A vocational-school editor turned Malfoy into a lady with AI, and pulled 30 million views overseas

In mid-November, a Harry Potter fan series made by a Chinese creator started spreading overseas. Across X, YouTube and TikTok the whole thing has piled up 30 million views, with individual episodes mostly in the millions. The singer JVKE left a comment thanking him for using one of his songs as the soundtrack, then followed him over to Bilibili to keep talking. Hollywood producer PJ Accetturo wrote "another masterpiece." Viewers in the US, Brazil, Spain and Korea started shooting their own reaction videos and asking for new episodes across time zones.
The person behind it goes by foyege. He studied film post-production at a vocational school, and his first job out of school was as an editor.
First, who this guy is
He spent time at a film and advertising company in Guangzhou, cycling through directing, camera operating and production. None of that sounds related to AI, and yet it is the precondition for all of this. He knows how many pairs of hands a scene passes through between the script and the finished cut, and he knows which of those steps can be skipped.
A lot of AI video goes wrong at the point where nobody decided what to shoot. The camera moves, the characters talk, and the scene still falls apart. What keeps people coming back to foyege's episodes is his judgment about shot breakdown and pacing. The models only draw that judgment.
The premise is the traffic
What he rewrote was the character rather than the plot. Malfoy is a woman now: blonde, arrogant, sharp-tongued, and her arc with Harry runs from sworn enemy to something like a crush.
That premise lands on demand that already existed. Fan writers have spent over a decade on the Harry and Malfoy rivalry, so audiences arrive with a high tolerance for hostility turning into something else, and a standing appetite for it. The gender swap pushes the appetite harder. One character who is aggressive and fragile at the same time is difficult to stop watching.
Structurally he takes the slow-burn road. The first two episodes load up the friction, the two of them sniping at each other while their eyes give it away early. Episode three speeds up the feeling and closes on a memory sequence that lets Harry see his own mind in flashback. That rhythm is how a seasoned idol drama gets built, and random generation does not get you there.
A few things in the sound and image deserve their own paragraph. The ball scene opens on a 360-degree orbit that establishes the depth of the hall first. When Malfoy enters, Harry glances back offhand, the cut lands on her back on the drum hit of the soundtrack, and six camera positions go by in twenty seconds, sliding from a profile close-up down to the snake-shaped belt at her waist. In episode two, where she stands up to the Death Eaters for Harry, there is almost no dialogue. Low strings and an electronic pulse carry the floor, and the moment the spell bursts the sound effect and the heavy drums arrive together.
One canvas, several models
What made me want to write about this was his workflow rather than the finished episodes.
For character design he skipped the shortcut of blending a celebrity face. He drew a dozen versions from a comic rough, picked the female Malfoy he had in mind, then turned it into a photoreal three-view asset. That step matters. The three-view is the anchor for every shot that follows, and once the face drifts the audience drops out of the story.
For story and storyboards, he runs the whole thing through his head first, then hands the AI a 2D composition description plus reference images. Note the order. The story comes from him. He breaks it into a shape the model can swallow.
The images land on LibTV's infinite canvas. He drives Seedance, Wanxiang and Kling from one canvas at the same time, with the three-view assets, storyboard references and long prompts all laid out in the same place, and when a take disappoints he switches models and rolls again to compare. A crying scene or a look between two characters, he admits, is more than one model can carry. He needs different models to each render the same scene, then picks whichever sits closest.

One judgment here I share: the canvas is the product. A video model, however strong, only answers whether a given shot can be generated. The place where creators actually get stuck is elsewhere. Is the character in this version the same person as in the last one? Where is the reference image for shot 12 saved? Do the earlier scenes need a re-render now that the line changed? Bounce between a dozen browser tabs and apps for that and the cost multiplies several times over. Pulling the workflow into one canvas saves attention, and the compute saving is a side effect.
Platforms bet on the person
One detail is easy to miss. When foyege had around two thousand followers on Xiaohongshu and was still a niche technical blogger, LibTV signed him exclusively. After the series blew up, platforms in China and abroad came courting, and he stayed.
Why move at two thousand followers? Because the value of this kind of creator sits in output and method rather than audience size. Someone with end-to-end experience who can also get a stable feel out of the models will keep shipping. Wait until the hit lands and the price is no longer that price.
LibTV also ran an AI creation program and hackathons with Douyin, pushing compute and tools toward working creators. It is the same play short-video platforms ran years ago when they courted creators: get the supply side moving first.
One thing that has to be said
This is unauthorized fan work on someone else's IP.
The rights to the Harry Potter characters and world sit with the author and the studios, and a gender swap, a romance or a spin-off plot does nothing to change that. The hotter the work gets, the tighter the line, and China already has precedents. On November 10, a portrait-rights suit filed by the actress Zhao Liying against the operator of Hongguo Short Drama and two other defendants opened at the Mentougou court in Beijing, and one core question is whether an AI-composited face counts as a likeness in the legal sense. Her lawyer put it plainly: an exact match is not required, and if ordinary viewers recognize the person at a glance, infringement is likely.
Where AI fan work stands today: the tech can shoot it, the platforms can carry it, and the rights have not been discussed. Those three moving out of step is what turns a hit into a dispute.
I am not going to rule on it for anyone. I put this layer in because it decides how far work like this travels. The 30 million views prove one thing. What got pried open is whether you can shoot it. Whether you can show it was never the locked door.
One person, one computer, a few screens: that really is enough now to make something tens of millions of people wait for. What is left is out of the creator's hands.
Related articles
The Jensen Discount: What Nvidia Loses When Its Founder Walks Out
Nvidia lost roughly $600 billion in market value when its founder stepped down. The earnings did not change. The multiple did, and that says where the value really lived.
Chinese text in video always turns to mush. Dreamina handed the typesetting to code
Dreamina launched Motion Studio on November 10. It no longer generates frames; the model writes code that lays out text, shapes and pacing. Chinese titles stop blurring, but whether the content is right is still on you.
A community spreadsheet of taps that turns off the ads inside Chinese Android builds
A no-root tool that switches off the ads Chinese phone makers ship inside the system UI. Its method is a hand-written tap script per phone model, maintained by volunteers every time a ROM moves. It works, and the maintenance is the point.
GPT Image 2.6 held the token price and raised the bill
GPT Image 2.6 kept every token rate from 2.5 and still costs more per image, because more of the work now happens in output tokens. Eleven days into the ramp, the pricing structure says more than the leaderboard does.