A script-to-storyboard workflow turns written scenes into visual frames before anything gets filmed or generated. You break the script into shots, fix the character and location as reference images, then generate one frame per shot. The board gets approved, and only then does each approved frame become video. In DesignerBox a brand storyboard costs less than any single video clip.
Many teams skip the board. They write a script, paste the first scene into a video model, and pay for an 8-second clip where the product is the wrong color. Then they fix the prompt and pay for another. Three rounds in, they have paid for three clips nobody approved, and the board that would have caught it costs less than one of them.
The board exists so that being wrong is cheap. This guide walks the script to storyboard workflow end to end: breaking the script into shots, locking what has to stay consistent, generating the board, and turning an approved frame into the shot. It is written for marketing leads and agency creative directors who need a client or a stakeholder to sign off before the expensive part starts.
Key Takeaways
- The storyboard is the cheap place to be wrong. Fix a shot on the board, before you pay for the clip.
- Human boards carry a rework clause for a reason. The Animation Guild’s published minimum for a half-hour subject’s storyboard is $4,686.94 from 2 August 2026, and an employer who wants a third revision pays 20% of the unit maximum for each one (animationguild.org, accessed September 2026).
- Lock the character before the frames, not after. Nano Banana Pro takes up to five character images to keep characters consistent, and outputs at 1K, 2K or 4K (ai.google.dev, September 2026).
- An approved frame is a video input, not a picture. Veo 3.1 takes a still as the first frame and accepts up to three reference images of a person, character or product, with audio always on (ai.google.dev, September 2026).
- Board the whole script, generate one shot. Approve the full board first, then spend video credits only on the shots that survived.
- Two-in-three video buyers now use generative AI for creative, up from half in 2025, and a third of ad assets use it this year (IAB 2026 Digital Video Ad Spend and Strategy Report, July 2026).
What a script-to-storyboard workflow is
A storyboard is a sequence of still frames, one per shot, that shows what the camera sees before anyone shoots it. A script-to-storyboard workflow does that with generated images instead of drawn panels. You supply the script, a reference photo of the product or character, and a description of each shot. The model returns a frame per shot with the same subject in every one.
A client approves the board, so the ad gets made only once.
That order matters more with AI than it did with pencils, because the expensive step moved. A drawn board was the costly part of pre-production and the shoot was budgeted separately. Now the board is nearly free and the generation is what burns the budget. Boarding first reverses the old economics.
What a storyboard costs when a human draws it
Commissioned storyboards are priced per unit of finished screen time, not per panel. The Animation Guild, Local 839 IATSE publishes its wage scales openly, so its numbers can be checked.
| Subject | Union minimum, storyboard only |
|---|---|
| Short, 4 to 7 minutes | $1,959.79 |
| Short, over 7 to 15 minutes | $2,468.16 |
| Half-hour subject | $4,686.94 |
| One hour or more (TV) | $6,990.89 |
| Theatrical, 75 minutes or more | $9,410.82 |
Rates for the 2 August 2026 to 31 July 2027 period, for programs licensed on or after 1 August 2022 (animationguild.org, accessed September 2026). These are US animation minimums, not advertising rates, and they are floors rather than market prices.
The clause worth reading twice is the revision one. The agreement lets an employer require two re-works after presentation at no extra cost. Beyond that, each additional re-work costs 20% of the original unit maximum (animationguild.org, accessed September 2026). Someone wrote that clause because changing your mind about a board three times is expensive enough to need a formula.
Per-panel prices are harder to check. Treat any per-panel dollar figure that does not name its publisher as unsourced.
Step 1: Break the script into boardable shots
A shot is one continuous camera position. Every time the camera cuts, moves to a new setup, or changes what it is looking at, that is a new frame on the board.
Read the script and mark a break at each of these:
- A cut. New angle, new frame.
- A location change. Even the same subject in a new place is a new setup.
- A subject change. Product to face, face to hands, hands to logo.
- A beat the voiceover lands on. If the script says something the viewer needs to see, it needs a frame.
A 30-second social ad usually lands somewhere between 6 and 12 shots once you do this. Resist boarding every micro-movement. The board’s job is to answer “does this sequence work”, not to specify every frame of motion.
Write each shot as one line before you generate anything: subject, framing, action. “Close on the bottle, hands entering from the left, cap twisting off.” That line becomes the prompt. For the full list of what each line needs, see the script breakdown for AI video ads.
If the script itself is not written yet, the offer and the message come first, and how to make an ad in seven steps covers both.
Step 2: Lock the character and the location before the frames
This is the step teams skip, and it is a common reason AI boards fail review. Generate eight frames from eight independent prompts and you get eight subtly different products, eight faces, eight rooms. The client sees drift, not a sequence.
Fix the references first. Upload the actual product photo. If a person appears, fix that person as a reference too, and do the same for the location.
Google documents Nano Banana Pro as taking up to five character images to keep characters consistent, with output at 1K, 2K or 4K (ai.google.dev, accessed September 2026). Google DeepMind also states plainly that the model can still struggle with small faces, accurate spelling, and fine details (deepmind.google, accessed September 2026), which is worth knowing before you board a shot that hangs on a legible label.
For frames that are edits of an earlier frame rather than fresh generations, Black Forest Labs documents FLUX.1 Kontext as preserving identity across multiple scenes and environments (bfl.ai, accessed October 2026). Its documentation says the model holds character consistency even after multiple edits (docs.bfl.ai, accessed October 2026). That is the behavior you want when shot 7 is shot 3 from a different angle.
Two companion guides go deeper on the failure modes here: keeping characters consistent across AI video shots and holding a location steady between frames.
Step 3: Generate the frames
With the references locked and the shot list written, generate the board as a grid rather than as loose images. A grid is what a reviewer can read at a glance, and it is what forces the shots to be judged as a sequence. Tools differ on what they take in and what the board comes back as, and our comparison of six AI storyboard generators sorts them by that split.
The brand storyboard app does this in one run, as a grid of 4 to 25 frames. The grid size sets the cost. Even a full 25-frame board costs less than one video clip.
Review the board against the script line by line. The three questions that catch the most problems:
- Does the product look like the product, in every frame?
- Does the sequence read without the script next to it?
- Is there a shot here that a video model will not be able to hold for eight seconds?
Reboard and rerun until the board passes. Every revision here happens before you spend video credits. Fewer re-rolls at the video stage is also how to make AI videos fast.
Step 4: Turn the approved frame into the shot
An approved frame is a first frame, and the approved set can be timed as an animatic with a scratch voice before any clip is made. This is where the board stops being a planning artifact and starts being production input. The steps from a product ad script to a joined cut, with one voice track and captions, are in the guide to script to video AI.
Veo 3.1 accepts a still image as the starting frame and takes up to three reference images of a single person, character or product. Durations are 4, 6 or 8 seconds, and 8 seconds is required, not merely available, once you go to 1080p or 4K or pass reference images. Audio is generated natively and is always on (ai.google.dev, accessed September 2026). Reference images work on both Veo 3.1 and Veo 3.1 Fast. Veo 3.1 Lite does not take them. The cheaper Fast variant does not cost you the reference input.
Seedance 2.0 works differently and suits a different kind of board. ByteDance documents it as producing 15-second multi-shot audio and video output, taking up to nine images, three video clips and three audio clips as simultaneous input (seed.bytedance.com, published February 2026, accessed October 2026). Multi-shot means cuts inside a single generation, so a three-frame section of your board can come back as one continuous piece rather than three clips to assemble.
DesignerBox runs both among its video models, and the model list shows what each one does. Board once, then pick the model that fits each shot.
Google’s own Creative Lab ran this exact split for its Pixel 10 spot Vanilla, building storyboards with Veo 3 and then hiring human directors to shoot the finished ad (Campaign US, campaignlive.com, September 2026). Our breakdown of the most-discussed AI commercials covers why the hybrid pattern keeps appearing.
Cost of a boarded 30-second ad
Here is what sets the cost of a 30-second ad with eight shots, boarded and then generated.
| Stage | What sets the cost |
|---|---|
| Reference | The image model you pick for the product photo edit |
| Board | The grid size, from 4 to 25 frames |
| Board revision | The same as the first board |
| Video | The model and the clip length. An 8-second clip costs 40 to 560 credits, depending on the model |
The video stage is almost all of the bill, and the model you pick moves it more than anything else. You see what you are about to spend before the run starts.
Clip length and model choice drive the bill more than shot count does. The credit side is covered in our guide to what AI video costs in credits.
Two plan gates matter for this workflow. Uploading your own photos and the commercial license start on the Pro plan. AI video, virtual try-on, upscaling, the image editor and the video editor start on the Premium plan. The board itself runs on any plan. On the free plan it runs on sample products. Plans and credits are on the pricing page.
Where AI storyboards still break
Four limits to know before the video stage. A board that promises too much gets rejected at the video stage instead of the board stage.
Text inside the frame. Google states Nano Banana Pro can struggle with accurate spelling (deepmind.google, September 2026). If a shot depends on legible packaging copy, treat the board frame as a layout placeholder and confirm the text at the final render.
Continuous motion across a cut. A board shows positions, not physics. A frame can look right and still describe a move that no video model will hold for eight seconds. The reviewer question in step 3 catches most of these.
Small faces in wide shots. Documented as a weak spot in the same source. Board the wide, but verify the face on the closer coverage.
Board frames are approval artifacts. They sit at board resolution and board fidelity. The finished asset comes from the video stage or a fresh generation at full resolution. Do not upscale a panel from the grid.
Boarding as a repeatable step
The version of this that pays off is the one that runs every campaign. Save the reference set, the shot-list format and the board size as a workflow. Then run it again for the next product with a new reference photo. A saved workflow runs the same way each time, so the fortieth product gets the same framing, the same character lock and the same review pass as the first. Workflows hold the whole sequence. Your brand rules sit in one record, and the workflow reads them on every run. The board, the approved frames, the clips and the cut stay in one place. The full workflow from the first product photo to the finished ad, in one subscription. The AI video ads page shows how a product photo becomes a short video ad.
The IAB reports that nearly two-in-three video buyers now use generative AI for digital video creative, up from half in 2025, with a third of ad assets using it this year and a projected 43% by 2027 (IAB 2026 Digital Video Ad Spend Report, July 2026). At that volume the board stops being a planning nicety and becomes the control point where a campaign either gets cheap or gets expensive.
Prompt craft is the other half of this. Our guide to writing AI video prompts that hold up covers the shot-description language that survives the jump from frame to clip. For the B2B end of this, AI video tools for LinkedIn covers what changes when the audience is a professional feed.
Start from a template, add your product photo and your script, and run it.
FAQ
What is a script-to-storyboard workflow?
It is the process of turning a written script into a sequence of still frames, one per shot, before any video gets generated or filmed. You break the script into camera setups, fix your product and character as reference images, generate a frame for each shot, and review the whole board as a sequence. Approval happens on the board, not on the finished video.
How many frames does a 30-second ad need?
Count camera setups, not seconds. A 30-second social ad typically resolves to 6 to 12 shots once you mark every cut, location change, subject change, and voiceover beat that needs a visual. No production body publishes a standard panel count, so build the count from your own script.
Can AI keep the same character across every storyboard frame?
Yes, with references locked before generation. Google documents Nano Banana Pro as taking up to five character images to keep characters consistent (ai.google.dev, September 2026), and Black Forest Labs documents FLUX.1 Kontext as preserving identity across multiple scenes and holding character consistency after multiple edits (bfl.ai, September 2026). Drift almost always comes from generating each frame as an independent prompt.
How much does an AI storyboard cost?
In DesignerBox the grid size sets the cost of a brand storyboard, and you see it before the run. A full 25-frame board costs less than any single video clip. For comparison, the Animation Guild’s published union minimum for a half-hour subject’s storyboard is $4,686.94 from 2 August 2026, with an additional 20% of the unit maximum charged for each re-work beyond the second (animationguild.org, September 2026).
Can a storyboard frame be turned directly into video?
Yes. Veo 3.1 takes a still image as the first frame of the generated clip and accepts up to three reference images for a character, object or scene, at 4, 6 or 8 seconds with audio generated natively (ai.google.dev, September 2026). Sora 2 Pro also took an input image that acted as the first frame of the video (developers.openai.com, September 2026), but OpenAI removed Sora 2 and Sora 2 Pro from its API on 24 September 2026 (OpenAI API deprecations, accessed September 2026). The approved frame becomes production input rather than a discarded planning artifact.
Which plan do I need to storyboard and then generate the video?
The board runs on every plan, because it is an image run. On the free plan it runs on sample products, and uploading your own photos starts on the Pro plan. AI video starts on the Premium plan. The video stage sets the plan you need, because video cost depends on the model and the clip length.
Should I board every shot or only the difficult ones?
Board every shot. The cost argument only works at full coverage, because the shots that break at the video stage are rarely the ones you would have flagged as difficult. An 8-second clip costs 40 to 560 credits, depending on the model. Partial boarding saves an amount too small to measure and forfeits the review pass that catches the expensive mistakes.
Sources
- Storyboard union minimums by subject length for 2 August 2026 to 31 July 2027, and the re-work clause, from the Animation Guild Local 839 published 2024-27 wage scales: (animationguild.org, September 2026)
- Nano Banana Pro’s five character images and 1K, 2K and 4K output: (ai.google.dev, September 2026)
- Nano Banana Pro’s documented weak spots on small faces and spelling: (deepmind.google, September 2026)
- Veo 3.1 taking a still as the first frame, up to three reference images, 4, 6 and 8 second durations, and native audio: (ai.google.dev, September 2026)
- FLUX.1 Kontext preserving identity across scenes: (bfl.ai, October 2026). Holding character consistency after multiple edits: (docs.bfl.ai, October 2026)
- Seedance 2.0’s 15-second multi-shot audio and video output and its image, video and audio input limits: (seed.bytedance.com, published February 2026, accessed October 2026)
- Sora 2 Pro accepting an input image as the video’s first frame: (developers.openai.com, September 2026)
- Sora 2 and Sora 2 Pro API removal on 24 September 2026: (OpenAI API deprecations, September 2026)
- Google Creative Lab’s Veo 3 storyboards for its Vanilla Pixel spot: (Campaign US, campaignlive.com, September 2026)
- Generative AI use among digital video buyers, the share of ad assets using it, and the 2027 projection: IAB 2026 Digital Video Ad Spend Report, news release of 14 July 2026 (iab.com, accessed October 2026)
- DesignerBox plan gates for video and the commercial license (DesignerBox pricing page, designerbox.ai/pricing, September 2026)
Model capabilities and storyboard rates re-checked as of September 2026. The FLUX.1 Kontext, Seedance 2.0 and IAB figures were re-checked on 2 October 2026. Industry figures from the IAB 2026 Digital Video Ad Spend and Strategy Full Report, July 2026. Individual results vary.
