Storyboard: Break the Script into a Shot List
This page delivers pre-production Document Three: the storyboard sheet (also called the shot-by-shot script). In 2.2 the short script template you wrote down what is visible in each beat; in 2.3 Dialogue Trimming you cut the lines until every sentence was necessary—Document Two is done. This page breaks it into a numbered AI storyboard sheet: five columns—shot number, shot size, visual, sound, and duration (this course's teaching template)—so every shot can be taken away and generated on its own. Splitting the storyboard still does not rely on inspiration, but on a sheet with fixed fields.
What You Will Be Able to Do After This Lesson
- Break the finalized script into a numbered AI storyboard sheet—delivering Document Three, with every shot ready to generate on its own
- Cut 30–60 seconds into 8–15 shots at 3–5 seconds per shot (the course design value), and backfill the shot count into the cost sheet
- Set the shot size for each shot and write a one-sentence subject-verb-object visual—the draft for generation prompts is in place at this step
- Place dialogue, voiceover, and ambient sound shot by shot, so the voiceover and subtitle stages can work straight from the sheet
- When needed, use AI to turn the script into a storyboard draft and pass the check lines yourself to finalize it (a Prompt designed for this course)
Your Current Struggles
- "The script is finalized, but when I open the generation tool I don't know what visual the first shot should be"
- "I had AI generate a 30-second clip in one go, the actions contradicted each other, and the whole segment was scrapped and retried"
- "I generated a folder full of footage, but which file corresponds to which shot, or which retry, is a complete mismatch"
When to Use This Technique
When you need to:
- Turn a finalized one-page script into a shot-by-shot executable list, generating and retrying each shot separately
- Before burning GPU time or credits, first calculate the shot count, per-shot duration, and retry budget
And you don't want to:
- Think up visuals as you generate, improvising every segment on the spot
- Discover during editing that the shots don't connect, then go back and burn the budget again
🎒 Preparation Before Starting
- Completed 2.2 the short script template: one A4 page laying out your 60 seconds, with the script finalized after passing four check lines
- Dialogue already trimmed in 2.3 Dialogue Trimming: Subtract Before Voiceover, and a full read-through fits within 10 seconds
- Have the 1.4 Cost and Trial-and-Error Log at hand—the shot count calculated on this page goes back into it
- A fresh A4 page or a blank spreadsheet document; this page is entirely on paper, no tools or GPU needed
Core Idea
The storyboard sheet answers "how to shoot each shot": the synopsis governs whether the story is right, the script governs whether it can be shot, and the storyboard governs how to shoot it. A shot is a smaller unit than a beat—a beat is one visual change, while a shot is the minimal work order for one generation and one edit; 1 beat is usually split into 1–3 shots.
Runway's official help documentation describes long-video production as: first do story conception and storyboarding, then prepare character references, then generate multiple short clips, and finally edit and combine them. Storyboarding comes before generation—first the shot list, then segment-by-segment generation; this is the fixed order of the official workflow.
Why set 3–5 seconds per shot (the course design value)? Two grounds:
- The generation side delivers short clips: the Wan2.2 official repository states speed in terms of "5 seconds at 720P"—for TI2V-5B without dedicated optimization, a single consumer-grade GPU generates a 5-second 720P video within 9 minutes; Runway's official workflow is likewise "multiple short clips edited and combined." No stage expects generating the whole film in one go.
- Editing needs headroom: a generated clip is only clean after trimming its head and tail, so the finished shot is set at 3–5 seconds, slightly shorter than the generated duration, leaving margin for editing.
Do the math: 30–60 seconds ÷ 3–5 seconds per shot ≈ 6–20 shots, and practice films usually land at 8–15 (the course design value).
Once the storyboard sheet is finalized, it becomes the production work order for the entire film—every later stage comes back to check it:
| Later Stage | Which Column of the Storyboard Sheet It Checks |
|---|---|
| 3.1 Character Card, 3.3 Scene Reference Images | Visual column—who appears in which shots, and where |
| 4.1 The First Shot, 4.2 Motion and Camera Prompt Writing | Visual column as the prompt draft, duration column converted to frame counts |
| 3.5 Footage Naming and Version Management, 4.5 Shot Retry and Replacement Strategy | Shot number—naming, retry logs, and replacement all depend on it |
| 5.1 Voiceover and Licensing, 5.2 Subtitle Creation and Proofreading | Sound column |
| 5.3 Editing and Volume Check | Shot number order + duration column |
Follow Along
The example below continues with the teaching example topic from 2.1, "Still Beautiful Even When It Collapses": on a stormy day under a pedestrian bridge, delivery rider Old Chen shields a birthday cake for his daughter. The dialogue is the two finalized lines after subtraction in 2.3. You can follow along with your own script throughout.
Step 1: Cut 60 seconds into 8–15 shots — first calculate the total number of shots
Why: AI video tools generate short clips of a few seconds at a time, not a whole film. Without a fixed shot count, you can't calculate generation, retries, or budget.
First calculate the total based on the target duration: target duration ÷ 3–5 seconds per shot (the course design value). A 30-second film needs about 6–10 shots; 60 seconds needs about 12–20. Then distribute the shots across the beats, 1–3 per beat:
| Beat | Content in the script | Assigned shot count |
|---|---|---|
| 1 | Under the pedestrian bridge in the rainstorm, Old Chen holds the cake box and looks down at his phone | 3 |
| 2 | He stuffs the phone into his pocket, gets up and dashes up the steps, his soles slipping | 2 |
| 3 | He falls into a puddle along with the box, and the box lid pops open | 2 |
| 4 | Crouching in the puddle, he gathers up the cake and pulls out his phone to answer | 3 |
Total: 10 shots × 3–5 seconds = 30–50 seconds, falling within the 30–60-second target.
The shot count is the seed of your budget: every shot must be generated and may need retries. The course rule in 4.5 Retry Strategy is at most 3 generations per shot — in the worst case, 10 shots require 30 generations. Backfill this number into 1.4 Cost and Trial-and-Error Log; locally, calculate by GPU time, and in the cloud, by credits (prices unverified — log by attempt count for now).
You should see: a new line in the script margin, "Total shots 10", plus the allocation per beat; total × per-shot duration falls within 30–60 seconds.
- Quick check: total shots × 3–5 seconds per shot falls within 30–60 seconds; each beat gets 1–3 shots; the total has been backfilled into the cost table
Step 2: Build a five-column AI storyboard — every field is a workstation
Why: The storyboard is the work order for all later stages — generation follows the picture column, voiceover follows the sound column, and editing follows the shot number and duration columns. Only with fixed fields can the stages match up.
Draw a five-column table with fixed column names (fields are the teaching template for this course):
| Field | What to fill in | Who checks it |
|---|---|---|
| Shot number | Numbered consecutively from 01, no gaps | Asset naming, retry log, editing order |
| Shot size | Extreme long shot / long shot / medium shot / medium close-up / close-up | Generation prompts, editing rhythm |
| Picture | One subject-verb-object sentence: who, doing what, to what | The draft for generation prompts |
| Sound | Dialogue (who speaks, whether off-screen) + ambient sound | Voiceover, subtitles |
| Duration | 3–5 seconds (course design value) | Frame count conversion for generation, editing ledger |
Keep the storyboard and the script on the same A4 page, or clip them together — every shot must be traceable to its beat, and when you change your mind, both are visible.
You should see: an empty five-column table with as many rows as the total shot count calculated in Step 1; the script placed alongside, ready to consult at any time.
- Quick check: five-column header built, shot number column starts from 01, row count = total shots
Step 3: Assign a shot size to each shot; the picture holds only one action
Why: Shot size determines how much information this shot gives the audience; the picture column is the draft for generation prompts — the image-to-video example prompt in Wan2.2 official repository explicitly states the shot size ("A close-up shot…"), so shot size is inherently part of the prompt.
Use the five most common shot sizes (standard film terminology, the teaching vocabulary for this course):
| Shot size | What's in the frame | When to use it |
|---|---|---|
| Extreme long shot | Environment dominant, figures very small | Opening establishing shot, establishing location |
| Long shot | Full figure + environment | Position changes, large movements |
| Medium shot | From the waist up | Walking, doing one thing |
| Medium close-up | From the chest up | Seeing expressions and handheld objects clearly |
| Close-up | Partial detail | Key objects, emotional beats |
Then break each beat's picture actions into shots, one action per shot. Compare with the script of "Still Beautiful Even When It Collapses":
| Beat | Picture action in the script | Broken into shots |
|---|---|---|
| 1 | Rain hammers down on the steps; Old Chen holds the cake box and looks down at his phone | 01 Extreme long shot · rain hammering the steps under the pedestrian bridge / 02 Medium shot · Old Chen holding the cake box, looking down at his phone / 03 Close-up · the phone screen lit up with his daughter's name |
| 2 | He stuffs the phone into his pocket, gets up and dashes up the steps, his soles slipping | 04 Medium shot · he stuffs the phone into his pocket and dashes up the steps / 05 Close-up · his soles slipping on the wet steps |
| 3 | Falls into the puddle together with the box, and the lid pops open | 06 Long shot·he falls into the puddle together with the box / 07 Close-up·the lid pops open, and half of the cream collapses |
| 4 | Crouches in the puddle gathering the cake back together; takes out the phone and gets connected | 08 Medium shot·he crouches in the puddle looking at the cake box / 09 Close-up·his hands gather the collapsed cream back into the box / 10 Medium close-up·he takes out the phone and gets connected |
Three rules (teaching template for this course):
- One shot, one action: split the action chain at its start and end points, numbering each separately
- The visuals follow the visual line of 2.2: complete subject-verb-object, filmable, no abstract words allowed
- Adjacent shots of the same subject change shot size: three consecutive "Medium shot·Old Chen" entries will make the film look like surveillance footage
You should see: 10 rows of shots now have shot sizes and visuals, and each row reads as an independently filmable moment.
- Quick check: each shot contains only one action; close-ups land on key objects (phone screen, shoe sole, cream, hands)
Step 4: Fill in the sound column—place dialogue, voiceover, and ambient sound shot by shot
Why: the sound column is the direct input for 5.1 voiceover and 5.2 subtitles. The dialogue finalized in 2.3 is assigned to specific shots in this step—who says it and whether it is off-screen, written clearly in each cell.
Three rules (teaching template for this course):
- Dialogue is annotated with who says it + whether it is off-screen
- Characters who never appear on screen can only have voiceover—the daughter in "Still Beautiful Even When It Collapses" is off-screen throughout the film (the convention set in the 2.2 header)
- Ambient sound is accounted for shot by shot; even silent shots have ambient sound
Referring to the two lines of dialogue finalized in 2.3, break the script's sound column down to shots:
| Shot | Sound column |
|---|---|
| 01 | Sound of torrential rain pounding the ground |
| 02 | Rain sound |
| 03 | Phone notification tone (screen lights up) |
| 04 | Rain sound getting closer, footsteps |
| 05 | A muffled thud of the shoe sole slipping |
| 06 | Splash of the fall |
| 07 | Sound of the box lid knocking against the steps; Old Chen (off-screen): "Don't collapse, don't collapse..." |
| 08 | Rain sound |
| 09 | Rain sound fading, phone vibration |
| 10 | Call-connected tone; daughter (off-screen): "It's beautiful even when it collapses, like a snow mountain." |
Note the way shot 07 is written: the visual is a close-up of the cake, Old Chen is not in the frame, and his line "Don't collapse, don't collapse..." is marked "off-screen"—if the person is outside the frame, the sound is voiceover, even if he is right nearby. The daughter never appears on screen in the whole film (the convention set in the 2.2 header), so no matter which shot her lines land in, they can only be marked off-screen.
You should see: 10 rows of sound all filled in, the two lines of dialogue each assigned to their own shots, and no empty cells in ambient sound.
- Quick check: every line of dialogue is annotated with who says it and whether it is off-screen; no sound column row is left empty
Step 5: Fill in durations and finalize by passing three check lines
Why: duration is the overall ledger of the storyboard. Each shot lasts 3–5 seconds (the course design value), and the total must fall back within 30–60 seconds—if it exceeds, go back to Step 1 to cut shots, not secretly shorten each shot.
Assign values within 3–5 seconds based on information load: a simple single action gets 3–4 seconds, and a shot with dialogue or an emotional beat gets 4–5 seconds. The duration column of "Still Beautiful Even When It Collapses":
| Shot | 01 | 02 | 03 | 04 | 05 | 06 | 07 | 08 | 09 | 10 | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Duration | 4 | 4 | 3 | 4 | 3 | 4 | 5 | 4 | 4 | 5 | 40 seconds |
40 seconds falls within 30–60 seconds. At generation time, seconds must be converted to frames—Wan2.2 TI2V-5B runs at 24 FPS (as stated in the official repository), so 4 seconds ≈ 96 frames; this conversion is put into practice in the shot generation section's workflow, and the storyboard only records seconds.
Before finalizing, pass three check lines (teaching template for this course):
Storyboard check lines
□ Ledger line: each shot 3–5 seconds, and does the total fall within 30–60 seconds?
□ One-shot line: one action and one subject-verb-object visual per shot? Shot numbers run continuously from 01 with no gaps?
□ Sound line: is dialogue annotated with the speaker and off-screen status? Is ambient sound accounted for shot by shot? Does the dialogue still fit within 10 seconds when read aloud?Whichever line fails, go back to the corresponding step to fix it; once all three pass, the storyboard is finalized—this is File Three. From the next lesson onward it is a read-only file: continuity checks are annotated on it, and generation starts according to its shot numbers.
You should see: the duration column fully filled in, the total falling within the target range, and all three check lines ticked.
- Quick check: all three check lines pass; the total shot count and worst-case generation count have been written back into the 1.4 cost table
Checkpoint ✅
- The storyboard has all five columns complete: shot numbers run consecutively from 01 with no gaps or duplicates
- One action per shot, one subject-verb-object sentence per shot — every shot can be taken away and generated on its own
- Shot sizes vary, with close-ups landing on key objects
- The sound column is filled in shot by shot: dialogue marks who speaks and off-screen, ambient sound has no blanks, and the full dialogue reads aloud in one pass within 10 seconds
- Each shot lasts 3–5 seconds (the course design value), the total falls within 30–60 seconds, and the shot count has been filled back into the cost table
Pitfall Reminders
❌ Two actions crammed into one shot
Symptom: the picture column says "He gets up, rushes up the steps, slips, and falls into a puddle" — one shot holds two things, beat 2 and beat 3.
Cause: the script's beats were copied wholesale into the picture column, forgetting that a shot is a smaller unit than a beat. A shot is the minimal work order for generation and editing: with two actions in it, one retry scraps the whole shot, and editing can't keep just the second half.
Correct approach: one action per shot; cut the action chain at its start and end points and number each separately — "gets up and rushes up the steps" is one shot (04), "falls into the puddle" is one shot (06), and the "sole slips" in between gets its own close-up (05).
❌ Adjectives sneak into the picture column along with the script
Symptom: the picture column contains "He stares at the cake in despair" or "The rain curtain is vast and sorrowful".
Cause: the abstract words already cleaned out once in 2.2 came back in a different table. The picture column is the draft for generation prompts — prompts take filmable subject-verb-object content, not emotional verdicts.
Correct approach: translate emotion into shot size and action. "Stares in despair" → close-up + "He stares at the cream that has collapsed on one side, his fingers hovering above the box lid, motionless." After filling it in, go through the picture line of 2.2 shot by shot.
❌ Shot numbers made up on the fly, with gaps and duplicates
Symptom: at shot 7 a "3A" gets inserted on the spot, and after deleting one shot the numbering jumps straight from 08 to 10.
Cause: treating shot numbers like row numbers. The shot number is the anchor for 3.5 asset naming (the "shot number_version number" convention), the key for the 4.5 retry ledger , and the ordering list for editing — once it's a mess, the three downstream stages all fall apart.
Correct approach: edits are allowed before finalization; after finalization, shot numbers are frozen — deleting a shot leaves the number vacant, a new shot takes the next number in sequence, and any full renumbering must update all references in sync.
Advanced Tips (Optional)
Have AI turn the script into a storyboard first draft — the correct way to use "AI script-to-storyboard" in this course: use it to break down a first draft, not to write the final version. Splitting a script into shots is mechanical work that AI does fast; shot-size choices, the total-duration tally, and the check lines are yours to go through.
Expected result: paste the finalized script and get a first-draft storyboard with all five columns, which you then revise shot by shot.
Use the prompt below (a Prompt designed for this course, as a teaching example; any conversational AI assistant will do — no particular tool required):
## Role
You are a short-film storyboard artist, skilled at breaking down a 60-second single-scene short script into a shot-by-shot list that can each be generated.
## Task
Break my finalized script into a first-draft storyboard: 3–5 seconds per shot, 8–15 shots in total.
## Input Information
### Required Fields
- Script table: [Paste the finalized beat table from 2.2: beat number | picture action | dialogue | sound]
- Target duration: [a value within 30–60 seconds, e.g. 40 seconds]
### Optional Fields
- Key objects to emphasize: [e.g. phone screen / cake box…]
## Output Specification
A five-column table: shot number | shot size (extreme long shot / long shot / medium shot / medium close-up / close-up) | picture (one subject-verb-object sentence, one action per shot) | sound (dialogue marked with who speaks and off-screen + ambient sound) | duration (3–5 seconds). Give the total duration at the end of the table.
## Constraints
- ✅ Use only the pictures, dialogue, and sounds already in the script; add no new plot
- ✅ Place dialogue as-is: no rewriting, no cutting
- ❌ No new characters, no lines not in the script, no two actions in one shot💡 The Boundaries of AI Assistance
AI's first drafts commonly have two flaws: two actions in one shot, and a total-duration tally that doesn't add up. When you get the draft, first run the three check lines from step 5, then decide shot by shot whether to revise — every shot in the final version must pass your own eyes for the course exercise to count.
FAQ
What is an AI shot script? How does it differ from an ordinary storyboard?
An AI shot script (called the AI storyboard in this course) breaks the finalized script into a shot-by-shot executable list, with five fixed columns: shot number, shot size, visuals, sound, and duration. The difference from an ordinary storyboard is that every shot can be taken individually to AI generation — AI video tools generate short clips of a few seconds at a time, not the whole film, so each shot is set to 3–5 seconds (the course design value), generated individually and retried individually.
What is the role of an AI storyboard artist? Can AI act as the storyboard artist?
Yes, but only let it produce the first draft, not the final version. In the "Advanced Tips" section on this page there is an "AI Script to Storyboard" prompt: paste the finalized script and have the AI output a first-draft storyboard of 8–15 shots in the five-column table format. Breaking down a storyboard is mechanical work that AI does quickly; but shot-size choices, the duration ledger, and the three check lines must be done by you — the final version only counts if you review every shot yourself.
How do you use AI shot scripting? What is the next step after getting the AI-generated first draft?
First run the three check lines from step 5: the ledger line (each shot 3–5 seconds, with the total falling within 30–60 seconds), the one-shot line (one action per shot, shot numbers running continuously from 01 without gaps), and the sound line (dialogue marked with who speaks and whether it is off-screen, with ambient sound accounted for in every shot). AI first drafts commonly have two problems: two actions crammed into one shot, and a duration ledger that does not add up — whichever line fails, go back to the corresponding step to fix it; only when all three lines pass is the storyboard finalized.
Is there a complete workflow for an AI shot scripting tutorial?
This page is the complete workflow, in five steps: step 1 calculates the total number of shots from the target duration (30–60 seconds ÷ 3–5 seconds per shot ≈ 8–15 shots); step 2 builds the five-column storyboard; step 3 sets shot sizes, with one action per shot in the visuals; step 4 places dialogue, voiceover, and ambient sound shot by shot; step 5 fills in durations and passes the three check lines to finalize. The required inputs are a script finalized with the 2.2 short script template and dialogue trimmed with 2.3 dialogue trimming.
Lesson Summary
✅ At this point, you have:
- Broken the finalized script into a numbered AI storyboard — delivering Document Three, and every shot can be taken away and generated on its own
- Cut 8–15 shots at 3–5 seconds each (the course design value), and filled the shot count and worst-case generation count back into the cost table
- Assigned a shot size to each shot and written a one-sentence subject-verb-object picture — the draft for generation prompts is in place
- Placed dialogue, voice-over, and ambient sound shot by shot, so dubbing and subtitles can be built straight from the table
Quick reference for the five fields of the AI storyboard:
| Field | What to Fill In | Corresponding Check Line |
|---|---|---|
| Shot number | Numbered consecutively from 01, frozen after finalization | One-shot line |
| Shot size | Five tiers: extreme long shot / long shot / medium shot / medium close-up / close-up | One-shot line |
| Picture | One action per shot, complete subject-verb-object | One-shot line |
| Sound | Dialogue (who speaks / off-screen) + ambient sound | Sound line |
| Duration | 3–5 seconds, total falling within 30–60 seconds | Tally line |
Next Lesson Preview
The 10 shots are laid out on paper, but how shots connect to one another — whether the end of one shot matches the start of the next — is the last check before generation begins. Next lesson: Shot Continuity Check: The Transition Checklist.
You will learn:
- How to check the connection of adjacent shots pair by pair (subject, position, lighting, direction of action)
- Which connection problems to fix on the storyboard, and which to leave for first-and-last-frame generation to rescue
- Only after finishing the transition checklist is it safe to start generating in shot-number order