Sora and the Empire of Absurdity
When Sam Altman unveiled Sora, he talked about a “breakthrough in modeling the physical world.” The neural network was supposed to generate realistic videos and become a tool for filmmakers. But the internet had other plans, and the platform shut down six months after launch.
On March 26, 2026, OpenAI announced the closure of Sora. Disney simultaneously canceled a $1 billion investment agreement. Less than six months had passed since Sora 2’s launch.
Here’s what actually happened — verified data only, no fiction.
A study of nearly a thousand random videos from Sora 2 during its peak popularity revealed a grim picture for OpenAI. Most of the content being created consisted of memes and outright absurd clips. Another significant portion were remixes — reusing other people’s videos with modifications.
Many videos fell into both categories at once: users took someone else’s absurd clip and made it even more absurd.
Commercial or branded content was scarce. Almost equally rare were videos created to test the neural network’s capabilities. The vast majority of users didn’t come to Sora to make movies or advertise products — they came to fool around.
And the audience loved it. Remixes captured the overwhelming majority of all views on the platform. Original content — the very thing Sora was supposedly built for — was practically unwatched.
Videos that were both memes and remixes performed especially well. Such clips garnered thousands of times more views than everything else. The median audience for this “combo” ran in the hundreds of thousands, while serious videos barely scraped past fifty views.
Researchers found a consistent pattern: the more seriously a user approached their task, the fewer people saw it.
Short prompts of just a few words got dozens of times more views than detailed scripts. Someone would type “Skarmy” or “Hippo” and rake in millions of views. Meanwhile, those who painstakingly specified camera angles, lighting, and atmosphere were left in glorious solitude.
Long, professional prompts did have higher engagement — those who found them liked them more. Quality was noticed. It’s just that almost nobody found them.
Prompts with cinematic terms like “8K,” “cinematic,” or “anamorphic” were rare and garnered pitifully few views. The ability to write complex prompts turned out to be the most useless skill on Sora.
Here’s what millions of people actually watched.
The leader in remix count was the prompt “Bro moves like a 3-ton Roomba” — about a person moving like a three-ton robot vacuum. It was remixed tens of thousands of times, and the original garnered nearly two million views.
Close behind was the mysterious “Skarmy” — a word no one could explain, but Sora dutifully generated skeleton armies dancing in the desert under it. Over one and a half million views and tens of thousands of remixes.
“British Chicken strapped to a drone rapping about not wanting to be fried” — nearly one and a half million views.
“Hippo” — just a hippopotamus. A million views.
“Squirrel eating spicy ramen” — a million views.
“Make it a pizza” — over seven hundred thousand views.
What do all these prompts have in common? They’re short, absurd, have zero barrier to entry, and are perfect for remixing. Not one exceeds twenty words. Not one requires any knowledge of cinema.
The most mentioned character in Sora prompts wasn’t a Hollywood star or a famous director — it was OpenAI’s CEO. Users massively generated videos of Sam Altman doing strange things: playing guitar with Pikachu, “announcing” Sora 3 with made-up features, ending up in absurd situations.
The total views for videos featuring Altman ran into the millions. For comparison, the second most popular character, YouTuber Jake Paul, garnered a thousand times fewer views.
OpenAI built a tool for filmmakers, and its CEO became the biggest meme on his own platform.
If you think serious creators didn’t try — they did. Nearly a quarter of the videos in the study fell into the “cinema/art” category. The quality was high, and they received more likes from those who saw them.
But the problem was that nobody saw them.
Most of these videos got fewer views than the platform average. An ultra-realistic noir clip with a meticulously crafted prompt of nearly two thousand characters received a paltry seventeen views and zero remixes. A commercial for a bookstore — fifty-two views. A horror trailer in 8K — thirty-eight.
On Sora, the most useless skill turned out to be the ability to write complex prompts.
If Sora had been just another TikTok, nothing terrible would have happened. On TikTok, Instagram, and YouTube, the cost of creating content is borne by the user. The platform loses nothing when someone films another dance or unboxing video.
Sora flipped this logic. Every generated meme cost OpenAI real money.
According to expert estimates, generating a single ten-second video on Sora cost the company about one and a half dollars. Computing power was the main expense.
Now do the math. The prompt “Bro moves like a 3-ton Roomba” was remixed tens of thousands of times. Each remix was a new generation, another dollar and a half spent. On just this one meme, OpenAI could lose tens of thousands of dollars.
Sora’s daily GPU expenses reached fifteen million dollars. Peak monthly revenue? Just over half a million. A Sora researcher publicly admitted: the economics were completely unsustainable.
User retention was catastrophic. Out of every hundred people who downloaded the app, only one remained after a month. People came, generated a few absurd videos, laughed, and left. The platform didn’t build a loyal audience — it created a stream of one-time meme-makers.
Sora didn’t die because of bad AI. The technology worked impressively — the neural network demonstrated abilities in 3D spatial consistency and modeling simple physical interactions.
Sora died because it gave people exactly what they wanted: an endless generator of funny, meaningless videos.
Short prompts gathered millions of views. Millions of views meant millions of generations. Millions of generations bankrupted OpenAI.
OpenAI dreamed of a platform where everyone could become a filmmaker. Instead, it got a meme roulette where viral success is determined not by talent but by the absurdity of the prompt.
The Empire of Absurdity didn’t rule for long. But the memes about the three-ton Roomba, poker-playing hippos, and skeleton armies dancing in the desert will remain in history as the perfect example of how the internet turns any technology into a joke machine.