Script and shot list
Turn a one sentence idea into beats and a shot list of 3 to 8 second shots, with a template, an LLM prompt and a full 30 second example.
Write the shot list before you write a single video prompt. Sources from Kling, MindStudio and Motionbox all call it the biggest quality lever you have. A shot list tells you how many clips to generate, what each one shows, and which references each one needs.
From idea to beats
A beat is one moment the viewer must understand. Most short videos need four kinds:
| Beat | Job | Where |
|---|---|---|
| Hook | Make people stop scrolling | The first 2 to 3 s |
| Setup | Show who, where and what they want | Right after the hook |
| Turn | Something changes or goes wrong | Middle |
| Payoff | The reveal, the joke, the product moment or the feeling | The end |
Ask an LLM for the beat sheet first and fix it before you go to shots. Fixing text costs nothing.
Turn this logline into a beat sheet of 6 to 10 beats for a <N> second video.
For each beat: purpose (hook, setup, turn or payoff), emotion, the single visual
that sells it, and whether it needs dialogue (prefer no). Mark the 2 beats
that are hardest for AI (several characters, hands, text) and propose simpler
visual substitutes.From beats to shots
Each beat becomes one or more shots. Keep these rules:
- Generate shots of 3 to 8 s. Shorter clips follow the prompt better. Two 4 s clips can beat one 8 s clip.
- Count shots from the runtime: a 30 s video is about 5 to 6 shots, a 1 min video about 8 to 15.
- Generate 3 to 6 s for story shots, then trim them to 2 to 3 s in the edit. MindStudio's rule: cut before the emotion needs it.
- One subject action and one camera move per shot.
- Every shot must stand on its own. The model has no memory of the shot before, so each line restates the character, place, light and style.
- Keep screen direction consistent. If a character walks left to right in shot 2, keep them moving left to right in shot 3 (the 180 degree rule).
Shot list template
Copy this table into a doc or spreadsheet. One row per shot.
| Shot | Duration | Framing | Action | Camera | Audio | References |
|---|---|---|---|---|---|---|
| 1 | 4 s | Wide | Who does what, in one sentence, ending in a still moment | One move, or static | Sound effects, ambience, short line, "no music" | Character sheet, location sheet, product photo |
| 2 |
What each column is for:
- Framing: how much of the subject we see (wide, medium, close up). See the camera vocabulary.
- Action: what changes during the shot. If you need two sentences, you probably need two shots.
- Camera: one move only (static, push in, pan, tracking, crane). Add a lens if you care about the look.
- Audio: the sounds you want, because models invent music when you leave this empty.
- References: which images you will attach. This column tells you what to build on the reference images step.
Let Claude or ChatGPT write the shot list
Paste this as the system prompt or the first message. Fill in the project bible once and reuse it for the whole video. PJ Accetturo, who made the Kalshi ad, asks for 5 prompts at a time because quality slips with more, so the prompt asks for batches of 5.
You are a director and AI video shot planner. Project bible:
* Runtime: <N> s, aspect ratio <16:9 or 9:16>
* Style line (reuse word for word in every shot): <style>
* Characters: <NAME: 5 to 7 fixed traits>, at most 2 on screen
* Locations: <at most 2, with lighting and 3 to 5 palette colors>
Rules: every shot is 3 to 8 s; one subject action and one camera move per shot;
no on screen text; each shot description is self contained (restate the
character, setting, lighting and style). Durations must add up to the runtime.
Output a table with these columns:
Shot | Duration | Framing | Action | Camera and lens | Audio (sound effects, ambience,
dialogue of 10 words or less, or "no music") | References to attach
Return 5 shots at a time and wait for me to say "next".Read every row before you accept it. Cut any row with two actions, a crowd, readable text, or a camera move plus a zoom plus a pan.
Worked example: a 30 s short
Idea, in one sentence: "RIO runs for the last night ferry in the rain, misses it, and ends up sharing his umbrella with a stray dog."
Beats: hook (ferry lights leaving), setup (Rio running), turn (he missed it), payoff (the dog, the umbrella).
| Shot | Duration | Framing | Action | Camera | Audio | References |
|---|---|---|---|---|---|---|
| 1 | 4 s | Wide | Ferry lights pull away from an empty terminal in drizzle | Static | Rain, distant ferry horn, no music | Location sheet |
| 2 | 5 s | Medium | Rio runs along the wet pier and stops hard at the railing | Tracking, following left to right | Splashing footsteps, rain, no music | Rio sheet, location sheet |
| 3 | 4 s | Medium close up, 85mm | Rio exhales, breath visible, watches the ferry go | Slow push in | Heavy breathing, rain | Rio sheet |
| 4 | 5 s | Close up, low angle | A small wet brown dog trots in and sits beside Rio's white sneakers | Static | Paws on wet concrete, rain | Dog sheet, Rio sheet (sneakers) |
| 5 | 6 s | Medium two shot | Rio looks down, half smiles, opens his umbrella over the dog | Static | Umbrella click, rain on fabric | Rio sheet, dog sheet |
| 6 | 6 s | Wide | Rio and the dog stand at the railing, ferry lights small on the water | Slow crane up | Rain, one far horn, no music | All three sheets |
Total: 30 s in 6 shots. At 2 to 5 takes per shot that is 12 to 30 generations, which you can price before you start. Music, if any, goes in during the edit.
Shot 3 written as a standalone prompt:
Medium close up, 85mm lens, slow push in. RIO, early 30s, short black curly hair,
thin silver hoop in left ear, rust orange work jacket over a grey hoodie, stands
at a rain wet railing at a night ferry terminal. He exhales hard, his breath
visible, and watches the ferry lights leave, then stays still. Sodium orange
and teal light, light drizzle, background lights soft. Stylized 3D animated
film look. Audio: heavy breathing, steady rain, no music.Notice that the prompt repeats Rio's full description and the location even though shot 2 already showed them. That is on purpose. See Prompting for the full structure.
Hand edit the LLM draft
LLM drafts are a starting point. The Big Yowie creator rewrites them using what they know about the model. Remove anything the model is bad at before you spend credits on it.
Sources
- Shot list first, shot counts, field list: Motionbox, Kling, MindStudio, CapCut
- Shorter clips follow prompts better: OpenAI Sora 2 prompting guide
- Batches of 5 and self contained prompts: PJ Accetturo via Business Insider
- Screen direction and hand editing: Kapwing