This Explainer Would Have Cost Thousands. It Took an Afternoon.
Here is the thing first, because it is the whole point.
A phone, a tap, and a packet of data travelling down through every system a card payment touches: the payment gateway, a fraud check that ticks off device, location and spending, the shop's own bank, the card network, and finally your bank. An approval races back up the outside, the phone says Paid, the shop gets its order, and a clock in the corner stops at 1.84 seconds. Then the camera pulls back and the whole system keeps quietly pulsing while you read it.
You have seen this style. Fintech launch films, product pages, the explainer at the top of a SaaS homepage. White canvas, line icons, one accent colour, everything easing into place. It is the look of a company that has a motion designer, and a motion designer for a thirty second piece like this has never been cheap.
The steps are real; the companies are not. Larkspan, Harbourline and Bramblewick Flowers do not exist, and the timings on each step are invented, though around two seconds end to end is about right. One honest simplification: those 1.84 seconds are an authorisation. Your bank says yes and holds the money, and the money itself moves later, usually a day or two after.
What I set out to do
I had just finished a two minute film for a football club's coaches, built the same way, and I wanted to know if the method held up somewhere that has nothing to do with football. So the brief was deliberately the opposite of a sports film: no photos, no footage, no colour beyond one accent. Just a clean white layout, some systems and some flows, beautifully animated. The sort of thing you would show a client to prove you can do premium.
The video is not generated. Claude wrote a program that draws every frame, and the program draws the same frames every single time.
That distinction is the one I keep banging on about, and I wrote the long version of it in Let AI do the hard work, use determinism for consistency. Short version: nothing in this video came out of a video model. Every card, every line, every icon is something you could open and change.
What the AI nailed
The video is a single web page. It has one job: given a moment in time, draw that moment. A headless browser asks it for 750 moments in a row, takes a picture of each, and ffmpeg stitches them into an MP4. No animation loop, no timers, nothing random.
It chose the restraint itself: a dotted off white canvas, cards with a hairline border and a soft shadow, indigo for the request, green for the approval, and the icons drawn as simple strokes in code rather than pulled from a library.
It follows the packet down through each system, which wakes up as the packet arrives, then follows the approval back up, then pulls out to show you everything at once. That pull back is the payoff, and it was designed as one.
The version on this page is 4:5 for the LinkedIn and Instagram feed. Adding the 9:16 version for Reels and TikTok was not a re-edit. It is the same file with one flag, which moves the headline below the platform's top bar and reframes the camera for the taller screen.
The finished file is 25 seconds and under 6 MB, sound included. White designs compress beautifully. It renders in a few minutes on an ordinary laptop, and the hundredth render costs exactly what the first did: nothing.
Where it went wrong
None of these were AI mysteries. All of them were visible the moment I looked at stills, which is why I now have it render a contact sheet of the whole thing before it renders a single second of video.
The lesson in that last one
A video that plays is not a video that is right. Count what you expected against what you got, every time, because the failures that matter are the ones that still look finished.
What I would tell you
If you have ever paid for, or been quoted for, a polished explainer of how your product or process works, it is worth knowing that the economics have moved. The design judgement still matters, and so does a person looking hard at every frame. The motion graphics part, the bit that used to need a specialist and a timeline in a desktop app, is now something you describe and then correct.
And because it is a program rather than a render you got lucky with, changing it later is an edit. Different amount, different brand colour, a step added to the flow, a version in Spanish. None of that means starting again.
If you want to see more of how we build things this way, the Starting Lineup work is where it started, and our AI integration page is the plain English version of what we do with it. Or just tell me what you are trying to explain.