How to Storyboard a Video with AI: A Complete Workflow (2026)

Storyboarding used to mean drawing. For most people that was the whole reason they skipped it.

It doesn't anymore. You paste a script, get a numbered shot list back, edit the parts that are wrong, and generate a frame for each shot — then turn the frames you approve into video.

This is the workflow, end to end. Five steps, the two places it still falls down, and a worked example you can follow along with.

We run Anyscene, so most of what's below came from watching credits disappear into boards that weren't planned properly.

Why a Storyboard Still Decides the Edit

A video fails in pre-production, not in the edit.

By the time you notice that shot four contradicts shot one, you've already paid for both. That was true when the cost was a crew day. It's still true now that the cost is generation credits — the bill just arrives faster.

What goes wrong without one

Here's the pattern we see most.

Someone generates eight shots from eight separate prompts. Then they notice the character's jacket changes colour three times.

The fix isn't one regeneration. You re-anchor the reference and redo everything downstream of the break.

What changed now that shots cost credits

Traditional storyboarding saved money by preventing wasted shooting days. AI storyboarding saves money by preventing wasted generations — same logic, much shorter feedback loop.

The rework math

Twenty minutes of planning beats six regenerations every time, and it isn't close.

What Is an AI Storyboard?

An AI storyboard is a numbered sequence of shots generated from a written script, where each shot carries its own framing, camera note and image prompt — and a generated frame you can review before committing to video.

That definition is narrow on purpose. The category has been stretched to cover tools that only do half of it.

Board vs animatic vs previz

People use these interchangeably. They're different deliverables.

This workflow produces a board, and optionally the moving shots that turn it into previz.

What the AI actually decides for you

Shot boundaries, framing suggestions, duration estimates, and a first-draft prompt per shot.

That's a real amount of work removed — but it's a draft, not a decision.

And what it does not

It doesn't know your brand, your budget, or which beat matters. It doesn't know that shot 5 is the one the client will fixate on.

Every tool that claims otherwise is describing a demo, not a workflow.

The Five-Step Workflow

Five steps, each needing an explicit action from you. There's no one-click button, and that's the design working rather than a feature missing.

Step 1 — Write the script so it breaks down cleanly

Any format works: screenplay, ad script, voiceover copy, or a list of beats.

What matters is that the action lines are concrete.

Concrete action lines beat scene labels

"She hesitates at the door" gives the breakdown something to work with. "Intro scene" gives it nothing, so it invents something generic.

Write what physically happens. The camera language comes later.

Step 2 — Choose style, ratio and models

Pick a visual style, an aspect ratio, and which models handle images and video.

Why model choice outranks every other setting

Veo 3.1 and Kling 2.6 have genuinely different strengths — dialogue and human motion versus stylised movement and camera work.

Getting this wrong costs more than any other setting on the page, and you can still change it per shot later.

Step 3 — Fix the breakdown before you spend

The script comes back as a numbered shot list with every field editable, including the generated prompt.

This is the step most tools hide, and the one that decides your bill.

Merge, split, reorder

Two shots that should be one? Merge them. One shot carrying three actions? Split it.

Editing text is free. Regenerating frames is not.

Consistency mode vs speed mode

Consistency is slower and much better at holding a character together. Speed is cheaper and drifts visibly.

💡 Pro tip: rewrite the prompts at this step using the shot prompt rules. It's the highest-leverage twenty minutes in the whole process.

Step 4 — Generate and confirm frames

Generate one at a time or in a batch. Review, regenerate what missed.

Why the confirm gate exists

Nothing proceeds to video until you've said a frame is right.

It's one extra click per shot, and it's what stops a bad frame from becoming an expensive bad clip.

Step 5 — Turn confirmed frames into video

Confirmed frames become the first frame of their video shot — the same image to video step you can run on any single picture.

One variable left

Because the frame is locked before the clip generates, motion is the only thing that can go wrong. That makes failures cheap to diagnose — if the clip is bad, it's the move, not the framing.

Then download everything and cut it in your editor.

A Worked Example

Here's what the process looks like on a short script.

📝 Note: this is a worked example built to show the shape of the output — the script, a typical breakdown, and the edits it would need. It is not a transcript of a specific generation run.

The script

INT. WORKSHOP — NIGHT

Rain against the windows. Mia sands a chair leg, stops, holds it up to the light. Not right. She sets it down and reaches for a different tool.

Forty words. Three beats.

The breakdown it produces

A script like that typically comes back as five or six shots:

What you'd change by hand

Almost always something. Here's what to look for.

The shots that need merging

Five and six are the same beat — she sees the flaw and reacts to it. Cutting between them adds a beat the script didn't ask for.

Merging saves a generation and tightens the sequence.

The shot that's missing

There's no reaction shot. The script says "not right," which is an internal state, and the breakdown gave it an object close-up rather than a face.

Add an MCU on her face between 5 and 6. Breakdowns under-generate reaction shots because scripts describe actions, not feelings.

The finished sequence

Six shots, about nineteen seconds, one character who looks like the same person throughout because the reference was locked before shot one.

That's the whole deliverable. Assemble in CapCut, Premiere or Resolve.

Storyboarding by Format

The five steps don't change. What changes is shot count and what you're optimising for.

Short-form vertical

Six to ten shots for thirty seconds. Frame 9:16 from the start rather than cropping later.

Keep sizes varied — vertical exaggerates the flatness of a single-size sequence.

Ads and product spots

Ad boards have structural obligations film boards don't.

The six-frame obligation

Miss the hook frame and nothing downstream matters, because nobody's still watching.

Explainers

Often stills plus voiceover, no video at all. Generate frames, skip step 5, save the credits.

Consistency mode keeps a recurring presenter stable.

Music and lyric sequences

The one case where drift between shots is a feature. Use speed mode and let the visuals diverge.

The Pre-Generation Checklist

Everything here happens while editing is still free. Run it once before you spend a credit.

Check the shot list

Does every shot name a size?

If a row says "shot of the workshop," it will come back as a medium shot, because that's what models default to.

Size is the one field where the default actively costs you.

Is any shot carrying two actions?

Look for "and," "then," and "while" in the action column. Each one is usually a shot boundary you haven't drawn yet.

Splitting costs a generation. Not splitting costs a frame that shows neither action clearly.

Are there reaction shots?

Breakdowns under-generate these, because scripts describe what people do rather than what they feel.

Read the sequence and ask where a face should be. Usually it's right after something goes wrong.

Check continuity

Is the character reference locked?

Before shot one, not after shot three. Everything downstream inherits from it.

Does the lighting clause repeat?

The same scene should carry the same lighting description in every shot. "Late afternoon, low sun through the window, warm rim light from camera left" — copied verbatim across all six.

This does more for continuity than any style keyword.

Do consecutive shots change size?

Two shots of the same subject at nearly the same size will jump-cut. Change by at least two steps: MS to CU cuts cleanly, MS to MCU jumps.

Check the money

Which shots actually need video?

Not all of them. Explainer sequences often work as stills, and a static product frame doesn't get better for moving.

Marking three of six shots as images-only halves that part of the bill.

Is the model right for the content?

Switching models after you've generated six shots means regenerating six shots.

📝 Note: the whole checklist takes about ten minutes on a six-shot board. Every item on it is free to fix at this stage and costs a regeneration afterwards.

Where AI Boards Still Fall Down

Three honest limits. None of them are solved by better prompting.

Character drift past a dozen shots

Sequential referencing compounds small errors. Around twelve shots the character has been photocopying itself.

Re-anchor to the original reference instead of continuing the chain.

Screen direction and the 180-degree rule

Models don't track screen direction across shots. Ask for a reverse angle and you may get your characters swapping sides.

You have to catch this at the board stage. The reference here covers the rule and when you can break it.

Anything that needs a real location

A specific building, a real product on a real shelf, a recognisable street. Generation approximates; it doesn't reproduce.

If the shot has to be that place, shoot that place.

Frequently Asked Questions

How long does it take to storyboard a video with AI?

The breakdown itself takes about a minute. Editing it takes ten to thirty, and that's the part that decides quality. Generating frames adds a few minutes per shot depending on model. Budget an hour for a six-shot board including the edits, not the five minutes the demos imply.

Do I need to be able to draw?

No. Drawing was the barrier that kept most people from storyboarding at all, and generation removed it. What replaced it is writing — you now need to describe a shot precisely, which is a different skill and a more learnable one.

How many shots does a 30-second video need?

Six to ten for narrative content. Ads run tighter at five to seven because each frame has a structural job. Explainers can go longer since stills hold longer than motion. Fewer than five in thirty seconds usually reads as static.

Can AI keep the same character across every shot?

Yes, with a character reference image and consistency mode, which passes each confirmed shot forward as the reference for the next. It holds well across a typical short sequence and degrades past roughly a dozen shots, so re-anchor to the original reference rather than continuing the chain.

What is the difference between a storyboard and a shot list?

A shot list is the text: numbers, sizes, moves, durations. A storyboard is that plus a visual for each entry. The list is what the crew works from; the board is what the client approves. AI workflows produce both from the same breakdown.

Can I use AI storyboards for client work?

Yes, and boards are one of the safer AI outputs to show clients because they're explicitly drafts. The gap is approval workflow — dedicated tools like Boords have version control and frame-level comments that generation tools generally don't. Export and share through whatever you already use.

Do I need a finished script to start?

No. A list of beats works — six lines describing what happens in order is enough for a usable breakdown. What doesn't work is a theme or a mood ("something energetic about running shoes"), because there are no actions in it to turn into shots. If you only have a concept, write the beats first; it takes ten minutes and it's the difference between a board and a mood board.

What do I do with the frames afterwards?

Download them and assemble in CapCut, Premiere or DaVinci Resolve. Generation tools are asset factories, not editors — transitions, text and music all belong in the edit, where changing them costs nothing.

Your Next Board

Take a script you've already written. Not a new one — one that exists, so you're testing the workflow rather than writing while you learn it.

Paste it into the AI storyboard generator, and spend your time on step 3. That's where the quality is decided and where the editing is still free.

If the prompts in the breakdown need work, the prompt library has forty you can borrow.