Video
How do I write a good video prompt?
Video prompts work best when they read like a short shot description: who or what, doing what, and how the camera sees it.
Subject, action, camera
Build the prompt in three parts. The subject: a red sports car on a coastal road. The action: it drives toward the camera and passes on the left. The camera: low angle, slow pan following the car. Adding pace — slowly, in one smooth move — helps the model keep the motion stable.
Name the camera move
Use familiar film terms: push in, pull out, pan left, tilt up, orbit around, tracking shot, handheld, static shot. One camera move per clip is more reliable than combining several.
Set light and mood
Describe light and atmosphere in a few words: golden hour, overcast, neon at night, soft studio light, mist. With image-to-video, keep these consistent with the starting image so the clip does not shift in colour.
Keep it short
A clear two- or three-sentence prompt usually beats a long list of details. Models struggle with many separate actions in five seconds. If you need a sequence of events, split it into several clips.
Change one thing at a time
When a draft is close, change one part of the prompt per attempt — only the camera, or only the action — so you can see what made the difference. Draft at the shortest length and lowest resolution while you iterate.
FAQ
- Should I describe the scene in image-to-video?
- Only briefly. The image already shows it. Spend the prompt on motion and camera.
- Can I ask for dialogue or sound?
- Only on models that generate audio. Describe the sound or the spoken line in the prompt on those models.
- Why does the video ignore part of my prompt?
- Too many instructions compete. Remove the least important ones and try again.