Behind the build
A launch film for an idea nobody could picture
Persona Lobby pressure-tests propositions against synthetic persona panels grounded in real population data. That sentence loses a room. This is how it became seventy-five seconds a founder can judge without trusting a word of it.
The product is genuinely hard to explain. Synthetic personas, psychometric grounding, census-weighted panels: every accurate description of it sounds like a research paper, and founders do not buy research papers.
So the brief was narrow. Land the idea in under ninety seconds, using one ordinary scenario a viewer already understands, and answer the sceptic's question, how do you know it is accurate, inside the film itself rather than in a follow-up call. The delivered cut runs seventy-five.
Open on a coffee shop loyalty app, not on a data pipeline. The abstraction arrives after the viewer already cares.
Name the real population data and the human validation step on screen, so the accuracy objection is met before it is raised.
Typography, pacing and sound finished to a standard that survives being dropped straight onto a homepage.
Compressing a technical proposition into a spoken seventy-five seconds, then cutting it again. Most of the work here was deletion.
A localised read with the pacing the edit needed, re-cut as often as the timing changed.
Cinematic B-roll and persona portraits that would otherwise have meant a location, a crew and a cast.
The product interface, the text choreography and the audio layering. Not generated, decided.
Twenty-plus versions went in the bin
The tools produced usable material within an hour. Getting to something worth publishing took days of rejecting output that was fine. That gap is the entire skill, and no subscription closes it.
The generated footage was pushed back, not forward
The obvious move is to lead with the most spectacular AI shot. It was faded into a soft background instead, so the words and the interface carry the message. Restraint reads as confidence; spectacle reads as a demo.
Typography and pacing were the real design work
Alignment, weight, negative space, the easing on every text cascade, where the voice pauses and where it does not. AI is poor at precise minimal layout, and precise minimal layout is what makes a film look expensive.
The message stayed in charge
Every effect that did not help a founder understand the product was removed, however good it looked on its own.
€10,000+ for an agency asset: location, crew, cast, voice talent, motion design
Under €200 in tokens across the stack
Three to four weeks, gated on scheduling
Days, gated only on judgement
One or two, because each reshoot has a hard cost
Twenty-plus, because a version costs almost nothing to discard
Committed up front, before anyone has seen a cut
Deferred to the edit, where it is cheap to be wrong
This sits on an AI consultancy's site because it is the same method, in a medium you can judge in seventy-five seconds. Scope one outcome tightly. Get a rough working version fast. Discard freely while it is still cheap to be wrong. Ship the version that survives.
That is how the workflows get built too. The tools are different, the discipline is identical, and the value was never in owning the tools.
AI-assisted production. Script, voice and visual assets were generated with the tools named above; structure, interface design, motion, edit and final mix were human decisions, reviewed and signed off before publication. How Artellis uses AI
If that method would suit something you're trying to get moving, the next step is a short call.