How to write AI video prompts that work

Start with one subject and one action

Most weak AI video results come from prompts that try to do too much. A clip lasts a few seconds, so give it a single subject and a single action: a fox leaping through snow, a bottle rotating on glass, a drone rising over a coastline.

Use the five-part formula

Write your prompt in this order: subject and action, setting, camera, lighting, style. For example: a lone astronaut crossing red dunes, an alien desert with drifting dust, slow drone shot gliding forward, low golden sunlight, cinematic film look.

This order matches how models read prompts: what happens first, how it is filmed second, and how it looks last.

Be concrete, not poetic

Replace abstract words with things a camera can see. Instead of a beautiful, emotional scene, write warm light on a woman's face as she looks out of a rain-streaked window. Specific nouns, materials and light direction give the model something to render.

Set the format and keep clips short

Choose aspect ratio and duration before you generate. Vertical 9:16 suits Shorts, Reels and TikTok. Three to eight seconds is the most reliable range for coherent motion.

Iterate one change at a time

If the result is close but not right, change one element, such as the camera move or the lighting, and generate again. This shows you which words control which parts of the output, and you will build a personal library of prompts that work.