AI mass-generates e-commerce videos: a 5-person team produces over a thousand per day, at under 5 yuan each

Many people know that in Xiamen there is a group of e-commerce players who, in order to continuously run short seed-content videos on Douyin, keep hundreds of models and dozens of photo studios on hand; the industry calls them the "Xiamen gang."

This is not about putting on a show. When advertising on Douyin, no one knows in advance whether a piece of creative will scale, so they can only publish seed content in bulk and then pick the standout creatives from among them to put ad spend behind—industry jargon calls this horse racing. Keeping a large number of models is about producing a large number of different videos on the same day and testing every direction at once. This kind of team has to produce hundreds of videos a day just to barely keep a brand continuously horse racing.

In the AI era, everything has changed. With the same output volume, AI can get it done in a few minutes. For example, a team of 5 people can produce over a thousand videos a day, with the cost per video kept under 10 yuan or even 5 yuan.

Volume has grown tenfold, yet the price has shrunk to a mere fraction.

When creative becomes as cheap as tap water, does the Xiamen gang's playbook of winning by piling on volume still hold? And where is the competitive threshold for content e-commerce going to shift to?

In this article, we'll talk about exactly that.

Creative costs pushed below 5 yuan—will AI mass production replace the "Xiamen gang" playbook?

In the e-commerce industry, the Xiamen gang has mastered a mature playbook for selling goods: find high-margin hit products, mass-produce creative, amplify with paid ad spend, and close sales in the livestream room.

Of these four steps, the most labor-intensive is creative. Buying traffic merely converts money into impressions; creative determines whether those impressions get clicked. So in many people's minds, the Xiamen gang's strongest capability is not making videos, but turning seed videos into industrial goods that can be mass-produced.

Public reports mention that a leading Xiamen gang group can put out over a hundred short videos a day, with the cost per video pressed down to 100–200 yuan.

From an outside perspective, Yunying She also found that many Xiamen gang brands indeed win on Douyin by volume. Take Dr. Wen as an example: Haini Group has opened a large number of accounts on Douyin, releasing over a thousand new videos a month, using "human-wave tactics" to push hit products to the extreme; another top-tier beauty brand operates 4 key blue-V official accounts, each updating 30 to 40 posts a day on average. Over a month, that adds up to several thousand videos for a single brand—in the past, that was the total creative output of an ordinary brand for a whole year.

Another case Yunying She saw comes from an AI team of only about 5 people, which, with the help of AIGC, can generate over a thousand videos in a few minutes, at a cost of less than 5 yuan per video. That figure may be only one-tenth, or even one-twentieth, of the cost of manual shooting.

Specifically, many merchants actually started trying AIGC for bulk content creation long ago. At first it was some white-label merchants; by this year, leading merchants in industries such as food, beauty, personal care, and gaming have also begun using AI content to carve up traffic, with gaming being the most representative.

Over the past six months or so, many people have probably come across game ads like those for Crazy Water World:

A put-upon bald protagonist is bullied in turn by his neighbors in a post-apocalyptic shelter, and his entire hope in life is a seawater filter. When he finally moves into a premium shelter, he finds it too small, then goes off to fish for a moray eel generator, and at the end of the video suddenly shouts "Crazy Water World launch," cuts to the game footage, and starts introducing the product.

These clips, just tens of seconds long, became so popular that netizens followed them like a drama series. A single video of the protagonist "turning his life around" can rack up over 300,000 likes on Douyin, and the topic's play count has even approached 1 billion views.

Similar AI ad content is too numerous to list. These seemingly scattered cases all point to the same thing: AI is taking over the step of mass-producing creative. In the past, this step relied on manpower—scriptwriters writing scripts, teams scheduling shoots, models appearing on camera; whoever had more people, more studios, and higher efficiency could blanket the market faster. Now, a team that knows how to use prompts can redo this step entirely, cutting costs to a mere fraction of what they used to be.

At present, leading Xiamen beauty brands represented by Dr. Wen and Kezhu have not yet used purely AI-generated videos for seed content at scale. Instead, many merchants in other fields are using AIGC to leverage traffic at low cost.

With costs down to 5 yuan, how does AI mass-produce creative?

Cutting the cost per video from over a hundred yuan to 5 yuan is not savings out of thin air; it comes from operators reorganizing content production from the ground up. After sorting through it, Yunying She finds it roughly comes down to three steps: first break things into parts, then produce videos in tiers, and finally get past platform review.

1) Break creative into building blocks and churn out volume with a set of proven templates

Most veteran operators typically split instructions into two categories. One is fixed blocks, which govern what the video looks like—video style, lighting, aspect ratio, camera movement, and so on. These are the same for every piece of creative and can be set once and reused.

The other is variable blocks, which govern what the video says—selling points, scenes, character actions, voiceover copy. These instructions keep changing with the product and target audience.

Specifically, once the fixed blocks are set, that's a template. The variable blocks are the parts of the template that can be swapped. What the industry often calls "swapping variables to produce dozens of videos" means deriving dozens of different creatives from the same template.

It is understood that there is no fixed number of templates a brand needs to maintain; generally, several to dozens are maintained at the same time, corresponding to different selling-point angles, audiences, and scenes. For example, the same bottle of serum can be broken into several template sets—ingredient-focused explainers, before-and-after comparisons, ordinary-user reviews—and each set can then derive dozens of pieces of content through variables. Usually, the more templates, the larger the traffic pools they can cover.

Generally speaking, there are usually two paths to creating a new template. One is to reverse-engineer it from a hit that has already performed well: when a certain creative takes off, break it down into "fixed blocks + variable blocks" and distill it into a new template—this is replicating success. The other is to build proactively: based on new selling points, trends, and audiences, assemble a new set of fixed blocks from scratch—essentially testing a new direction.

There is another point that is often overlooked: templates are not eternal. The more creatives of the same style you run, the more fatigued users become from seeing them, and the platform will also judge them as homogenized, causing impressions to decline accordingly. To decide whether a template should be retired, mainly look at two signals: the run rate falls below expectations, or completion and engagement drop noticeably. At that point, it means the template's freshness is exhausted, and it is time to change the style and change the hook.

2) To cut costs, they prioritized models by price

Once the template is set, it is time for production.

An increasingly common practice in the industry is tiered trial and error: first use a cheap, fast model to produce a 5–10 second version, testing only the hook and direction. Whichever ones perform, switch to a better model and make the full version of the content.

The so-called hook is the attention-grabbing element in the first few seconds of a video, such as "What is the most important step in skincare?" or "Over 50,000 commuters are dressing like this." Direction refers to which audience, which selling point, and which scenario the creative focuses on. These two elements are the top priority of the content and often require repeated tuning.

Models also come in "expensive and cheap" tiers. Many operators recommend using cheap models to test hooks and direction, such as Grok Video and the entry version of Seedance, at a few yuan per video; and using models with better content quality for the final product, such as Veo, Sora, and the HD version of Seedance, which are more expensive but deliver more stable content quality. Some practitioners have calculated that testing ten hooks on the cheapest model costs only as much as finely polishing one finished video with a high-priced model.

Many cases prove that this production approach is relatively cost-saving and reliable. For example, one operator used this method for three months of testing, cutting creative production costs by 60% and more than doubling monthly output. At the same time, content that has been repeatedly tested is also more likely to pass the platform's review, and the approval rate has risen considerably.

3) Once the content is done, how do you get past the platform?

With the template broken down and the direction tested, the finished product reaches the final step: passing review. According to Yunying Society, this year Douyin, Xiaohongshu, and other content platforms have noticeably tightened their review of AIGC.

English text is being prepared.

At the same time, the platform has drawn a safety line: if AI is used only for visual enhancement and scene setup, and the efficacy matches real capabilities, it is allowed through. AI content that has been labeled with notices such as "AI-generated" or "demonstration effect, for reference only" and does not mislead users is also allowed to run.

If Douyin is policing whether the ad is real, Xiaohongshu is watching whether the account is a real person.

Since 2023, Xiaohongshu has required creators to proactively declare AI content, and in April this year it issued complete governance rules. In July, labeling was extended from notes to accounts: accounts operated by AI virtual humans must clearly display a label on their homepage, and if unlabeled accounts are identified, they will be throttled. WeChat Channels followed suit in August this year: e-commerce short videos that do not proactively add AI labels will have their distribution restricted.

Having understood the platform's sensitive elements, Yunying Society recommends that operators do three things before publishing AIGC-related content, to ensure the content is not throttled for violations.

Second, before the finished video is released, check it yourself three times. The first pass checks whether the visuals have collapsed or the text is garbled; the second pass checks whether the colors, labels, and aspect ratio are correct; the third pass checks whether the duration and content are compliant. Only after all three passes is the finished product ready to go out. The purpose of human review is to hold the bottom line and avoid situations like Supor's, where the official account mass-published AI-generated vulgar, borderline videos, triggering a public opinion storm and ultimately damaging the brand image.

Conclusion

In Yunying Society's view, the most important point in this round of change brought by AI is that AI has pushed creatives down to 5 yuan, drastically lowering testing costs. In the past, a single content creative cost tens to hundreds of yuan, and a team could only try a limited amount per month. Now, at 5 yuan apiece, the cost of trial and error is low enough to freely try more possibilities. But cheaper trial and error does not mean cheaper hits: the more creatives there are, the harder it becomes to judge which piece of content is worth putting more spend behind.

There is another thing that is often misread: many people think that once AI arrives, new players will squeeze out old players. The reality is that merchants and brands that once championed the Xiamen-style playbook are now also using AI to produce creatives at scale, and they are picking it up fast. In a sense, what AI replaces is the filming step; the brands that use the Xiamen-style playbook to acquire customers and sell goods are still at the table.

Overall, Yunying Society's judgment is that AI's core change to content e-commerce is that creatives are no longer scarce. When production capacity is no longer a barrier, what is truly valuable is judgment and taste. Previously, it was about who shot more and shot faster; now it is about who can pick, from a pile of AIGC, the piece of content that will perform best.

This article is from the WeChat public account "Yunying Research Society" (ID: U_quan), author: Yunying Society Editorial Department, published by 36Kr with authorization.