Explainer video production for software

2 explainers — and neither was made for a client

Both explainers on this page are for products that do not exist. The studio invented the software, drew every screen in them and cut both films — and on the narrated one, wrote the script it is spoken from. Each piece is labelled (Concept) wherever it appears. The consequence is sharper for this format than for the others the studio sells: an explainer is a claim that a real product survived being explained, and a product invented in-house cannot argue back. There was no support team to say “that is not how customers actually use step three”, because there are no customers and there is no support team. That is a real difference, not a technicality.

What concept work does prove is the part a brief cannot describe — structure, pacing, how legible an interface stays at a phone size, whether the film has the discipline to stop. The full portfolio is filterable by industry, deliverable and format, and the one engagement a client did pay for is film work rather than an explainer: a brand promo and a market-update format for a mortgage lender.

The cut is locked to the voice, not the voice to the cut

In most explainer pipelines the audio is laid against a finished animation and somebody nudges keyframes until the two roughly agree. It works until the script changes. Move a sentence, cut a clause, re-record a line, and every downstream cut is a few frames wrong in a way that reads as cheap without a viewer being able to say why.

The film above is assembled the other way round. The narration is rendered first and timestamped, and every narrated scene’s boundary is computed from those timestamps rather than typed in — so the step-two chapter starts on the sentence that begins “Two”, and there is no state in which the cut and the voice disagree. The one hand-set number in the whole timeline is where the closing card lets go, which was shortened on purpose to take dead air off the end. Re-cut the script and the film re-times itself on the next render. The craft note on that piece goes through the frames, including what it gets wrong.

Be precise about what that does and does not buy, because the neighbouring claim is one it is easy to state a notch too strongly. The build holds the CUT to the voice; the caption file shipped beside the film is cued by narration segment, not by word. The same machinery drives the narrated work on the data-driven films, where the voice has to land on a figure at the moment the figure resolves — a world-electricity film (Concept) is the hardest version of that problem in the library.

The interfaces are drawn — and for your product they may not need to be

Every screen in the film above is drawn for the frame: a shared board with its columns labelled, a task card joined by a lit line to the one person marked free, a sync node fanning out to the plates it just updated. None of it is a screen recording, and the reason is honesty rather than preference — the product is fictional, so there is nothing to record, and a shot of a real screen would be the one element in the piece that could not be stood behind.

For a product that exists, a capture of the actual software is often the better answer, and a studio that will not say so is selling its own preference. What drawing buys, when it is the right call, is legibility: type set for the frame instead of for a browser at whatever zoom the recording happened at, one element lit while the rest stays quiet, and a flow shown in the order the script needs rather than the order the app enforces. The usual answer for a real product is both — the interface recorded, and the moment that matters rebuilt so it can be read at a phone size.

Drawn does not mean simplified into meaninglessness. A six-city multi-market template (Concept) and a narrated data film that prints its own source codes (Concept) are drawn the same way, at real density, from real datasets.

The figure a viewer repeats is the one that has to be labelled

When an explainer has a number in it, that frame is the most dangerous one in the film. A big figure counting up under a confident line is the thing a viewer remembers, screenshots and repeats — and if it was invented for the edit, the film has put a claim into the world that nobody in the company ever signed off. Laneflow (Concept) handles that in the frame: the payoff chapter sets the figure counting and prints a chip directly beneath it naming what the figure is a sample of.

Now the part a portfolio page normally leaves out. That chip is small type under a figure many times its height, so it is a footnote — what makes it defensible is that it is a footnote UNDER the number rather than an end card nobody screenshots. And the other film on this page does it less well: Cadence (Concept) labels that figure too, and more thoroughly than the film above does: the line directly beneath its headline number reads “illustrative figure · concept commission”, naming the provenance as well as the status. Where it falls short is one plate. The square cut is a real square build — its chrome re-lays and it prints its own 1080×1080 — but the 24-week ridgeline is laid out at the wide film’s coordinates and clips at both edges, so its title reads “RATIONS LANDSCAPE”, its legend loses the word it starts with, and the annotation marking that change illustrative is pushed past the right edge. That plate is ours to re-render.

Nobody speaks the narration on the film above

One of the two explainers here carries narration, and that voice is a text-to-speech render — an open-weights speech model, run locally, from the same script printed verbatim on the piece’s own page. No voice is cloned from a real person and no performance is imitated. It is said here, in a heading, rather than in a footnote, because on this format the voice is not a garnish: an explainer is a voice with pictures under it, and a buyer choosing this studio is choosing that voice more directly than on any other page of this site.

The picture is a different matter and the distinction is the whole point. No generative model draws a frame of this work — not the interfaces, not the type, not the transitions; every frame is a component rendered from code, which is why the same build returns the same film. The synthetic voice is a production choice on concept pieces made with no budget and nobody to book. If a project wants a human read, the build composites the voice track it is handed and nothing else about the film changes — and a script written to be spoken by a person is a better script either way. One person writes the script, draws the screens and renders the film, which is why there is somebody to ask.

How an explainer project runs

It starts with the script, because on this format the script is the product and everything visual is a consequence of it. You send what the software does, who it is for and the one thing they must understand; what comes back is a written walkthrough with the steps numbered, and that is the thing to argue about — arguing about it later, over rendered frames, is what makes explainers expensive. Once the steps are agreed the interfaces are drawn against them, the voice is laid, and the film assembles itself around the timings. Tell us what your product does — the message reaches the founder, not a queue.

Afterwards you own the source, not just the export, and on this format that matters more than usual: a product changes, and a film you cannot open is a film you have to buy again. The field notes are where the engineering behind it gets written down, and the studio’s own identity system is the reference for how the token discipline works when a film, a mark and a document all have to agree.

How long should explainer videos be?

There is no correct number, and a studio that hands you one is quoting a habit rather than a measurement. Runtime is an output here, not an input: it falls out of how many ideas the product genuinely needs, one to a beat, and it can only be settled honestly once that count is known. Choosing the length first is what manufactures the padding above, because the film is committed to filling a slot before anybody has decided what belongs in it.

The explainer at the top of this page, Laneflow (Concept), came to 23 seconds. That figure is measured off the encoded file this page serves rather than taken from a brief, and the wide and the vertical are separate builds rather than crops of one another, yet both land on it. Read it as evidence of a method, not as a length to aim at: it is what one product's ideas amounted to once nothing was added to round the figure up.