Prompt Craft for Video

The five-part prompt formula that separates usable clips from slop.

Video tutorialTrack course · Youri van Hofwegen

AI Video Creation — Beginner to Advanced

Speed

0.75x1x1.25x1.5x2x

Skip intro Next lesson

Watch 00:00 · 2 checkpoints Watch on YouTube More on this topic

Transcript & captions .txt

9/9

1. 00:00Prompt Craft for Video. The five-part prompt formula that separates usable clips from slop.
2. 00:05A video prompt has to describe a subject, an action, an environment, a camera and a look.
3. 00:12Missing any of these means the model invents it — and the model's defaults are boring.
4. 00:18Prompt formula (js): const prompt = […
5. 00:23One action per clip. Two actions produce mush or a cut you didn't ask for.
6. 00:29Describe what IS in frame, not what isn't — negatives are weakly obeyed.
7. 00:34Front-load the subject: early tokens carry more weight.
8. 00:39Reuse the same LOOK block across every shot for visual consistency.
9. 00:44Note: Keep a prompt library. Your grade block, camera block and character block should be copy-paste constants across a whole project.

A video prompt has to describe a subject, an action, an environment, a camera and a look. Missing any of these means the model invents it — and the model's defaults are boring.

const prompt = [
  'SUBJECT: a lone courier in a rain-slick chrome jacket',
  'ACTION: walking slowly toward camera, glancing over her shoulder',
  'ENVIRONMENT: neon alley, wet asphalt, steam vents, night',
  'CAMERA: 35mm, slow dolly-in, shallow depth of field',
  'LOOK: cyberpunk teal and magenta grade, volumetric haze, film grain'
].join('. ');

• One action per clip. Two actions produce mush or a cut you didn't ask for.
• Describe what IS in frame, not what isn't — negatives are weakly obeyed.
• Front-load the subject: early tokens carry more weight.
• Reuse the same LOOK block across every shot for visual consistency.