What generative video still cannot do
Models have got extraordinarily good at the cinematic shot, and are still unusable for the thing most businesses actually need to publish: a fact, held still, on brand.
Watch any generative video reel and you will see the same set of shots: a drone over a coastline, a slow push into a face, a product rotating in studio light. They are genuinely beautiful, and they are all the same kind of thing. They are footage.
Footage is not what most companies need to publish
The video a business needs on a Tuesday is not cinematic. It is a number that went up, a fixture list, a price change, a new opening, a result. It needs:
- Text that is exactly the text. Not text-like shapes. Not a headline with one letter wrong.
- A number that does not move. If it says 4.7% it must say 4.7% in frame one and frame ninety.
- A chart that is the data. Not a chart-shaped drawing that resembles the trend.
- A logo that is your logo, in the right colour, at the right size, inside the safe area.
Every one of those is a hard requirement, and every one of them is where a generative model is at its weakest. The failure is not occasional, it is structural: the model is producing something that looks like the thing, and for footage that is enough, while for a fact it is the one thing that cannot be allowed.
The split that actually works
The interesting answer is not that AI cannot do video. It is that the two halves of the job belong in different places.
A model is excellent at the part that is language: reading the event, deciding what is worth saying, writing the headline, picking the channel, choosing when to post. That is judgement over text, and it is what an assistant is for.
A composition drawn by a designer is excellent at the part that is layout: where the number sits, how the chart animates in, what happens when the headline is long, how the whole thing reflows from 16:9 to 9:16 without a human re-cutting it.
Give the assistant the first half and the template the second, and you get a video that is both worth publishing and safe to publish without watching it first. That is the shape of the thing we build.
