ChatGPT put out a new image model recently, and it took me straight back to a year ago when the first one came out and the entire internet started Ghibli-fying everything.

It was everywhere…

Then at the end of 2025, Gemini's Banana model did it again with a second wave.

But AI video is still waiting for that moment…

Sora came close when it launched, there was real buzz.

But OpenAI shut down the consumer app in March 2026, just six months after Sora 2, which tells you something.

OpenAI has sort of paused AI video efforts too.

So why hasn't the viral moment happened yet?

The tech actually got good. Just not in a shareable way.

The gap between early 2025 and early 2026 is actually real and significant.

Clip length went from 4-10 second bursts to coherent two-minute generations.

Native audio arrived and it actually works: sound effects that match what's on screen, lip-sync that doesn't make you cringe.

Resolution crossed broadcast quality, with Veo 3.1 hitting true 4K at 60fps in January 2026.

All genuinely impressive but none of it translates into something you send your group chat.

Video is just a harder medium.

Most people have a rough gut sense of whether an image looks good.

Fewer people have a trained eye for what makes a video actually work: pacing, camera movement, whether the motion feels intentional or just random.

A bad AI image is immediately obvious but a mediocre AI video is off in a dozen subtle ways that are harder to name, and harder to fix by just re-running the prompt.

Images are also a finished product on their own. A video almost always needs to fit into something larger. That adds a layer of complexity before you've even opened the tool.

And honestly, it just takes too long.

Getting something genuinely good out of an AI video tool right now requires time and iteration that most people aren't going to put in.

Generation times are still measured in minutes. The prompting required to get coherent motion, consistent subjects, and the right camera angle is significantly more demanding than typing "Ghibli style."

You can get to something impressive, but you have to really work for it.

The Ghibli moment worked because the barrier was zero and the payoff was immediate. AI video hasn't closed that gap yet.

What needs to happen for the viral moment to land

A few things need to come together at the same time.

#1 Generation speed needs to feel near-instant

The moment you're waiting two minutes to see your result, you've already lost most people.

#2 The output needs to be consistently good on a short, simple prompt without the user needing to understand camera terminology.

The bigger issue, though, is who's going to get us there. The major AI labs -- OpenAI, Google, Anthropic -- have largely treated video as a side project.

OpenAI shut down the Sora consumer app.

Google buries Veo inside Gemini.

Anthropic never offered AI video (only now with Claude Design)

None of them are betting the company on it… What actually moves this category forward will probably come from dedicated video models and the companies building exclusively around that problem, not from a general-purpose AI lab adding video as another feature on the list.