AI Storyboard Generator From a Script: A Video Producer's Workflow

Split the script into numbered scene beats with a chat assistant, generate one frame per beat with an image tool, and keep a human artist for the frames that must be exact or must match: product close-ups, screens and on-screen text, hands doing a specific job, recurring characters, camera moves and the key frames the client signs off.

The split matters because a storyboard is a planning document. Its job is to get the client to agree the sequence, the message and the look before anyone books a crew. An image generator is quick at mood and composition, but each frame is a separate image, so continuity, exact product detail and screen direction stay with a person.

Follow me on Instagram@sagnikteaches

From a 150-word script to twelve numbered beats

Take a 60-second launch film for a client's soil-moisture sensor and its companion app. The voiceover is 150 words, which at an unhurried narration pace of about 150 words a minute fills the minute. A beat is one visual idea that holds the screen for roughly three to eight seconds, and a minute-long explainer usually lands between 10 and 16 of them. Treat that as a rule of thumb, not a standard, and adjust it to the pace your editor likes.

Connect on LinkedInSagnik Bhattacharya

The first job for the assistant is the split, not the pictures. Paste the approved script and give it rules, including the one that stops it rewriting the client's words:

Subscribe on YouTube@sagnikteaches

You are a storyboard planner for a video production company.
Below is a client voiceover script (150 words) for a 60-second film.

Split it into numbered beats. One beat = one visual idea.
For each beat give: beat number, seconds (total must be 60),
the voiceover line it covers, shot size (wide, medium, close-up, insert),
what we see in one sentence, and any on-screen text or logo.

Rules:
- Do not add voiceover lines or change the client's words.
- Write EXACT beside any beat that shows the real product, a logo,
  a phone screen, a chart or written text.
- Write MOVE beside any beat where the camera moves or the action
  changes between the first and last second.
- Write CONTINUITY beside any beat with a character who appears
  in more than one beat.
- Output a table. Then list any script line you could not turn
  into a picture and say why.

SCRIPT:
[paste the approved script]

Here is an illustrative result for the sensor film. The last column is the route each frame will take: route A means the generator alone is enough, route B means the generator drafts it and an artist corrects it, and route C means an artist or designer makes it.

Check the seconds column before anything else. Assistants can be unreliable at adding up a column, so a first pass may total 63 or 66 seconds. If it does, ask for a re-timed table rather than trimming by hand, because a hand edit leaves the voiceover lines out of step with the beats.

Check, too, that no voiceover line was dropped or invented. Paste the table's voiceover column into a blank document and compare it with the client's script. If a line has gone missing, the picture for it will be missing from the board, and you will find out in the approval round.

Which frames a generator can draw and which go to an artist

Capability is the wrong test, because image tools improve every few months. The useful test is risk: what does it cost if this frame is slightly wrong? Six questions sort most frames. Does the frame need to match a real product, logo, screen or written wording exactly? Does a person or object have to look the same in five or more beats? Does the action have to be physically right? Does it move over time? Does it have to cut cleanly against its neighbours? Will someone downstream build from it?

A yes to any of them moves the frame towards an artist. These are the frame types where I would not rely on a generator alone, with a cheaper route for each:

Apply the six questions to the twelve beats and the flags from the table above do most of the work. Beats 4, 5, 6, 8 and 12 carry EXACT flags, so they take route C. Beats 2, 7, 9 and 11 feature the gardener across several shots, so they take route B: generate against a locked character sheet, then let an artist correct the drift. Beats 1, 3 and 10 are atmosphere with no continuity burden, so route A is enough.

The saving comes from what route C actually means. Beats 6, 8 and 12 are layouts, not illustrations, and the motion designer has to build them for the edit anyway. Only beats 4 and 5 need a person to draw, and for beat 5 a phone photo of a hand pressing the real sensor into a flowerpot gives the artist an accurate pose to trace. Pulling the designer's work forward costs very little. The bill for drawn frames is two frames, not twelve.

A four-frame test to run on your own tool

Before trusting any generator with route A or B, spend half an hour on four test frames using a past client's assets, not a live job. Test one is the product: give it three photos and ask for the product on a kitchen counter, then lay the result over the photo and check the outline. Test two is the words: ask for a phone screen reading exactly "Dry. Water in 2 hours." and check each letter. Test three is the character: draw the same person in six different situations and compare faces, clothes and props side by side. Test four is a camera move: ask for the first and last frame of a slow push-in and see whether the room and the subject stay the same.

Score each test pass, partial or fail. Suppose the tool passes the character test in four of six frames and fails the text test twice in three tries. That tells you the character beats can go through route B, with a fix in every frame, and that anything containing words stays on route C. Re-run the test when the vendor announces a new model, since the result is a property of the tool this quarter, not a rule.

A consistency sheet that travels with every prompt

Consistency of characters and brand is the complaint producers hear most about AI boards, and the fix is dull: one short sheet, pasted at the head of every frame prompt, plus reference images uploaded alongside it. Filled in for the sensor film, with illustrative values, it looks like this:

CHARACTER: The gardener. Mid-fifties, short grey hair, round wire glasses,
tan canvas apron over a navy shirt. Carries a green watering can in every
scene she appears in.
LOOK: Soft natural daylight, shallow depth of field, muted greens.
Warm orange appears only on the sensor's status light.
PALETTE (from the client's brand guidelines): greens #2E7D32 and #A5D6A7,
off-white #F6F4EE, accent #F57C00.
FORMAT: 16:9, no borders, no captions.
NEVER INCLUDE: text, logos, brand names, watermarks.

A frame prompt is then the sheet plus one line each for shot, setting and action. For beat 7 it might read: "Medium shot, eye level. The gardener stands at a kitchen counter holding a phone in one hand and a mug in the other, morning light through a window on her left. Leave the phone screen blank." Asking for a blank screen is deliberate. A generated screen will be wrong in some way, and a blank rectangle is easy to replace later.

A typical first result, again illustrative: the face and glasses match the sheet, but the apron has turned olive green, the watering can is missing, and the mug is held in the wrong hand. Resending the whole sheet with a line such as "apron is tan, the can sits on the counter behind her, mug in her right hand" usually fixes it in a couple of attempts. If the third try is still wrong, stop regenerating. That frame belongs on the artist's list.

Tools help in different ways. Boords, a storyboard-specific tool, has an Asset Library where you save characters, locations and props once and type "@" in a prompt to reuse them. Midjourney's documentation says its Edit model takes up to four reference images (as of November 2026; check the current page). General assistants such as ChatGPT or Gemini accept an uploaded reference image in the same conversation, and the strength of the match varies, which is exactly why the test above exists.

Keep a consistency log as you go, one row per frame, so the check does not live in anyone's head:

If the client also wants vertical cut-downs, make a second board in the new frame shape. Cropping a 16:9 frame to 9:16 usually slices through the subject, and regenerating one frame at a time breaks the match you worked for. The new board needs its own consistency log.

What a 12-frame board costs on each route

Image tools are cheap next to the hours around them. These are list prices in USD as checked on the vendors' own pages in October and November 2026; allowances change, so check before you buy.

Count images, not frames. A 12-frame board at about four attempts per frame is 48 images, so Boords Solo's 250 monthly images stretch to roughly five boards. One caution on Boords: its conversion from script to storyboard takes a snapshot of the script, and later script edits do not update existing frames. Lock the script version before you convert, or convert the revised draft into a new board.

The tool is the small number. Time and artist fees decide the cost. These sums use made-up rates, so swap in your own: producer time at $45 an hour, an artist at $60 an hour, and about 45 minutes for an artist to draw a frame from scratch.

The artist time in the last row is 45 minutes each for beats 4 and 5 plus 15 minutes of correction on each of the four continuity beats. The generator-only route is $150 cheaper and leaves five frames at risk, the ones the client will study hardest. Time a pilot on one real board and replace my numbers; the ratio is what matters, not the dollars.

The approval round: what the client signs and what they do not

Clients treat any picture as a promise. The cover note has to say what the frames promise and what they do not. The wording I would use puts it in two lists. The client is approving the running order, the message in each beat, the on-screen wording, the product shots that must appear and the type of location and talent. The client is not approving faces, exact colour, the graded look or any generated product, logo or screen, all of which come from their own supplied assets.

Subject: [Film name] storyboard v1 for your approval

Hi [first name],

The storyboard is attached: 12 numbered frames, each with its
voiceover line underneath.

Please approve or comment on:
1. The running order and timings
2. The message in each beat
3. The on-screen wording (frames 6, 8 and 12)
4. The product shots we have planned (frames 4 and 5)

Frames are indicative of composition, mood and sequence. Products,
logos, screens and text will be produced from your supplied assets,
and faces are placeholders. Two rounds of changes are included.
Please send one consolidated set of comments by [date].

Thanks,
[your name]

Each round is a new version: v1, v2, v3, with the changes listed at the top. Boords advertises no-signup client reviews, comments and version history on its paid plans. A PDF plus a shared comment sheet with one row per frame does the same job. Whatever you use, insist on one consolidated list of comments, and see the tutorial on consolidating client revision feedback with AI if the comments arrive in six emails. The number of included rounds belongs in the proposal, and the tutorial on drafting the scope, budget and timeline is the place to set it.

Once the client signs off, freeze the board. The frame numbers become the shot numbers, and the approved beats feed the shoot-day plan described in building a shot list from the script. If the script itself is still moving, finish that first: going from a brief to a script draft sits one step earlier in the same workflow.

A shootability check before the board goes out

An AI frame can show something your crew cannot produce, and a client who has approved it will expect it. Spend fifteen minutes with the director or the director of photography on every frame before the board is sent. Ask whether the lens implied is possible in that room, whether the location exists and is booked, whether the light in the picture matches the time of day you can shoot, and whether the camera move needs kit that is not in the budget. Ask also whether any prop or costume is in the shot that is not on the list.

An invented example of how this goes wrong: a generated frame showed a kitchen with a long window wall, a low dolly move towards the gardener and a reflection of her face in the glass. It looked good, and the client approved it. The location the team had booked had one small window, and the only camera support in the quote was a gimbal. The reflection was impossible without a studio build. The fix at the board stage costs a sentence. Change the frame to a side-on medium shot, note "gimbal push-in, no reflection" under it, and send it as v2 before the client falls for the original.

Write the result of the check under each frame in plain words, for example: "Frame 9: gimbal push-in, 35mm equivalent, garden bed B, needs 10 minutes of sun after 4pm." Those notes become the first draft of the shot list.

Rights, data and the sentence your contract needs

Three things to settle before a client's script goes into any tool. First, whether the tool keeps or trains on what you paste. Business plans of ChatGPT and Claude, and Gemini in Workspace, do not train on business content by default, while personal ChatGPT and Claude plans have a model-training switch in their privacy settings that you should turn off for client work. Second, the tool's own terms. Midjourney's revenue rule above is one example, and Adobe says its commercial-safety claim covers its own Firefly models, not the partner models offered inside the same app. Third, the contract: one sentence stating that pre-production frames may be AI-assisted and are used for approval only, with the final film produced from the client's supplied assets and your own footage.

That sentence also shapes the disclosure conversation. If the frames are internal planning tools and never published, disclosure to the public is not the issue, but your duty to the client is, so tell them. Keep prompts free of living artists' names and other films' titles, because a board built on "in the style of" prompts is hard to defend when a client asks where the look came from. If a client wants AI-made images in the finished film, that is a different job with different rights and a different quote, and the software comparison for small studios is a better place to start.

Storyboard questions from producers

How many frames does a one-minute film need?

Most producers land between 10 and 16 frames for a minute, one per beat, where a beat is a single visual idea holding the screen for roughly three to eight seconds. It is a rule of thumb rather than a standard. Fast-cut ads need more frames, slow interview-led films fewer, so count beats in the script first and let the frame count follow.

Should the client be told the frames are AI-made?

Yes, in the cover note and in your terms. Say the frames are indicative, made with AI assistance and checked by your team, and that products, logos, screens and on-screen text will be produced from the client's own assets. It avoids a dispute later and costs one sentence. Check the client's contract for any clause about AI use in pre-production.

Can I use free AI image tools for client work?

Free tiers are fine for a private test, but read the usage terms before client material goes in. Many tools limit commercial use, add watermarks or use inputs for training unless you switch it off. Client scripts are often unreleased, so a business plan with training off by default, or the training switch turned off, is the safer route.

Is a hand-drawn storyboard still worth offering?

For frames that carry the idea, yes. A client who is paying for a brand film often wants the hero frames drawn with care, and a human storyboard artist also thinks in cuts, screen direction and camera moves. Offer it as the premium route for key frames while AI covers the routine ones.

Further reads

• Best AI Filmmaking Tools by Job for a Small Production Business — Which AI tools suit which production jobs, beyond storyboards.
• Video Production Pricing Sheet: Build a Rate Card With AI — Where storyboard rounds and revisions sit on a rate card.
• AI Rough Cut From a Transcript: First Edits for Small Video Teams — The edit stage after the shoot, starting from a transcript.
• Videographer Software: Client Review, Approvals and Invoices Compared — Review and approval tools compared, for the sign-off step.
• AI Subtitle Generator for Client Videos: Accuracy and Delivery Checks — Delivery checks on captions once the film is finished.
• AI Video Content Agency: Services a Small Studio Can Add and Price — AI services a small studio can add, and how to price them.
• AI Tools for Filmmakers: Script Breakdowns, Shot Lists, Call Sheets — Assistants or production platforms for breakdowns, shot lists and call sheets on a small paid shoot, with worked examples and what to check by hand.
• AI Voice Over for Videos: Can a Production Company Use It for Clients? — AI voice over suits client videos when the voice is licensed or consented, the client agrees in writing and platform labels are followed. The steps.
• AI Commercials for Small Brands: A 30-Second Ad From Script to Edit — Take one small-brand offer from script and storyboard through generated clips, voiceover, edit and disclosure checks.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Boords pricing and AI storyboard generator pages, Midjourney plan comparison and Edit model documentation, ChatGPT, Google AI and Adobe Firefly plan pages (checked October to November 2026).