The AI Reunion Wave China Can't Decide How to Feel About

For about a week at the start of October 2026, Chinese social platforms filled with the dead. AI-generated short films brought back recently departed public figures, set them in familiar scenes, and gave grieving audiences a moment of reunion they could watch on a phone. The most shared pieces pulled in hundreds of thousands of likes. Then the mood turned, and the same technology that had been praised as a gift started drawing accusations of exploitation.

The sequence is worth following, because it shows how fast a feel-good AI trend can curdle, and it lays out the boundaries a lot of countries are about to have to define.

How the wave started

The trigger was real loss. On September 24, the actor You Benchang, best known for playing a beloved folk character on Chinese television, died at 93. A memorial video on Bilibili drew more than 343,000 likes and close to a million views. Two days later, a tribute to the singer Liu Huan followed. Fans were already in a state of collective mourning, and the platforms were full of people who wanted to say goodbye.

Within days, the first AI tribute films appeared. A Douyin creator released a piece that reunited the departed on screen, and it collected about 134,700 likes, more than 5,400 comments, and thousands of saves. On Bilibili, creators posted their own versions under an AI-themed production event, and the views piled up. By October 1, a format had taken hold: a door opens, and behind it stand figures the audience has lost, waiting to greet each other. The list of names kept growing as more creators joined in.

Where it turned

The backlash arrived in the same week, in the form of a different kind of headline. On September 30, a story about a grandmother who was placed into a film by AI and left speechless climbed the trending lists. The point of that coverage was not whether the likeness was convincing. It was the word "speechless," the sense that something had been done to her rather than for her. Days later, a Swedish engineering company used generative AI to bring the actress Greta Garbo back for an advertisement, igniting a fresh argument about the commercial use of a deceased celebrity's image. And on September 30, the 84-year-old composer Xu Jingqing, who wrote the score for a classic adaptation of Journey to the West, publicly pushed back on the unauthorized reuse of his work, putting the line between homage and infringement squarely back on the table.

There was also a quieter problem that the wave carried along: misinformation. A black-and-white clip shown at a farewell ceremony was real footage shot by a photographer in 2024, not an AI creation. Yet it circulated widely as an example of a successful AI resurrection. In a trend built on emotion, the line between what was made and what was real blurred almost immediately.

Why this is not just a China story

Grief is a universal market. The same tools that let a fan say goodbye to an idol also let a brand sell a product using a face that can no longer consent. Deepfake workspaces have made likeness reproduction cheap, and the legal systems that might restrain it are mostly still catching up. Some jurisdictions require disclosure that an image is synthetic. Very few have clear rules about who owns a deceased person's digital likeness, or how long those rights last, or who gets to grant permission.

Add generative video and the stakes rise. A realistic moving likeness is far more unsettling than a still image, because motion reads as presence. The audience's brain fills in the rest, and the reaction that follows is stronger. Creators figured this out fast, and so did advertisers.

The reasonable position is not that AI tributes should be banned. The capacity to bring comfort is real, and plenty of the early videos came from genuine affection. The question is consent and honesty. Who agreed to this use, is the audience told it is synthetic, and does the work honor the person or merely monetize the audience's feelings. Those are questions a platform policy can partially answer and a court eventually has to finish answering.

What the rules actually cover

Most existing law was written for a world where a person's face could not be reproduced at scale by anyone with a laptop. That is why the response to this wave has been so improvised. In many places there is clear protection against using a living person's likeness for commercial purposes without consent. What happens after someone dies is murkier. Some jurisdictions recognize a post-mortem right of publicity that lasts for a set number of years, administered by an estate. Others leave it undefined, which means the answer depends on a patchwork of copyright, defamation, and personality-rights rules that were never designed for generative video.

Disclosure rules add another layer, and they are getting stricter. Several jurisdictions now require synthetic media to be labeled, especially when it depicts real people. In practice, that is the easiest fix and the one platforms can enforce fastest, because it does not require deciding who holds a deceased person's rights. A label does not stop the creation, but it changes how the audience reads it, and for a trend built on the illusion of presence, that label is not a small thing.

The consent question is the one technology keeps dodging. A person who is alive can grant or refuse permission. A person who has died cannot, which leaves the decision to an estate, a family, or a company that bought the rights. The Garbo case is instructive precisely because it turned a private question into a commercial one. Once a likeness can appear in an advertisement, the debate is no longer about whether the technology works. It is about who gets to say yes.

What the platforms learned

The arc of this trend is a preview of how AI content will keep arriving: a genuine technical capability meets a genuine human need, spreads fast because it is cheap to make and easy to feel, and then collides with rights and trust. The creators who lasted past the backlash were the ones who labeled their work clearly and stayed away from living likenesses and commercial use. The ones who got in trouble did the opposite.

For anyone building with generative video, the takeaway is not to avoid the emotional territory. It is to treat a face, especially a face that can no longer speak for itself, as something you need permission to use. The technology made that easier to ignore. The reaction to this wave suggests audiences will not let them ignore it for long.