01
What separates a prompt that works from one that does not
Concrete visual detail beats adjectives, every time. “A beautiful beach scene” gives a model nothing to decide; “slow dolly-in at golden hour, low camera, wet sand reflecting the sky, one figure walking away from the lens” gives it four decisions already made.
So each prompt this returns names four things explicitly: the shot and camera movement, what the subject is doing, the lighting, and the mood. That is the structure most text-to-video models respond to, and writing it out is most of the craft.
02
This is useful even if you never use Veedio
The prompts come back as plain text you can paste into whatever model you are already paying for. We are not going to pretend our own pipeline is the only destination for them, and a tool that is only useful if you buy something is not a free tool.
What our pipeline does with a prompt is different anyway: it writes a script, generates a frame per scene, voices it and captions it. If what you want is one five-second clip from a frontier model, use the frontier model — this page will still write you a better prompt for it.
03
What we do not claim about other models
We have not benchmarked Veo, Sora, Kling, Seedance or Runway against each other and we are not going to publish a table pretending we have. Their prompt handling differs, their guidance changes with each release, and a page asserting otherwise would be out of date within a month.
What is stable enough to be worth saying: every one of them rewards a named shot, a specific action, a described light source and a stated mood, and every one of them does worse with a list of adjectives. That is what these prompts are built from.