The 6-beat script for 60-90 second product demos (template + worked example)

Last week a founder asked me to script their demo video. I sent back six beats with timecodes. They shot it themselves in one sitting and it outperformed their polished launch film — because structure beats production value every time someone is deciding whether to scroll.

Here's the template, plus a worked example from a real first-run demo I scripted this week (product anonymized to "a recipe app" — the structure transfers).

The six beats

0-5s — cold open. The single most surprising result the product produces. No logo, no intro, no "hi, I'm…". The first frame has to earn the second one.

5-15s — the friction. The world before the product, in one real shot. Not a stock photo of someone stressed — the actual ugly manual way the problem gets solved today.

15-30s — the turn. First interaction, real input, unedited. Screen only, cursor visible. This is where you prove nothing is faked.

30-60s — payoff montage. 2-3 fast cuts of the result being used in context. Cut on action, not after it — the cut itself creates pace.

60-80s — the honest wall. One thing the product can't do yet, on camera. This is the credibility beat. Almost nobody includes it; that's exactly why it works.

80-90s — CTA. Logo, one line, one action. If your CTA needs two sentences, your positioning isn't done.

Worked example: a first-run demo

The app: point your camera at your fridge, get recipes you can cook with only what's in it. Here's how the beats filled in:

• Cold open: fridge door opens → snap → "3 recipes from this" appears. VO: "This fridge just became a menu."
• Friction: one shot of a person staring into an open fridge. VO: "You check the fridge three times a day. You still order in."
• The turn: the app opens, photo taken, the scan list builds unedited. VO: "One photo. No typing."
• Payoff: three cuts — the expiry-aware flag on old produce, a recipe card with every ingredient already checked, the "only what I have" filter on with every card still cookable.
• Honest wall: a turned-away jar gets a wrong read — and the app asks instead of guessing. VO: "It reads labels, not minds."
• CTA: logo, "Cook what you already have.", install link.

The honest wall was free: the founders already had that footage from their own QA testing. If you've ever watched users try to break your product, you have yours too.

Three rules that outrank the beats

1. Portrait, 9:16. The same video cropped to landscape loses 60% of its placements.
2. Captions on everything. Most first views are muted. No captions, no message.
3. Every second is real UI. AI-generated b-roll and mockups read as fake in under a second, and viewers punish it. The turn (beat 3) is where trust is won — keep it uncut.

That third rule is also why "one prompt makes your trailer" tools keep disappointing people: they can generate footage, but the honest wall and the uncut turn are the two beats that require your product to actually work on camera. Use the tools for everything around those beats.

I cut demos like this end-to-end (script, VO, capture, edit) — the portfolio is here, first cut in 48h. And if you're earning in crypto, Superteam Earn pays USDC for content and demo-video bounties — that's my referral link, it credits me if you make your first submission.

Take the beats and shoot it yourself, though. Structure is the part that was expensive.