Storyboard a short explainer before generating its visuals
You've been asked for a one-minute explainer video, and an image tool will happily produce a smiling customer in seconds. Decide what each scene has to say first, or the pictures will make promises your policy doesn't.
What you'll make
a four-to-six-scene storyboard in which every scene has one message, a source rule, a visual idea, its on-screen words and what it must not imply.
What you'll need
the facts or policy the explainer is about, who will watch it and how (video, slides, a help page), and a work AI chat. You can plan entirely in chat before touching an image or video tool. About twenty minutes.
Example result.
| Field | Scene 4 of 4 |
|---|---|
| Message | Your refund arrives within 5 working days of your parcel reaching us |
| Source | R4 |
| Visual idea | A parcel arriving at a warehouse, then a calendar with five days marked |
| On-screen words | “Refund: within 5 working days of arrival” |
| Must not imply | An instant refund, or a refund from the day you post it |
1. Ask for the storyboard against numbered rules
Number the rules in your source (R1, R2…) so every scene can point to one. Then ask:
Storyboard a short explainer for [audience and medium]. After watching, they should understand [outcome].
Use only these rules: [rules]. Give [number] scenes. For each scene give: the one message, the rule number it comes from, a visual idea, the on-screen words, and what it must not imply.
Keep every exception and time limit as written. Do not add promises, benefits or features that the rules don't contain. Don't create any images yet.[audience and medium] is who and how, such as “customers, 60-second video”. [outcome] is the one thing they should be able to do. [rules] is your numbered list.
Why this works: “what it must not imply” makes the model say out loud the overpromise a picture could make, before anyone draws it.
What you should see: a table with one row per scene, each pointing to a rule number.
2. Read just the messages, in order
Read the message column on its own, top to bottom. Does each scene add something, and does the order make sense to someone who knows nothing? Then check each scene’s rule number against the source: does the message say what that rule says, with its limits?
Why this works: if the messages don’t make sense as a list, no visual will rescue them. Checking each against its rule catches the scene that has quietly rounded “within 5 working days” up to “fast”.
What you should see: a sequence that would make sense read aloud, and every message matching its rule.
3. Make the visuals one scene at a time
Once the storyboard is right, take it to the tool that makes the final piece: an image tool such as ChatGPT or Gemini for stills, PowerPoint or Google Slides with recorded narration, or the video tool your organisation uses. Make one scene, check it against that scene’s “must not imply”, then move on. Keep words as editable text over the image rather than inside it.
Why this works: a generated picture can contradict a correct script. Checking one scene at a time keeps the contradiction small and easy to fix.
What you should see: a visual for each scene that shows its message and nothing its “must not imply” rules out.
Check the result
- Every scene points to a rule number, and its message matches that rule.
- Every exception and time limit in the source appears in some scene.
- No scene, visual or on-screen word promises more than the rules.
- The messages make sense read in order on their own.
The usual failure: the last scene oversells. Endings pull towards a happy promise, so “refund within 5 working days of arrival” becomes “refund within 24 hours”, with a delighted customer checking their phone. Spot it by reading the final scene’s message against its rule word for word.
Scene [number] changes rule [rule] from "[rule wording]" into a stronger promise. Rewrite that scene's message, visual idea and on-screen words so they say exactly what the rule says. Keep the other scenes as they are.[number] is the scene, [rule] its rule number, and [rule wording] the rule’s exact text.
Worked example: the 24-hour refund
Fictional example. A homeware shop wanted a 60-second returns explainer for customers. Its rules:
R1 Return within 30 days of delivery.
R2 Items must be unused and in their original packaging.
R3 Start the return in your account under Orders, then print the label.
R4 Refunds go to the original payment method within 5 working days of the item reaching our warehouse.
R5 Made-to-order items can't be returned unless they are faulty.The first storyboard had four scenes: who can return (R1, R2), starting the return (R3), packing and posting, and the refund.
- Scene 4 promised too much. Its message was “Get your money back within 24 hours”, with a customer smiling at a refund notification. R4 says 5 working days, counted from arrival at the warehouse. The repair prompt, with scene 4 and R4 filled in, gave the row in the example result at the top of this page.
- An exception was missing. R5 appeared in no scene. Adding “except made-to-order items, unless faulty” to scene 1 kept the order sensible: who can return comes first.
- The visuals came last. Only once all four messages matched their rules did the scenes go to an image tool, one at a time, each checked against its “must not imply”.
Task kit for this job: Download (ZIP) · Reusable instructions
Go deeper
Keep the picture and the narration saying the same thing
If the narration says something is conditional, the picture and on-screen words mustn’t look certain. A correct voice-over over a picture of an instant refund still teaches “instant”. Review each scene’s words, picture and narration together.
Keep people and products consistent across scenes
Once the message is sound, describe the people and products that recur, and change only what each scene needs. Check each output against the last. See consistent image sequences.