How did an accountant create a video with 7.4 million views using AI by featuring dragons as a dish?

"You must have some trendy catchphrases to keep people engaged."

In late October, a video titled "Cooking an Ancient Mosasaur into Six Dishes (Part 1)" went viral on Bilibili. It reached 7 million views just three days after its release. What's crucial is that this is a completely AI-generated video, lasting 6 minutes and 23 seconds. According to past patterns, it's quite rare for a video with these two factors to gain such high traffic.

After all, many people are "repelled" by AI-generated content. However, among the nearly 5,000 comments under this video, most are expressions of shock at both the rapidly improving image quality of AI and the author's control over AI.

This video is indeed different from most previous AI videos. It's not about cutting stones or cats cooking. Instead, it features chefs from several countries in a cooking competition, with the ingredient being a mosasaur. Yes, the mosasaur - an ancient creature that went extinct 65 million years ago - cooking it into dishes, have you ever seen that?

The video starts with a grand scene of foreigners sawing meat and chopping ribs taller than a person. The camera zooms in, rotates, and switches rapidly. As the characters appear, conflicts break out. The plot is compact and full of tension, instantly grabbing the audience's attention.

● The beginning of the video "Cooking an Ancient Mosasaur into Six Dishes (Part 1)" 

The most difficult and important thing is to keep the audience engaged throughout the 6 minutes and 23 seconds. For this purpose, the Bilibili UP master "Huangpu River Salmon" (hereinafter referred to as "Salmon") used many "tactics" -

For example, there are various trendy catchphrases throughout the video. First on the scene is the Indian chef making "Nine-Turn Intestines"; the Chinese chef is "Cynthia" from Shanghai. When she appears, there is background music and a voice - over, like the well - known line "The most high - end ingredients often only require the simplest cooking methods"... Because "Salmon" firmly believes "You must have some trendy catchphrases to keep people engaged."

On top of that, the most basic thing is the AI - generated images that can almost pass for real. "It can't be exactly the same, but I strive for 90% authenticity." So much so that some netizens once argued, "You clearly shot this for real, but you insist it's AI - generated." "Salmon" didn't argue and just said, "I caught the dragon in the South China Sea a few days ago."

It's not easy to present such high - quality images and maintain the consistency of multiple characters. It's said that when generating one of the special - effect scenes, "Salmon" filled an entire page of a document with prompts. And when feeding the prompts to the large - scale model, it usually takes multiple attempts to generate a perfect and usable video clip. It's like opening blind boxes, which is also called "card - drawing" in the field of AI generation.

● The prompts written by "Huangpu River Salmon" 

Currently, most video models can only generate 5 - second or at most 10 - second clips. In theory, this 6 - minute and 23 - second video requires at least 75 5 - second shots to be spliced together. But in fact, "Salmon" generated more than 1,000 pieces of materials for this video, and for some shots, the card - drawing process was repeated over a hundred times.

Under this video, which has already reached 7.4 million views, someone commented, "This is the most amazing AI video I've ever seen since the birth of AI."

Actually, the 28 - year - old "Salmon" works as an accountant by profession. Three years ago, when AI first emerged in the public application field, he quickly followed the trend and started learning. From image generation to 3D modeling and then to animating, it used to take him a month to produce a short clip. Now, he can "make several short clips in a day."

Especially after the Spring Festival this year, he found that he could truly complete a small advertising video entirely with AI. Although it was a bit rough, it could be delivered. He witnessed the rapid progress of large - scale models, so he "made up his mind to get closer to this industry." After all, his childhood dream was to be a director.

So, the success of "Cooking an Ancient Mosasaur into Six Dishes (Part 1)" was not accidental. Apart from the earth - shattering progress of large - scale models in a short period, it was also a well - planned hunt for traffic by "Salmon."

One month before this video, he uploaded three videos themed around cooking dinosaurs at a frequency of once a week, like an A/B test as a prelude -

In the first episode, "Cooking a Tyrannosaurus Rex into Three Dishes," it was simply about cooking, putting the idea into practice, without a main character;

In the second episode, "A Shanghai Chef Cooks a Pterosaur into Three Dishes and a Soup," the main character, Shanghai chef Cynthia, was added, and there were trendy catchphrases. The audience's response was intense;

In the third episode, "A Cantonese Cooks a Triceratops for Three Meals a Day," the main character was removed, but elements from "Chibi Maruko - chan" were added, and the effect declined.

After these three episodes, "Salmon" basically grasped the logic of creating a viral AI video. He would deliberately leave some continuity errors in the video. For example, Cynthia often puts her hand above a boiling hot pot, just to get people talking. And the lively discussions will push the video into a wider traffic pool.

Of course, there is also an unignorable change in the external environment - the significantly increased acceptance of AI - generated content by major content platforms and audiences. Most of the popular videos on Bilibili contain AI elements, and there has even been an AI - generated NPC singing competition; Kuaishou also claims, "We hope to empower more creators on the platform with AI and look forward to more high - quality long - form AI content being produced on the platform."

One month ago, the AI video generation models Sora2 and Veo3.1 started to compete fiercely, and their capabilities soared. The sensitive "Salmon" clearly made a judgment - the time was ripe.

On November 9th, "Salmon" updated "Cooking an Ancient Mosasaur into Six Dishes (Part 2)." By noon on the 13th, the views had already approached 2.34 million, and his Bilibili followers had increased to 71,000. Cooperation invitations have been pouring in from the background. At present, his transformation into a full - time AI designer is on the right track.

At the beginning of November, the "Future Human Laboratory" had a conversation with "Huangpu River Salmon," discussing the ingenuity, production, and future plans behind this viral video. The following is the content of the conversation, sorted out and published -

"I thought it would be good, but I didn't expect it to be this good."

Future Human Laboratory: The first three videos were basically uploaded once a week. There was a 16 - day interval before "Cooking an Ancient Mosasaur into Six Dishes (Part 1)." Where did you spend most of the time?

Huangpu River Salmon: The production didn't take too long. Most of the time was spent on sorting out the script and prompts in the early stage. I silently thought about this script in my mind for about a week. Who would win, who would lose, what each person's personality was like, what dishes they would make, and how the final competition would be... After I conceived the general framework of the whole story, I started working on the two characters (the Indian chef and Cynthia) in the first episode. Once I started, I managed to finish it in a little over a week, of course, staying up late every day.

Future Human Laboratory: Did you expect this video to be this popular?

Huangpu River Salmon: I thought it would be good, but I didn't expect it to be this good. I estimated that it would reach at most 1 million views. I posted it at 10:30 p.m. and then went to sleep without paying attention. The next day, it had already exceeded 1 million views. My private messages were flooded. There were some cooperation offers, but more people were asking how I made it and asking me to share. They said it seemed like the AI I was using was a different product from theirs.

Future Human Laboratory: I once talked to a blogger who specializes in AI product reviews. We think the reason this video went viral lies more in the arrangement of the whole story, the dense trendy catchphrases, and the rapid switching of images. I noticed that you switch the shot almost every 1 - 2 seconds. The AI presentation is actually secondary.

Huangpu River Salmon: Yes. In my understanding, the videos that people are willing to watch are either abstract and funny, for entertainment purposes. The other type is well - produced. The images are made by AI, but there shouldn't be too many mistakes. I strive to make the AI - generated images more realistic. It can't be exactly the same, but I aim for 90% authenticity. The third aspect is to incorporate some trendy catchphrases on the Internet, which is like riding on the wave of popular topics. Combining these three factors, I think, is the reason for the success of this video.

You must have some trendy catchphrases to keep people engaged. If it's just about cooking, I think it would get at most a few hundred thousand views.

Future Human Laboratory: You really understand how to attract traffic. Are there any other designs in the video to attract traffic?

Huangpu River Salmon: I would deliberately add some things that seem like flaws to the real - looking AI - generated images. For example, Cynthia often likes to put her hand above a boiling hot pot. Then people will say there is a continuity error. Because if there are a lot of comments on this part, it may be pushed to the next traffic level. So I will deliberately add some things that are likely to trigger comments and discussions.

 ● Cynthia puts her hand above the pot

Future Human Laboratory: You're so sly.

Huangpu River Salmon: This scene is actually quite interesting. People think it's like "Iron Sand Palm" or something like that. Actually, I have some well - made footage where she doesn't put her hand on the hot pot. But after some consideration, I kept this flawed shot.

Future Human Laboratory: So you deliberately left some room for people to discuss.

Huangpu River Salmon: Yes. Including the scene of frying black truffles, many people discussed this part. Some said that black truffles lose their flavor when fried, while others said they can be fried. Some also questioned the combination of chanterelles and black truffles. Many professional chefs said that it can indeed be done this way, and some even said they wanted to replicate it. Actually, I just wondered if frying black truffles would be delicious and randomly added a few characteristic Yunnan mushrooms. I threw in these absurd things to let netizens discuss.

Future Human Laboratory: How did you discover these patterns?

Huangpu River Salmon: It's also based on my previous works. When the chefs were cooking, sometimes I didn't notice that their hands would pass through the open flame. But at that time, I noticed that as soon as such a scene appeared, people would start posting bullet comments saying there was a continuity error or joking about "Iron - Handed Chef." So I thought that adding some appropriate AI - related continuity errors later would also let people know it's not a real - shot video.

Because some people would send me private messages saying, "You clearly shot this for real, but you insist it's AI - generated," or "You clearly mixed some real - shot footage with a small amount of AI - generated content and claimed it was all AI - generated to deceive people." I have two purposes. One is to make people laugh, and the other is to prove that it's all AI - generated.

Future Human Laboratory: Do you feel happy when someone says your video is real - shot?

Huangpu River Salmon: It can be seen as a recognition of my ability, so I'm a little happy. I would also reply to them, "Yes, I shot it for real. I caught the dragon in the South China Sea a few days ago or just captured it in Guangzhou."

Passing for Real

Future Human Laboratory: How did you come up with the idea of cooking dinosaurs?

Huangpu River Salmon: At that time, I saw some domestic AI works. Everyone was doing the same things, like cutting stones with a knife or cats cooking, creating an ASMR - like effect. After one became popular, others started copying.