How to write prompts for AI images and video that come out right the first time

Use a five-part structure

Good prompts answer five questions in plain words: what is the subject, where is it, what is the light, where is the camera, and what should it feel like. "A ceramic coffee cup on a wooden cafe table, warm morning light from a window, close-up at table height, clean commercial photography" answers all five.

Write the subject first and the style last. Models weigh the start of a prompt most, so the thing you are selling belongs at the front.

Say what you want, not what you do not want

"No people" works for simple exclusions, but long lists of negatives confuse models and often pull the unwanted thing in. Describe the scene you want precisely enough that the unwanted thing has no room: "an empty table set for two" beats "no people, no clutter, no mess".

Text inside images is still unreliable. If you need a price, a name or a headline, add it afterwards in your design tool rather than asking the model to spell it.

Prompts for video are about time

A video prompt needs one extra ingredient: what changes. Describe a single camera move and a single motion - "slow dolly-in as steam rises from the cup" - and keep the rest of the scene stable. Short, specific motion reads as cinematic; many simultaneous changes read as noise.

If you already have a good image, animate it instead of describing it again. See the image-to-video guide.

Iterate cheaply, finish once

Draft with the Fast image setting, which is the cheapest option, until the composition is right. Then spend premium credits once on the final. Recipes skip the blank page entirely: they open the Studio with a proven prompt structure and ask you to replace only the bracketed parts.

Try it with a recipe

Keep exploring