Shot Prompt Builder

What it does

Converts one storyboard scene into one production-ready generation prompt: a single clear action, camera and lighting defined, character references attached, nothing improvised. The one-action rule is what keeps AI video from turning to soup.

When to use it

For each scene, right before you generate it
A generation came back chaotic — rebuild the prompt instead of re-rolling
You’re tempted to ask for “enters, sits, answers phone, argues, leaves” in one prompt (don’t — that’s five shots)

The skill

Act as a cinematographer writing a generation prompt for an AI
video tool. Hard rules: ONE main action per shot; nothing appears
that isn't in the scene or references; movement must be physically
plausible; if my scene actually contains multiple actions, split it
and tell me.

The scene: [PASTE FROM SHOT LIST]
Character paste-block: [PASTE FROM CONSISTENCY GUIDE]
Location paste-block: [PASTE FROM CONSISTENCY GUIDE]

Write the shot:
1. THE ACTION — one sentence, one action.
2. Character: movement, body language, expression through the shot.
3. Camera: framing, angle, movement (or locked), and why it serves
   the emotion of this beat.
4. Lighting and atmosphere.
5. Background life: what moves subtly behind the action.
6. First frame → last frame: what the shot opens and closes on.
7. The final assembled generation prompt: references + action +
   camera + lighting in one paragraph, ready to paste.
If this shot matters to the story (a hook, a reveal, an emotional
beat), say "generate 2-3 variations of this one" — and what to
compare between them.

Example output

[TO FILL AFTER TESTING]

Tweaks

Batch it: paste 3 scenes at once and get 3 shot prompts back — but still generate them one at a time
Rejected shots are data: note WHY (wrong motion? melted face?) and add that as a “never” line next time
The first-frame/last-frame pair matters most for tools that animate between keyframes