How to Write AI Video Prompts That Produce Better Results
A practical structure for subject, action, camera, lighting, style, sound and constraints — with examples you can use right away.
The prompt framework
Seven parts you can mix and match depending on what matters for your shot.
Subject
Name the subject clearly: who or what it is, what it looks like, and where it is.
Action
Describe exactly what happens — the specific movement or event, not just a mood.
Camera
Add a camera movement (pan, zoom, tracking shot, static) to guide how the scene is framed.
Lighting
Mention the light source and time of day — golden hour, neon, overcast — to set the mood.
Style
Add a style word — cinematic, photorealistic, anime, cartoon — to steer the overall look.
Sound
If the model supports audio, describe the specific sound you want, not just 'with sound'.
Constraints
State what shouldn't happen — e.g. 'no extra characters, no text on screen' — to avoid common mistakes.
Weak vs strong prompt
Weak
A cool video of a car.
Strong
A matte black sports car drifting through a neon-lit Tokyo street at night, rain reflections on the asphalt, camera tracking alongside at car height, cinematic, motion blur on the wheels.
Prompts don't transfer identically between models
Prompt structure that works well on one model doesn't always transfer exactly to another — some models weigh camera language more heavily, others respond more to mood and style words. If a result doesn't look right, try the same prompt on a different model before rewriting it from scratch.
Quick checklist before you generate
- Subject is specific, not generic
- One clear action, not several competing ideas
- Camera movement stated
- Lighting or time of day mentioned
- A style word is included
- Sound is described specifically, if you're using audio
- Anything that shouldn't happen is stated explicitly