Most prompts that fail are not too short — they are unordered. A video model reads your prompt like a shot list, so the fix is to give it one. Every prompt in the Prompt Studio catalog uses the same seven lines. Here’s what goes on each, with the mistakes we see most.
Key takeaways
- Put the subject and its action first; the model weights the opening lines most.
- One camera instruction per clip. Two competing moves produce mush.
- The constraints line is where you stop the model doing the thing you don’t want — use it.
1. Subject
Who or what the clip is about, in physical terms: ‘a small round robot with one antenna’, not ‘a cute robot’. Adjectives like cute, cool or cinematic don’t render; features do.
2. Action
What the subject does, as one verb phrase. ‘Pours coffee, spills it, looks at the camera’ is three beats — fine for a 10-second clip; five beats is too many.
3. Environment
Where, plus one or two texture details that sell realism: ‘a rain-wet Tokyo side street, neon reflections, steam from a vent’.
4. Camera
Framing and one move: ‘medium shot, slow push-in’. If you need a second angle, that’s a second generation, not a second line.
5. Lighting
Time of day and quality: ‘late afternoon, warm, low sun from camera left’. This line does more for the ‘look’ than any style word.
6. Timing
Beats with rough seconds: ‘establish 0–2s, action 2–7s, reaction 7–9s’. Models respect pacing far more than most people expect.
7. Constraints
What must not appear: ‘no on-screen text, no extra people, no camera shake’. Every catalog prompt ships with this line because it removes the most common failure modes in one go.
Putting it together
Write the seven lines, paste them in that order, generate twice, and change one line at a time between runs. If you’d rather start from a tested brief, every prompt in Explore follows this exact structure with the variables marked.