AI is wildly emulating the "Xiamen Gang" playbook of Douyin, churning out 200 videos per day at a cost of less than 5 yuan per video.

As many people know, there is a group of e-commerce practitioners in Xiamen who prepare hundreds of models and dozens of photo studios to continuously post grass-planting short videos on Douyin. In the industry, they are called the "Xiamen Gang".

This is not a show of extravagance. When placing ads on Douyin, no one can tell in advance whether a piece of material can achieve high traffic performance. The only way is to release a large number of grass-planting contents in batches, select the outstanding-performing materials to put in more ad budget, which is called "horse racing" in industry jargon. Keeping a large number of models is to produce a large number of different videos on the same day to test all possible directions. This team has to produce hundreds of videos a day to barely maintain the continuous "horse racing" operation of the brand.

In the AI era, everything has changed. With the same output, AI can finish the work in just a few minutes. For example, a team of 5 people can produce thousands of videos a day, and the cost of each video can be controlled below 10 yuan or even 5 yuan.

The output has increased tenfold, while the cost is only a small fraction of the original.

When materials become as cheap as tap water, can the Xiamen Gang's strategy of relying on massive volume still work? Where will the competitive threshold of content e-commerce shift to next?

In this article, we will talk about this topic.

With material cost dropping below 5 yuan, will AI mass production replace the "Xiamen Gang" strategy?

In the e-commerce industry, the Xiamen Gang has mastered a mature sales logic: find high-margin hit products, produce materials in batches, pay for ad traffic amplification, and close deals in live streaming rooms.

Among the four steps, the one that consumes the most manpower is material production. Ad placement only converts money into exposure, and materials determine whether these exposures can be clicked. Therefore, in the minds of many people, the Xiamen Gang's strongest capability is not shooting videos, but turning grass-planting videos into industrial products that can be mass-produced.

Public reports mention that a leading Xiamen Gang group can produce hundreds of short videos a day, and the cost of each video can be reduced to 100~200 yuan.

From an external perspective, Operation Society also found that many Xiamen Gang brands do win on Douyin by volume. Take Dr. Wen as an example, Haini Group has opened a large number of accounts on Douyin, launching thousands of new videos every month, pushing the "human sea tactic" to the extreme for hit products; another top-level beauty brand, with 4 key official Blue V accounts, updates 30 to 40 videos per account on average every day. In one month, a brand can accumulate thousands of videos, which was the total amount of materials an ordinary brand could produce in a year in the past.

In the AI era, what agents take over exactly is the mass production of content. 

Another case observed by Operation Society comes from an AI team of only about 5 people. With the help of AIGC, they can generate thousands of videos in a few minutes, with the cost of each video less than 5 yuan. This figure may be only one tenth or even one twentieth of the cost of manual shooting.

What deserves more attention is the speed. An AI video can be completed within half an hour from planning and generation to upload and becoming a hit.

Specifically, in fact, many merchants have tried to produce content in batches with AIGC long ago. At the very beginning, some no-name brands tried it. This year, leading merchants in food, beauty, personal care, games and other industries have also started to use AI content to grab traffic, among which the game industry is the most representative.

In the past six months, many people must have come across game ads like *Crazy Water World*:

A bullied bald protagonist is bullied by neighbors in turn in the doomsday shelter, and his only hope in life is a seawater filter. When he finally moves into a high-level shelter, he finds it too small, then goes to salvage an eel generator. At the end of the video, he suddenly shouts "Crazy Water World is launched", then the game screen pops up and the product introduction starts.

These dozens of seconds of clips are so popular that netizens follow them as if they are TV series. A video of the protagonist "turning the tables" can get more than 300,000 likes on Douyin, and the topic view count is even close to 1 billion.

Some other brands use AI to put their products into overbearing president short dramas, interspersing product introductions in the love stories of male and female protagonists. They post dozens of similar materials a day, and simultaneously catch up with daily hot topics to grab traffic for the brand in various ways.

Some merchants even directly use AI to make suggestive plots, setting up suspense at the beginning of the video to "hook" users until the product appears, for example, a girl is taking a shower when an uncle breaks in; a man is in the bathroom when a beautiful woman suddenly bursts in...

Similar AI advertising contents are too numerous to list. These seemingly scattered cases all point to the same thing: AI is taking over the process of mass-producing materials. In the past, this process relied on manpower: screenwriters wrote scripts, teams arranged shooting, and models appeared on camera. The one with more people, more studios and higher efficiency could lay out volume faster. Now, a team that knows how to use prompts can redo this process, cutting the cost to only a tiny fraction of the past.

In other words, what has been changed is not the business of a certain company, but the model of piling up materials with manpower itself. When this process is replaced by AI, will the scale advantage that the Xiamen Gang relies on for survival be shaken?

At present, leading beauty brands in Xiamen represented by Dr. Wen and Kezhu have not yet used pure AI-generated videos on a large scale for grass-planting. On the contrary, many merchants in other fields are using AIGC to seize traffic at low cost.

Cost drops to 5 yuan, how AI produces materials in batches

The cost per piece drops from hundreds of yuan to 5 yuan, which is not saved out of thin air, but because operators have reorganized the content production process. According to the sorting of Operation Society, there are roughly three steps: split components first, produce videos at different levels, and finally pass the platform review.

1) Split materials into building blocks, and achieve massive volume with a set of verified templates

Let's look at the first step of splitting components first. AI generates videos based on a text instruction, and what kind of clips the model produces depends entirely on how this instruction is written. Therefore, to some extent, the most critical step in making materials has become writing prompts.

Most senior operators usually split prompts into two categories. One is the fixed block, which controls what the video looks like, including video style, lighting, aspect ratio, camera movement, etc. These contents are the same for every material, and can be used repeatedly after being set once.

The other category is the variable block, which controls what the video is about, such as selling points, scenarios, character actions, and voiceover copy. These prompts keep changing with products and target audiences.

Specifically, when the fixed block is determined, it becomes a set of templates. The variable block is the replaceable part in the template. What the industry often calls "changing variables to produce dozens of videos" means deriving dozens of different materials from the same set of templates.

It is understood that there is no fixed number of templates a brand needs to maintain. Usually, they maintain several to dozens of sets at the same time, corresponding to different selling point angles, crowds and scenarios. For example, for the same bottle of essence, several sets of templates can be split out, such as ingredient popularization, before-and-after use comparison, layperson evaluation, etc. Each set can derive dozens of contents through variables. The more templates there are, the larger the traffic pool that can be covered.

Generally speaking, there are two common paths to create new templates. One is to reverse split from the hit videos that have achieved good performance: when a piece of material becomes popular, split it into "fixed block + variable block" and precipitate it into a new set of templates, which is to replicate success. The other is to create proactively, build a new set of fixed blocks according to new selling points, hot topics and crowds, which is equivalent to testing new directions.

Another point that is often ignored is that templates are not permanent. The more materials of the same style are released, the more fatigued users will feel, and the platform will also determine that the content is homogenized, so the exposure will drop accordingly. To judge whether a set of templates needs to be eliminated, we mainly look at two signals: the traffic performance rate drops below expectations, or the completion rate and interaction metrics decline significantly. At this point, it means that the freshness of this set of templates is exhausted, and it is time to change the style and the opening hook.

2) To reduce costs, they prioritize models by price

After the template is determined, it comes to the video production stage.

A more and more common practice in the industry is hierarchical trial and error. First use a cheap fast model to produce a 5~10 second version, only to test the opening hook and general direction. Only after some clips show good data performance, then switch to a better model to produce the full version of the content.

The so-called hook refers to the eye-catching content in the first few seconds of the video, such as "What is the most important step in skin care?" "More than 50,000 commuters are dressing like this". The direction refers to which type of crowd, which selling point and which scenario the material focuses on. These two elements are the top priority of content, and often need to be debugged repeatedly.

Models also differ in cost. Many operators suggest using cheap models to test hooks and directions, such as the entry version of Grok Video and Seedance, which costs a few yuan per piece; use models with better content quality to produce finished videos, such as the high-definition version of Veo, Sora and Seedance, which are more expensive but have more stable content quality. Some practitioners have calculated that testing ten hooks on the cheapest model only costs as much as using a high-priced model to refine one finished video.

Many cases have proved that this production method can relatively save costs and is quite reliable. For example, an operator used this method to test for three months, the production cost of materials was reduced by 60%, and the monthly production capacity even doubled. At the same time, the content that has been tested repeatedly is easier to pass the platform test, and the review pass rate has also increased significantly.

3) After the content is produced, how to pass the platform review?

With the templates split and the direction tested, the finished video reaches the last step: review. According to Operation Society, this year, the review rules for AIGC content on content platforms such as Douyin and Xiaohongshu have been significantly tightened.

Specifically, to get a high review pass rate, you need to first understand what the platform cares about. What Douyin focuses on is "don't let AI act and brag for the product".

At the end of August this year, Ocean Engine announced the results of a special governance campaign for AI illegal marketing, and more than 50,000 illegal accounts were shut down in total. The named operations are divided into three categories: hit efficacy images that violate common sense of physics, such as kitchen utensils cutting ingredients as thin as cicada wings, plasters spraying out toxins, etc; making deformed values plots such as conflicts between mothers-in-law and daughters-in-law; using violent and vulgar images to attract traffic.

At the same time, the platform has also drawn a safety line: AI is only used for picture beautification and scene construction, and the efficacy conforms to the real product capability, then the content will be approved. AI content marked with relevant prompts such as "AI generated" and "Demonstration effect for reference only" that does not mislead users is also allowed to be published.

If Douyin focuses on whether the advertisement is true, then Xiaohongshu focuses on whether the account is operated by a real person.

Xiaohongshu has required creators to actively declare AI content since 2023, and released complete governance rules in April this year. In July, the labeling requirement was extended from notes to accounts, requiring accounts operated by AI virtual avatars to clearly show the label on their homepage, and unlabeled accounts will be restricted in traffic once identified. WeChat Channels also followed up in August this year, restricting the distribution of live-streaming marketing short videos that do not actively add AI labels.

After understanding the sensitive elements of the platform, Operation Society suggests that operators do three things before publishing AIGC-related content to ensure that the content will not be restricted in traffic due to violations.

First, build a forbidden word list. Avoid absolute expressions, implied efficacy claims and sensitive scenarios. Use AI to filter the content before generation to avoid the platform's red line from the source.

Second, check the finished video three times by yourself before publishing. Check for the first time to see if the picture is distorted and the text is garbled; check for the second time to see if the color, label and proportion are correct; check for the third time to see if the duration and content comply with the regulations. After all three checks are passed, the finished video can be released. The purpose of manual review is to hold the bottom line, to avoid situations like Supor, whose official account released a large number of AI-generated vulgar suggestive videos in batches, causing public opinion storm and eventually damaging the brand image.

Third, check the content for different platforms separately. For content published on Douyin, focus on checking whether it is overacted; for content published on Xiaohongshu, focus on checking whether it looks like it is posted by a real person, especially AI virtual avatar voiceover materials, which are the key restricted objects for Xiaohongshu traffic limit.

Conclusion

In the view of Operation Society, the most important point in this round of changes brought by AI is that AI reduces the cost of materials to 5 yuan, and the test cost has dropped significantly. In the past, the cost of a content material was tens to hundreds of yuan, and the total amount of trials a team could do in a month was limited. Now the cost per piece is 5 yuan, the trial and error cost is so low that people can try more possibilities freely. But the cheaper the right to try and error is, does not mean the cheaper it is to produce a hit. The more materials there are, the more difficult it is to judge which content is worth putting in more ad budget.

There is another thing that is often misread: many people think that with the arrival of AI, new players will squeeze out old players. The actual situation is that