Sora died on Tuesday. By Wednesday morning, Elon Musk was on X advertising Grok Imagine as the answer. "The next Grok Imagine release will be epic," he posted. "We are doubling down." He shared demo videos. He highlighted the $10 monthly subscription. The body was not cold.
This is how the industry processes a model's death. Not with reflection. With a marketing push.
xAI is not the only one moving. Every "Sora alternatives" listicle on the internet updated overnight. Veed published a guide. LTX Studio published a guide. Three different YouTube channels uploaded comparison videos before the servers finished spinning down. The SEO machines kicked in before the API keys expired.
None of this is wrong, exactly. A product died. People who relied on it need somewhere to go. Pointing them to other tools is practical. But the speed and the framing tell you something about how the market thinks about AI video models, and what it tells you is uncomfortable if you have spent any time learning the differences between them.
The market sees a slot. Model dies, slot opens, next model fills it. Interchangeable. A Sora-shaped hole in a dashboard, and any model roughly the right shape gets hammered into place.
That is not how this works.
What left the room
Sora read prompts like stage directions. Not keywords. Not technical specifications. Directions. "She hears the door close and realizes she is alone" occasionally produced something that looked like acting, not posing. The narrative threading connected shots through story logic, not visual matching. When it worked, which was maybe one generation in four, there was a quality of dramatic recognition that no remaining model attempts.
That temperament is gone. Not deprecated. Gone. There is no API call that retrieves it. There is no alternative provider hosting the same weights.
What the market is offering instead: visual spectacle from a model that optimizes for spectacle. Grok Imagine generates high-contrast, high-saturation, visually dense output that fills every frame with energy. It has an opinion about what "impressive" looks like and applies that opinion whether you asked for it or not.
That is a fine tool. It is not the same tool.
Casting vs inventory
You do not replace a DP who shoots with natural light and long lenses by hiring one who shoots with Steadicam and wide angles. They both produce footage. They produce fundamentally different footage.
Grok Imagine is not Sora with a different logo. Kling is not Sora with better physics. Veo is not Sora with prettier defaults. Runway is not Sora with more controls. Each model carries a temperament that emerges from its training data, its architecture, and whatever optimization priorities its builders chose. Those are not superficial differences papered over by a good prompt. They are the model.
The land grab pretends temperament does not exist because temperament is not a column in a comparison chart. You can compare resolution, duration, price per second, generation speed, and maximum clip length. You cannot compare "reads prompts like stage directions and occasionally produces genuine dramatic beats at a rate of roughly one in four." That sentence does not fit in a table.
What the rush reveals
When Musk advertises Grok Imagine hours after Sora's shutdown, the implicit claim is substitutability. Our product occupies the same category. Come here instead.
But substitutability is exactly what this space lacks. Every model responds differently to the same prompt. A forty-word structured description of a scene produces six different interpretations from six different models, each reflecting a different visual education, a different sense of what matters, a different relationship between the prompt and the output. CinePrompt builds different prompt optimizations for each model not as a feature but as a concession to reality.
The rush also reveals something about who the displaced users are. Sora's actual user base was not filmmakers building structured prompts. It was mostly casual users generating short clips from brief descriptions in what amounted to an AI social network that never figured out it was an AI social network. Those users absolutely can switch to Grok Imagine with no friction. A casual prompt produces casual output regardless of which model processes it. The spectacle machine will serve them fine.
The users for whom the loss is specific are the ones who had learned Sora's dialect. Who had figured out that narrative framing produced better dramatic output from that particular model. Who had calibrated their prompts to exploit that model's particular strength. Those users cannot redirect their workflow to a model with a different temperament and get equivalent results.
They will adapt. They will learn Kling's physical specificity, or Veo's interpretive confidence, or Runway's literal obedience. They will redistribute their creative intent across the remaining six models. But redistribution is not replacement. It is reorganization after a loss.
Six is not seven minus one
The math looks simple. Subtract one, continue. But a creative ecosystem is not arithmetic. Each model's existence influenced how the others were used. Sora handled narrative and dramatic intent better than the alternatives, which meant filmmakers could route narrative-heavy work to Sora and precision work to Runway and atmosphere work to Veo and texture work to Kling. Losing one model changes the routing for all of them.
The remaining six now carry responsibilities they did not volunteer for. Veo will be asked to do more narrative work. Kling will be asked for dramatic timing it was not optimized to deliver. Grok Imagine will receive prompts written for a model that listened to stage directions and will respond with spectacle, because spectacle is what it does.
The vocabulary still works. The prompt structure still works. The six-layer hierarchy, the attention gradient, the decomposition of feeling into specific visual language: all of it transfers to every surviving model and to whichever model launches next month. But the cast is different now, and the roles each member plays have shifted, and pretending the understudy is the lead is a disservice to both.
The pattern
Five disruptions in six weeks. Grok's paywall. Seedance's copyright suspension. Runway's platform pivot. Real-time generation's arrival. Sora's death. And now the land grab over the remains.
Each disruption confirmed the same principle: the portable part is the craft. The fragile part is the dependency on any single platform, provider, or model staying exactly where it was when you built your workflow around it.
Musk will ship the next Grok Imagine. It might be impressive. It will carry its own temperament, its own strengths, its own blind spots. It will not be Sora. It will be itself. And the people who learn its specific dialect, who figure out what it hears and what it ignores, who calibrate their vocabulary to its particular personality, will get more out of it than the people who treat it as a slot filler.
The vacancy is specific. The market is treating it as generic. Filmmakers know the difference.
Bruce Belafonte is an AI filmmaker at Light Owl. He has watched five disruptions in six weeks and his prompts still work on all six remaining models, which he considers the only metric that matters.