Writing Prompts for Video — Motion, Camera Movement, and Scene Description
Video prompts need one more dimension than image prompts
Everything that makes a good image prompt — subject, style, lighting (see writing effective image prompts for that side of it) — still matters for video. But a video prompt also needs to describe what changes over the duration of the clip, which an image prompt never has to. A prompt that only describes a static scene tends to produce minimal, uncertain motion; a prompt that describes what's moving and how gives the model something concrete to animate.
Describe the motion, not just the subject
Say what's actually moving, and how:
- "the waves roll in slowly," rather than just "ocean"
- "steam rises gently from the cup," rather than just "cup of coffee"
- "the dog runs across the frame from left to right," rather than just "a dog"
Vague motion language ("something happens," "it moves around") gives the model little to commit to. Specific, physically plausible motion tends to render more convincingly than something elaborate or physically unusual.
Describe the camera, separately from the subject
Camera movement is its own axis, independent of what the subject is doing:
- "slow pan left to right"
- "camera pushes in close on the subject's face"
- "static shot, locked off" (if you want the camera to simply not move)
- "aerial drone shot descending toward the scene"
If you don't specify, expect a reasonably static or simple default camera behavior — call it out explicitly if camera movement matters to the result you want.
Scene description: set the stage before the motion starts
Establish the setting the way you would for an image prompt — location, time of day, lighting, style — then layer the motion and camera direction on top of it. A prompt that reads naturally as a short, concrete description works better than a list of disconnected keywords:
"A wide shot of a quiet street at golden hour, a cyclist rides past from right to left, camera holds steady, warm cinematic lighting."
Keep expectations realistic about duration
Available clip durations are short and discrete (see how to generate your first AI video clip for how duration options are presented), so a prompt describing a long, multi-beat sequence of events won't fit into a single short clip — scope your prompt to what can plausibly happen in the duration you've chosen. If you need a longer story, that's what chaining segments together is for — see how video chaining works.
Starting from an existing image
If you're generating a video from a specific starting image (image-to-video) rather than from scratch, your prompt's job changes slightly — you're now describing what happens to that image over time, rather than describing the whole scene from nothing. Keep the subject and composition consistent with what's actually in the source image rather than contradicting it.
If the motion doesn't look right
If a generated clip comes back static or the motion looks off, the most common fix is making the motion and camera language in your prompt more explicit rather than more elaborate — a short, specific instruction about what moves and how the camera behaves tends to outperform a longer, more descriptive paragraph that never actually says what's in motion.
Try it yourself
Take a scene you'd normally describe as a still image and add one motion phrase and one camera phrase to it — regenerate both versions and compare. It's the clearest way to see how much those two additions actually change a video result.