How to write a prompt for an AI video model
A video prompt describes motion: what the subject does, how the camera moves, the light and whether there is sound. Write two to four sentences. Kling 3.0 accepts up to 2,500 characters and Veo 3.1 up to 1,000. Looks and colors come from the photo or the references, so do not repeat them in the prompt.
Updated:
The order to write in
Subject and place
One sentence: who or what is in the center and where the action happens.
Motion
What changes during the clip: a step, a turn, a drop falling, a door opening. One main action per shot.
Camera and light
A push-in, a pan or a locked frame. Soft daylight or warm backlight. Naming it stops the model from choosing for you.
Sound
If you want sound, describe it: sea noise, music, a short line. If you do not, say so.
What usually breaks a clip
- Too many actions at once: the model does one and blurs the rest.
- Describing a look that is already in the photo: the face starts to drift.
- Contradictions such as fast and slow in the same sentence.
- On-screen text without quotes: the model writes it as it likes.
Where it is in ClickBait
Frequently asked questions
How long should the prompt be?
A few sentences are usually enough. A short, precise prompt beats a long list of details.
Should I write the prompt in English?
Models understand English best, but they also work with other languages. For exact words and names, English is safer.
What if the clip does not match the prompt?
Remove one extra action, describe the motion more precisely and run it again. Two small fixes usually work better than one huge rewrite.
Should I write technical settings in the prompt?
Format, length and quality are set in the settings, not in the text. Do not repeat them in the prompt.
Do I need to describe the character's looks?
If the character is in the photo or the references, skip the looks. Describe them only when there is no reference.