Writing prompts for AI video: 10 rules

2026-07-13

Same model, same credits — yet one person gets a cinematic masterpiece and another gets chaotic mush. The difference is the prompt. These 10 rules will fundamentally change your results.

Anatomy of a prompt

A working prompt follows this structure: [Subject] + [Action] + [Environment] + [Camera] + [Lighting] + [Style]. Example: "An old sailor (subject) turning the boat's wheel (action) in a stormy sea (environment), camera pulling back from close-up (camera), gloomy grey light (lighting), cinematic drama (style)".

The 10 rules

Be specific — not "beautiful landscape" but "sunrise over a snowy mountain peak, clouds in the valley".
One scene, one action — don't write a script that can't fit in 5-10 seconds.
Learn camera language — "dolly in", "pan left", "aerial shot", "tracking shot", "static".
Describe the light — "golden hour", "overcast", "neon glow", "candlelight" — lighting defines the mood.
Set the motion speed — "slow motion", "timelapse", "gentle movement".
Give a style reference — "shot on 35mm film", "documentary style", "Pixar animation style".
Don't write negatives — "don't make it bad" doesn't work; describe what you want, not what you don't.
Iterate — the first result is a sketch: keep what works, refine what doesn't.
Match prompt to model — flagship models handle long detailed prompts; faster models prefer shorter, punchier ones.
Learn from others — grab the prompt of any video you like in the community gallery via Remix and adapt it to your subject.

Ready-to-use examples

Nature: "Aerial shot of turquoise waves crashing on black volcanic beach, morning mist, cinematic, slow motion". Portrait: "Close-up of an elderly craftsman carving wood, warm workshop light, shallow depth of field, documentary style". Product: "Luxury watch rotating on dark reflective surface, dramatic side lighting, macro details, premium commercial style".

Start with these examples — copy, swap in your subject, generate.