I used to write scripts the way I was taught. Flowing prose. Scene headings. Dialogue floating in the middle of the page. Descriptions that read like a short story. Beautiful documents — honestly, they looked great. And completely, utterly useless the moment I sat down to generate. I’d stare at a paragraph like “Maria enters the café, nervous but hopeful, and everything changes” and realize I’d written something a human director could interpret but a generator could not execute. Which café? What does she look like? Where’s the camera? What, concretely, is happening in the frame? It took me a month of friction to learn the rewrite: for AI video, the script is not literature. It is a shot list. A table of decisions made once, on paper, so that prompting becomes transcription instead of improvisation.
So this guide lays out the format I use for every multi-shot project now. The columns. How to handle dialogue and sound. Where the continuity notes live. And how a finished shot list turns into Seedance or Veo prompts without a second round of creative writing. If prose scripts have ever left you re-deciding your film at the prompt box — camera angle here, wardrobe wording there, at render prices — this is the fix. This is genuinely the fix.
Prose Scripts Describe Films; Shot Lists Direct Them
A prose screenplay describes a film for humans who will spend weeks interpreting it. Actors bring faces. A director of photography brings the camera. The document can afford to be suggestive, because an army of specialists will convert suggestion into specifics. In AI production? You are that entire army. And the generator is a talented, literal-minded collaborator that will convert your specifics — or, in their absence, invent its own. Every unspecified decision in a prose script becomes an invention at generation time. And invented decisions do not coordinate with each other. That is how a character ends up with three jackets, two haircuts, and a café that rebuilds itself between shots.
A shot list moves every one of those decisions upstream. Onto one page. Where they can agree with each other. Instead of discovering at shot six that you never decided the time of day, you decide it once, write it in a header, and copy it down the column. The creative work doesn’t disappear — it concentrates. Rather than making forty small decisions scattered across a day of prompting (each one slightly worse than the last, because you’re tired), you make them in one focused session, looking at the whole film at once. One sitting. One clear head.
There’s a cost argument too, and it’s blunt. Prompts written by transcription generate right the first time more often. Every “let me just re-decide this at the prompt” moment risks a wasted render. On character-heavy pieces I’ve watched my first-take success rate climb — simply because the prompt box stopped being a writing prompt and became a copy-paste exercise. Boring at the keyboard. Thrilling in the render queue.
The Shot List Format: Size, Subject, Action, Camera, Light
Here’s the core table. Five columns carry a shot; a sixth (duration/notes) rides along in practice. I keep it in a spreadsheet or a simple document table. The tool doesn’t matter. The columns do. That’s the only part I’m strict about.
- Shot number. The shot’s identity everywhere: storyboard panel, generation file name, timeline clip. One number per shot, never reused, so “Shot 07” means the same thing to every version of you.
- Size. Wide, medium, medium close-up, close-up. The shot-size grammar covered elsewhere in this series; here it is simply a required field, because a shot without a declared size gets the model’s default, and defaults are how sequences flatten.
- Subject. Who or what is in the frame, named with the exact reference wording: “Maya (ref A), denim jacket” — not “the woman.” Names and wardrobe here feed directly into prompts and into consistency checking.
- Action. One verb-driven sentence, present tense, describing the single main movement: “Maya stops mid-reach for the door handle, seeing the note.” If the action column needs the word “and” twice, the shot is probably two shots.
- Camera. Position and movement: “Locked, eye level” / “Slow push-in from medium to medium close-up” / “Over Rafi’s shoulder.” Keep movement to one instruction; competing moves produce soup.
- Light. Source, direction, and time: “Window light camera-left, late afternoon.” This column looks fussy and saves more scenes than any other; light disagreement between shots is the seam viewers feel without naming.
A real row, from a bakery short I storyboarded with this system: 04 | Close-up (object) | Maya’s hand + door handle, paper note taped to glass | Hand reaches in, stops before touching handle | Locked, slight low angle | Cool dawn through glass, warm spill from streetlamp left. Read that row and you can already write the generation prompt. Which is precisely the point. Nothing in that row is flavor. Every word will be spent.
Duration belongs in the list as a target, not a fact. “4s” means this shot should carry four seconds of the story. Generated clips will arrive longer — they always do — and you trim to the plan. The plan matters because the total column should roughly sum to your target runtime before you generate, so a 30-second piece doesn’t quietly become a 47-second piece through nobody’s decision. I say roughly. Leave a few seconds of slack for connective shots you’ll add in the edit. And never let slack live in the ending.
Dialogue And Sound Columns
Two more columns turn a shot list into a shooting script proper: Dialogue and Sound. They live in the same table because dialogue and sound must be planned against the same frames the picture is planned on. A line’s length determines a shot’s minimum duration. A sound cue often determines a cut. Same frames, same table.
The Dialogue column holds the exact line, the speaker, and the delivery note in three words or fewer (“quiet, not looking up”). Exact means exact — write the final wording here, not a summary like “she explains the plan,” because the line’s length in syllables is a production fact. If you’re generating voice separately (my default for full scenes), this column is also your recording script, and the voice session runs straight down it. If a shot has no dialogue, the cell says so. Blank cells get filled by assumption later — and assumptions are how silent beats acquire surprise narration in the edit.
The Sound column is where most AI shot lists are emptiest, and where mine earns its keep. For each shot, note the dominant sound event (door latch, rain starting, kettle), the ambience (café murmur, night street), and any music note (music out, sting on the reveal). Planning sound per shot does three jobs. One: it reminds you to source or generate those sounds before the edit. Two: it exposes continuity problems while they’re still text — ambience that changes identity between shots in the same room. Three: it makes the Premiere Pro assembly dramatically faster, because you’re executing a list instead of hunting feelings at midnight. The dialogue coverage guide in this series leans on reaction shots; the sound column is where you plan the room tone that glues them together.
One rule for both columns: they describe the finished video, not the generation. A shot generated silent still gets its sound cell filled, because the cell describes what the viewer hears at that moment in the final piece. Keeping “what generation produces” and “what the audience experiences” in separate mental boxes is the entire discipline of AI filmmaking in one sentence. One sentence. Tattoo it somewhere.
Reference And Continuity Notes Per Shot
The final column is Notes, and it is the conscience of the whole document. This is where continuity lives. Which character reference set this shot uses. Which location plate. What must match the previous shot — wardrobe state, prop position, screen direction, eyeline. And what state changes happen inside the shot that later shots must inherit: jacket now off, letter now opened, window now broken.
Write these notes as instructions to a forgetful collaborator, because the generator is one. “Same café plate as Shot 02, Maya enters from frame left” will do more for your sequence’s coherence than any single prompt trick. When a character’s appearance is anchored by reference images — the workflow detailed in the reference-images guide — the Notes column records which references attach to which shots, so a character introduced in Shot 03 with reference set A never silently switches to set B in Shot 09 because you grabbed the wrong file on a tired evening.
Continuity notes also cover the unglamorous logistics that save edits: eyeline direction per dialogue shot (looks frame-right), movement direction (crosses left to right), prop hand (letter in left hand). These are the details the eyeline and cutting-on-action guides in this series depend on. And they have exactly one natural home: written once, in the row, before generation. Continuity invented in the timeline is continuity paid for in re-renders.
A practical tip: after drafting the full list, read down single columns as a sanity check. Read only the Light column — does the film’s day progress sensibly? Only the Size column — does the framing breathe, or does it park on mediums? Only the Dialogue column — does the conversation actually read? Column-reading catches structural problems in ninety seconds that shot-by-shot reading hides completely. Try it once and you’ll do it forever.
From Shot List To Prompts Without Rewriting
With the list complete, prompt-writing becomes assembly. My per-shot routine:
- Start from the row. Size + subject + action + camera + light, in sentence form, is 80 percent of a working prompt. The row was written to be spent this way.
- Add the anchors. Attach the reference images named in Notes; paste the character or product description string verbatim from the project’s header block. Identical wording, every time — the consistency guides explain why this matters; the shot list makes it effortless, because the canonical string lives in exactly one place.
- Add the motion handles. From the Action cell and its neighbors, note how the shot should start and end (“begins mid-stride; ends with hand reaching out of frame”) so the clip arrives with cut points built in.
- Write nothing new. If the prompt needs an idea that is not in the list, stop and put it in the list first. That sounds bureaucratic; it is actually protective. It keeps one document as the single source of truth, so when Shot 04 changes, Shots 05 and 09 hear about it.
This is the working end of the prompt-to-final-cut workflow — the list is the bridge between storyboard and generator. Together with your storyboard panels it means every creative decision happens twice-checked and once: once visually on the board, once textually in the list, and never again at render prices.
File discipline rides along free. Generations save as Shot04_v1, Shot04_v2; the list’s Notes column records which version won and why (“v2 — hand cleaner”). Six weeks later, when a client or your own curiosity asks why the film looks the way it does, the answer exists. Professionalism, in AI production, is largely this. Being able to answer that question.
A One-Page Shooting Script Template
Everything above collapses into a one-page template. Header block first: project title, target runtime and format, the canonical character description strings, product or prop strings, location list, and the global light/time scheme. Then the table: Shot | Size | Subject | Action | Camera | Light | Dialogue | Sound | Notes (refs + continuity). A 30-second piece usually fits in eight to twelve rows. Genuinely one page. Which is the point — a shooting script you can hold in your head beats a binder you cannot.
Copy this skeleton and fill it before your next generation session:
- Header: Project / Runtime / Characters (name + fixed description string + ref set) / Locations / Global light scheme / Ending type (loop, cliffhanger, reveal).
- Per row: one shot, one main action, named size, one camera instruction, light in five words, exact dialogue or “none,” the shot’s key sound, and continuity owed to neighbors.
- Footer: total planned duration, generation order (hardest shots first, ending early), and the version-note line where winning takes get recorded.
I’ll be honest about the adjustment period: the first shot list feels slower than free-prompting, the way the first storyboard did. By the third project? It is simply how films get made in this studio of one. I open the template, fill the header from the character sheets, spend twenty minutes deciding everything once — and then generation day is a calm transcription exercise with coffee. The prose script had its century. For a filmmaker whose crew is a model, a timeline, and one pair of eyes, the shot list is the script now. Write the table. Let the prompts write themselves.