The field notes · October 4, 2026

How to write an AI video prompt

A practical framework for subjects, actions, camera movement, and continuity.

Start with a single scene

A useful first prompt describes one subject doing one main action in one setting. A dancer turning in a rainy street is easier to evaluate than a sequence that jumps between locations. Begin with a short clip and expand after you get a stable result.

Describe motion explicitly

Write what moves and what stays still. “The camera slowly pushes toward the subject while the subject remains still” is more specific than “cinematic scene.” Avoid combining an orbit, a zoom, and a handheld pan in the same short shot.

Add light and framing

Include the shot size, light direction, and format when those matter. A full-body shot helps evaluate dance movement. A close-up helps focus on a product texture. Choose 9:16 in your video tool for a vertical output; mentioning it in the prompt alone may not change the export dimensions.

Use a simple prompt structure

Subject → action → setting → camera → light → continuity constraints. For example: “An unbranded glass bottle rests on stone. Mist passes behind it. The camera makes a slow arc. Soft side light, stable bottle shape, no lettering.”

Review before publishing

Look for warped hands, changing faces, drifting product geometry, and unwanted text. Edit one prompt variable at a time. Keep a record of the model and settings alongside the prompt so successful attempts are easier to reproduce.

Browse the prompt collection to try this framework with a starting point.