Your AI-Made Cover Image Needs a Label in China

Somebody on your team needed a cover image, could not find a photograph that worked, and generated one. It went up on Xiaohongshu, it looked fine, and nothing happened.

China has had a labeling rule for AI-generated content since September 2025, and almost no overseas operator we speak to has heard of it. The gap matters more for you than for a Chinese business, for a reason that has nothing to do with intent: the tool you used was probably not built to comply with it.

The short version: since 1 September 2025, AI-generated and AI-synthesised content published in China has to carry a label, both a visible one and one embedded in the file. The heavy duties fall on the AI services and the platforms, not on you. What lands on you is narrower and still real: use the platform's own AI-declaration control when you post, and understand that a Western image tool leaves you carrying nothing. "Synthesised" is broader than "generated", so an AI-extended background counts too.

What the rule says

The Measures on Labeling AI-Generated and Synthesised Content took effect on 1 September 2025, issued jointly by the Cyberspace Administration and three other ministries, with a mandatory national standard sitting underneath them.

Two kinds of label are required. An explicit one that a person can perceive: for an image, text containing "AI" or 人工智能 together with 生成 or 合成, placed at an edge or a corner. An implicit one written into the file's own data, which nobody sees and which travels with the file.

The explicit label has a size floor that surprises people. The text has to be at least five percent of the image's shortest side in height. On a note image 1080 pixels wide, that is around 54 pixels of type — larger than most people would choose if they were designing it to be discreet.

Who the rule is actually aimed at

This is the part worth understanding before you panic, and also before you relax.

The obligations sit mainly with two groups. The AI service that generated the content is required to attach the labels. The platform that distributes it is required to check, to provide a way for users to declare AI content, and to act when content is unlabeled.

You, posting a cover image for your tour, are neither. Your duty in practice is the one the platforms impose: declare it. Every major Chinese platform now has a control for exactly this, and their user agreements say what happens if you do not use it.

The gap that is specific to you

Here is why this is more of a problem for an operator in Reykjavik than for one in Chengdu.

A Chinese AI tool generates an image with both labels already attached, because the law requires the tool to do it. The person posting inherits compliance without thinking about it.

A Western tool was not built against this standard. Whatever you generated in it arrives with no Chinese-language visible label and, in most cases, nothing in the file either. So the image you are about to post is unlabeled, and the only thing standing between you and an unlabeled AI post is the declaration control you may not know exists.

Two habits follow. Use the platform's AI toggle every time, and stop exporting through tools that strip file metadata — if your generator did embed something, a careless resize can remove it.

"Synthesised" is wider than "generated"

The rule covers 生成 and 合成. Generation and synthesis. Most people read the first word and skip the second, and the second is where ordinary editing lives.

Things that are likely to count, even though the picture is fundamentally a photograph you took:

• A background extended by generative fill so the image fits a different crop.
• A person, a bin or a parked car removed by a generative tool.
• Sky replacement.
• AI upscaling that invents detail that was not in the original.
• A synthetic voice reading a caption over your video.
• Subtitles or captions produced by a model.

We are not going to draw the line for you on any given edit, because the line is not ours to draw and it is still moving. The workable posture is that if a model added or invented pixels, declare it.

What actually happens if you do not

Worth being accurate here, because the alarming numbers circulating online attach to something else.

For a small overseas operator, the realistic consequence is at platform level: reduced distribution, the note taken down, and for a repeated pattern, the account. Xiaohongshu, Douyin, WeChat Channels, Weibo and Bilibili all run detection and all say in their terms that unlabeled AI content can cost you the account.

The large administrative fines you may have read about are attached to breaches of other laws — cybersecurity, personal information, data security — that happen alongside. They are not the tariff for an unlabeled cover image, and anyone telling a tour operator otherwise is selling something.

The practical risk to you is the same one we described in why notes get removed: quiet loss of reach, on an account you were building patiently.

A separate question we are leaving for later

Whether you should be using a generated image of a place you sell at all is a different argument from whether you labeled it. A traveler who arrives to find the light, the crowd and the color of the water unlike your picture has a complaint that no label prevents. We will come back to that argument on its own, because it is a different one.

For now: the labeling duty applies whether or not the image was a good idea.

What to do this week

Four things, none of which takes long:

1. Look through what you have already published. Anything generated, still up, and undeclared is worth fixing rather than waiting to see.
2. Find the AI-declaration control in each platform you post to, and make using it part of the posting routine rather than a thing somebody remembers.
3. Decide who signs off on images. In most small teams nobody does, which is how a generated image reaches a Chinese account without anyone weighing it.
4. Write down which tools your team uses. You cannot reason about whether your images are labeled if you do not know what made them.

The boundary of this

These are working habits, not legal advice. The Measures are recent, the mandatory standard beneath them is detailed, platform rules sit on top and are revised without notice, and enforcement is still developing. Where a campaign carries real money, the next step is somebody qualified in Chinese content law, not more reading. The wider map of what applies to an overseas seller is in compliance basics.

Where CN1X fits

We publish to accounts registered to your company, so this lands on us in practice. Anything we produce with a generative tool is declared through the platform's own control and labeled where the standard calls for it, and the images in a batch that were made that way are identified when we send it. Where you send us a generated image, we will say so before it goes up.

If you have Chinese accounts with generated images already published and nobody has checked them, tell us where the account is — an audit of what is already up is a short job and a better use of an afternoon than a takedown notice.