Script to storyboard: how to break a scene into shots (and where AI takes over)
A storyboard is a list of decisions; the drawings only carry them. From a three-line script, mark beats, give each a shot, write start and end states, copy the template, see what AI can take.
Ask three people what a storyboard is and you get three answers: a table with one row per shot, a strip of drawings in order, or the director's script with shot numbers scribbled down the margin. They look different, and they answer the same question: which shots does this scene become, and what is in each one?
This article leaves drawing aside. The decisions behind the frames are what a storyboard is for; how good the pictures look comes a distant second. We are going to take a three-line script and walk it all the way to shots, then hand you a template you can copy.
Are a storyboard, a shot list and a shooting script the same thing?
Put the three things back where they belong:
| Name | What it looks like | What it answers |
|---|---|---|
| Shot list | A table, one row per shot | Which shots, how long, what size and movement |
| Storyboard | Frames in order, one per shot | What each shot looks like |
| Shooting script | The script with shot numbers and technical notes | What the crew follows on the day |
All three carry the same decision: a continuous scene gets cut into shots that can each be executed on their own. The shot list is the leanest form of that decision, the storyboard is its picture, the shooting script is the decision pasted back onto the pages.
So the question underneath "how do I make a storyboard" is how this scene gets cut, and why that way. Once the cut is right, a table, a strip of sketches or margin notes are all just containers. When the cut is wrong, good drawings do not save it.
Script first or storyboard first? Three lines count as a script
Script first. A storyboard translates a script, and drawing frames before there is one means choosing shots for a story that does not yet know what it is about, guessing in every panel.
"Script" here need not mean a formatted screenplay. A paragraph that says who, where, what happens and how it ends is enough:
A convenience store, late at night. Ann puts a drink and a rice ball on the counter, reaches into her pocket, and the wallet is not there. She goes through her coat and her bag while the clerk waits. Finally she pushes the items back, says sorry, and walks out.
Three lines. Everything below grows out of them.
Step 1: find the moments the audience cannot miss, the beats
A beat is a moment the audience has to see. Actions and beats are different units: searching the coat, searching the bag, searching the other pocket are three actions and one beat, "she cannot find it".
Mark the beats in that paragraph:
- Items go on the counter. The scene has started; we are in a convenience store.
- The hand goes into the pocket and stops. The wallet is not there.
- Searching while the clerk waits. The awkwardness is building.
- She gives up and pushes the items back. The decision is made.
- She walks out the door. The scene is over.
Five beats. The only test is: if you remove this moment, does the audience lose the thread or the feeling? If yes, it is a beat. If no, it is connective tissue; fold it into the beat next door.
This is the step that needs a human most. We will come back to it when we get to AI.
Step 2: one shot per beat, and size, angle and movement each answer one question
With the beats marked, give each one a shot. Start with one shot per beat; split a beat into two shots only when it clearly needs it. Splitting later is much easier than merging later.
Every shot needs three decisions, and each answers a different question:
- Shot size answers "how close is the audience". Wide establishes the place, medium establishes the relationship, close-up delivers the emotion or the detail.
- Angle answers "whose side is the audience on". Eye level observes, low angle gives power, high angle takes it away, over-the-shoulder means "I am listening to you".
- Camera movement answers "how does time move inside the frame". Static watches, push-in closes in, tracking walks alongside. The vocabulary for movement is standardised across the industry; the camera commands guide goes through it term by term, so we will not repeat it here.
Apply that to the five beats:
| Shot | Beat | Size | Angle | Movement | Length |
|---|---|---|---|---|---|
| 1 | Items go on the counter | Medium | Eye level, over the clerk's shoulder | Static | 3 s |
| 2 | Hand into pocket, stops | Close-up (hand and pocket) | Slightly high | Static | 2 s |
| 3 | Searching, clerk waits | Medium | Eye level, facing Ann | Slow push-in | 5 s |
| 4 | Items pushed back | Close-up (items sliding away on the counter) | High | Static | 3 s |
| 5 | Out the door | Wide (from inside, toward the door) | Eye level | Static | 4 s |
Seventeen seconds, five shots. A few of the reasons behind the choices:
- Shot 3 is the centre of the scene, so it gets the longest duration and the only camera move: a slow push-in that lets the awkwardness close in.
- Shots 2 and 4 are both close-ups, one for "discovery" and one for "giving up", with different subjects (hand, items), so they will not blur together.
- Shot 5 goes back to wide and uses space to close the scene.
That table is already a usable shot list. Whether you draw it depends on whether the people you work with need pictures.
Step 3: write the start and end states for each generated shot
On a traditional shoot, continuity between shots is guarded by the script supervisor and the editor. When the shots are generated, each one is produced separately and nobody is on set, so "where things ended in the previous shot and what the next shot has to pick up" has to be written into the storyboard, or every shot starts from zero and guesses.
Shots 2 to 3, for example:
- Shot 2 end state: Ann's right hand is still in her coat pocket, eyes down.
- Shot 3 start state: Ann's right hand comes out of the coat pocket, she looks up at the clerk.
Two lines give you a continuity target to check across the independently generated shots. It deserves an article of its own, and a simple storyboard to connect every shot is that article, including how to tell which kind of problem you have when two shots refuse to join. Here, just remember: add two columns to the shot list, start state and end state.
How do you write a storyboard? Eleven columns, and two you cannot skip
Put the three steps together and this is the table. One row per shot, columns as follows:
| Column | What goes in it | Common mistake |
|---|---|---|
| Shot number | Scene plus running number, e.g. 3-2 | Reordering without renumbering, so the edit never matches |
| Beat | The reason this shot exists, one sentence | Writing a picture description and forgetting what the shot has to prove |
| Size | Wide / full / medium / close / extreme close | Three close-ups in a row and the audience loses the room |
| Angle | Eye level / high / low / over-the-shoulder / POV | Everything at eye level, so the emotion has no highs or lows |
| Movement | Static / push / pull / pan / track / crane | Every shot moving, so movement means nothing |
| Length | Seconds | Everything at 5 seconds, so the rhythm is flat |
| Description | Who is where doing what, one or two sentences | Adjectives instead of actions |
| Dialogue / sound | Everything audible in this shot | Forgetting room tone, so every shot has a different sound floor |
| Start state | Where people, objects and positions begin | Blank, so the next shot guesses |
| End state | Where the action stops | Same |
| Notes | References, risks, alternatives | — |
Filled in for shot 3 of our scene:
| Column | Content |
|---|---|
| Shot number | 1-3 |
| Beat | Searching while the clerk waits; awkwardness building |
| Size | Medium |
| Angle | Eye level, facing Ann, clerk soft in the foreground |
| Movement | Slow push-in |
| Length | 5 s |
| Description | Ann moves from coat to bag, faster each time; the clerk's hands rest on the counter and do not move |
| Dialogue / sound | No dialogue; the hum of the drinks fridge, fabric rustling |
| Start state | Right hand comes out of the coat pocket, she looks up at the clerk |
| End state | Both hands stopped on the open bag, eyes on the items on the counter |
| Notes | High risk: the clerk's posture tends to drift across shots; prepare a reference image of the clerk |
That is the full answer to "what format should a storyboard be in". Drop columns if you like, but keep start state and end state: they make the intended handoff explicit for separately generated shots.
Can AI turn a script into a storyboard? It can draw the frames, not decide the cut
Now look back over the whole process and separate "a person does this" from "AI can do this":
Marking beats: a person. A language model can read the script and list candidate beats, quickly. Check whether its list treats every action as a separate beat or gives minor actions too much weight. Use its list as a draft. Which beats survive and which one is the centre is your call.
One shot per beat: a person decides, AI fills the table. Give a model your beat list and a few rules ("the centre gets the longest duration and the only camera move") and it will fill in size, angle and movement into a decent table, and sometimes suggest options you had not considered. Judge the proposed shots by how they serve this scene, not just whether each choice sounds defensible on its own.
Drawing the frames: AI can take this. This is what the storyboard tools are actually good at: turning each row of the shot list into an image. The precondition is shot list first, then feed it in. The reverse (let it draw a strip of frames, then reverse-engineer shots out of the pictures) is currently the most common way people use these tools, and the most wasteful: you get a row of handsome pictures with no beats, then spend your time inventing reasons for them.
Start and end states: a person writes, AI reads. These two columns are exactly what a video generation model consumes. They describe what the next shot should inherit; the generated transition still needs checking.
So "can ChatGPT generate a storyboard" has a precise answer. Script, then you mark beats, then you assign shots, then AI draws, then you check that every frame still proves its beat. AI can draft the table and frames; you decide whether the sequence serves the story.
Ninety years ago the storyboard was invented so that a room full of people who could not see the same film could agree on one. Today there is one more collaborator who cannot see it, and it is not even in the room.
The only thing it receives is your table. Write down how the scene is cut, and why each cut is there, and it has a chance of making what you meant. How pretty the pictures are has never mattered much.