How to Write AI Video Prompts That Actually Work

2026-05-31 · 6 min read

The difference between a bad AI video and a great one is almost always the prompt. Here's the structure that works across every frontier model.

The six-part prompt formula

Every strong AI video prompt names six things: frame, subject, action, lighting, lens, motion. Miss any of them and the model fills the gap with whatever it 'feels like' — which is usually generic.

Frame

Wide shot, medium close-up, extreme close-up, low angle, high angle, drone overhead, POV. Name the camera position first. The model can't read your mind about how 'close' the camera should be.

Subject and action

Be specific. Not 'a man' — 'a 30-year-old man in a black overcoat'. Not 'running' — 'sprinting flat-out, arms pumping, breath visible in cold air'.

Lighting

Golden hour, blue hour, neon practicals, soft window light, harsh midday sun, candle-lit interior. Lighting determines mood more than any other variable. Always name it.

Lens

Anamorphic, wide angle, telephoto, macro, fisheye, 35mm, 85mm. Lens choice changes everything — depth of field, distortion, intimacy.

Motion

Static, dolly-in, dolly-out, tracking shot, crane, handheld, slow-motion, time-lapse. If you don't name motion, the model defaults to a gentle drift that often kills the shot.

A full example

Bad: 'a guy running through a city'.

Good: 'low-angle tracking shot of a 30-year-old man in a black overcoat sprinting through a neon-lit Tokyo alley at night, rain on the pavement reflecting pink and cyan neon, anamorphic lens with subtle flare, slow-motion 60fps'.

Same idea, completely different output.