Every prompt in the Prompt Studio catalog is written as a director’s brief: subject, action, environment, camera, lighting, timing and constraints, with square-bracket variables you fill in. That structure is what makes it work across models — but only if you use it the way it’s built. This is the routine we use ourselves.
Key takeaways
- Fill in every [VARIABLE] before you generate; an empty bracket is the most common cause of a bad result.
- Paste the whole brief, in order — models weight the first lines most, and the order is deliberate.
- Change one variable at a time between runs so you can tell what moved the result.
1. Open the prompt and read the constraints line first
The constraints line tells you what the prompt is designed to avoid — text on screen, extra people, camera shake, a second cut. If your idea needs one of those things, pick a different prompt rather than fighting this one.
2. Fill in the variables
Variables look like [CHARACTER_NAME], [SETTING] or [PRODUCT]. Replace each one with a short, concrete phrase: ‘a grey tabby cat’ beats ‘my cat’, and ‘a rain-soaked Tokyo side street at night’ beats ‘a city’. Keep each variable under about eight words so the brief stays readable to the model.
3. Paste the whole thing
Copy the complete brief into the model’s prompt box — Veo and Sora take it as one block; Kling does too, and also accepts the camera line in its separate camera-movement control if you prefer. Don’t trim the lighting or timing lines to save space; they’re doing most of the work on look and pacing.
4. Generate twice before judging
Video models are non-deterministic. Run the same brief twice. If both results miss in the same way, change the brief; if only one misses, run it a third time and keep the best.
5. Iterate one variable at a time
When a result is close, change exactly one thing — the lighting variable, the camera move, the character — and run again. Changing three things at once means you can’t tell which one helped. Most of our best clips come from the third or fourth variation of a single brief.
What to do with the output
Export at the model’s highest setting, trim to the first strong frame, and post natively (upload the file, don’t share a link). Add captions in the platform’s own tool — models still get on-screen text wrong, which is why our briefs tell them not to try.