How to write an AI video prompt
A step-by-step pattern for video prompts: subject, action, camera, light and length, plus what to leave out and how to check a draft before you spend credits.
Step by step
Lead with the subject
Start with who or what is in the shot, in a few concrete words, for example a lone hiker or a glass bottle of perfume.
Add the action with a motion verb
Say what the subject is doing. Motion verbs such as sprinting or unfurling give a model more to work with than static adjectives such as fast.
Name the camera movement
State the move outright, for example slow dolly in, static wide shot or handheld tracking shot, instead of leaving it implied.
Set the light and mood
A few words such as golden hour or harsh fluorescent light anchor the look more than a pile of style adjectives.
Keep it short
Aim for one or two sentences, under about 60 words. Detail past that is usually ignored.
Say what you want, not what to avoid
Instructions such as no blur or not cartoonish rarely work in the main prompt. Rephrase them as the thing you do want.
A worked example
Draft: A woman walks through a city.
Rewritten: A woman in a red coat striding across a rain-soaked crosswalk at dusk, slow tracking shot from the side, neon reflections on wet asphalt, moody evening light.
The rewrite leads with the subject, uses a motion verb, names the camera move and sets the light, and it is still a single sentence.
Let the prompt assistant do the first pass
Turn on Enhance in the video generator and the prompt assistant rewrites your draft using these same patterns when you press Generate. It is aimed at raising your first-try success rate. It does not guarantee a result, so look at the output and re-roll with a tweaked prompt if it is not right.
Use a style preset for the look
If you want a consistent look across several clips, a style preset adds a ready-made prompt recipe to whatever you describe. You can browse every recipe, with the exact text, on the styles pages.
Check the cost before you generate
The exact credit cost for your model, resolution and length is shown before you run anything. Test an idea at the cheapest resolution first, then re-generate the keeper at higher quality.
Frequently asked questions
How long should an AI video prompt be?
One or two sentences, under about 60 words. Most video models ignore detail past that, so a short concrete prompt tends to beat a long one.
Should I use negative instructions like no blur?
Not in the main prompt. Models have no reliable way to act on what to avoid, so describe what you want instead, for example sharp focus rather than no blur.
Does the same prompt work on every model?
The pattern is general and a good starting point on any model, but each model reads prompts a little differently. Run the same prompt through two models in the picker and keep the result you prefer.
Keep reading
Try it yourself
Sign-up is free, no card required, and the exact cost is shown before every generation.
Get started free