How to write a video prompt that actually renders

Most prompts that come back looking wrong are not a model problem. They are missing one of four ingredients: what is in frame, how the camera moves, how the scene is lit, and how fast the shot feels. Write those four and the output stops being a lottery.

Subject first, adjectives later

Start with the noun that must appear. “A paper boat in a rain gutter” beats “something nostalgic and rainy” because the model can draw a boat. Style words are seasoning — add them after the subject is settled.

Name the camera move

“Slow dolly in”, “handheld follow”, “static wide” all produce visibly different results. If you say nothing, you get the average of every training clip, which usually looks like a slow push forward. Say it explicitly.

Light is a lever, not decoration

“Overcast morning”, “single hard key from the left”, “neon spill from a shop window” change the grade before any stylistic instruction does. One light description is worth three style adjectives.

Pace lives in the duration

A four second clip cannot hold a three-beat story. Match the idea to the length: one action per four seconds, one establishing shot for anything longer. If you need a sequence, generate the shots separately and cut them together.

A worked example

A paper boat drifting down a rain gutter at night, slow dolly following it, neon spill from a shop window, shallow focus, 35mm.

Subject, camera, light, and a pace hint in one sentence. That is the whole technique.