How to Write Text-to-Video Prompts That Actually Work

Why your prompt matters more than you think

An AI video model can only render what it understands. Vague prompts like "a cool city video" leave the model guessing about the subject, the motion and the mood — so you get generic results.

A good prompt reads like a short shot description from a director: what we see, what moves, and how the camera captures it.

A simple five-part prompt structure

Build every prompt from these five parts, in this order:

• Subject — who or what is on screen ("a red vintage motorcycle").
• Action — what happens ("speeding along a coastal road").
• Camera — how it is filmed ("low tracking shot from the side").
• Lighting — the time and mood ("golden hour, long shadows").
• Style — the overall look ("cinematic, 35mm film grain").

Before and after

Before: "motorcycle on a road".

After: "A red vintage motorcycle speeding along a coastal road, low tracking shot from the side, golden hour with long shadows, cinematic 35mm film grain."

The second prompt gives the model a clear subject, motion and camera path, which leads to smoother, more intentional clips.

Common mistakes to avoid

Most weak results come from a handful of habits:

• Packing several scenes into one prompt — generate one shot at a time.
• Describing feelings instead of visuals ("an inspiring video").
• Contradictory styles, such as "photorealistic cartoon".
• Forgetting the camera, which often produces static, flat shots.

Try it yourself

Pick one idea, write it with the five-part structure, then change a single part at a time — the camera or the lighting — to see how each word shapes the result.