emerge will look like.

And Jingying Technology's exploration may also be answering a more fundamental question in the content industry: when AI takes over the execution layer of content production, will the real competition ultimately come down to who can tell a good story?

This article is from the WeChat official account "Hard AI" . For more cutting-edge AI information, please go here.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.