Deliverable · Ad creative
Ad creative arrives as a set, not as a single film
The short answer
Video ad creative at Blue Snow Developers is built as a variant system, not one film at a time: one composition, a locked brand layer, and only the data changing per market, rendered from code rather than generated. 2 pieces are published here; neither was made for a client.
Almost nobody commissions a single advertisement. They commission a market, a season, a test — a set of cuts that all have to be correct — and the interesting question is never about the first cut. It is whether the hundredth is built the way the first was, or assembled by somebody tired on a Friday. Everything here is motion graphics — nothing on this page was filmed — so the hundredth cut is a re-render, not a reshoot.
The first cut is easy. The hundredth is the actual job
Anybody can make one good advertisement. The work only becomes difficult when the same idea has to exist forty times — one per city, per product line, per language, per offer — and every one of them has to carry the mark correctly, the small print correctly and the figure correctly. That is where the ordinary process quietly falls over, because the ordinary process is a person opening a file and changing things by hand — and on the fortieth pass that person is not making the same decisions they made on the first.
Count how many of a thing you are buying and the format usually answers itself. A brand film is bought once and lived with for years; a logo animation is bought once and used every single day; an explainer is bought once and watched deliberately; a promo is bought for a campaign and a place. Ad creative is the one nobody buys in the singular — which changes what "finished" means. Not a file: a build that can be run again on new inputs and land in the same place. Everything the studio makes, scoped and quoted the same way puts them next to each other.
Nobody has commissioned any of this, and that has to be said first
Both films below are self-initiated studio work. They exist because the format had to be built end to end before it could honestly be offered, not because a brand paid for them, and they are labelled (Concept) wherever they appear. Neither has ever run as a paid advertisement. A grid of finished-looking frames does the implying on its own, and nobody has to say anything untrue for a reader to walk away believing brands have bought this.
What concept work can and cannot prove is worth being precise about, because the honest answer is not "nothing". A finished frame is still evidence — of drawing, of typesetting, of whether a number on screen arrives with the thing that makes it checkable. What it is not evidence of is demand. Nobody has paid for this format yet, and there is no click-through rate to quote. The library marks real client work apart from concept work on every tile, and the one real client on this site is a mortgage lender's promo rather than anything on this page.
Where the vertical work on this site actually lives
Not on this page: every ad-creative piece published here is a landscape cut and nothing else, and a great deal of paid social is not. That is a gap in this pair rather than a limit of the studio, and it matters because the cheap fix is the wrong one. Squeeze a wide frame into a tall one and the composition does not travel with it — the margins go first, and the margins are the load-bearing part of an ad rather than the decorative one.
The vertical craft is on this site, it is just filed under other formats. A vertical data drop is built as a stack for the feed — one sourced number, typeset to be read at arm's length on a phone — and a local-business film built tall from the first frame runs its chapters one to a screen with a progress row underneath. Neither of those started life wide. A vertical variant of either film on this page would be a new build on the same data, which is the cheap half of the work — and it is cheap precisely because the expensive half, deciding what the composition is, has already been done.
What a composition is, as opposed to six files laid out six times
Most claims about consistency have to be taken on trust, because you are shown one output at a time. This film refuses that: it puts six markets in front of you simultaneously, so the thing being claimed is visible rather than described. Run your eye down the column of brand marks and they do not move a pixel between New York and São Paulo. Run it across the maps and every one is a different place, drawn at that city's real coordinates. Fixed where the brand lives, variable where the data lives — and a set built that way is a system, whereas a set that merely looks that way is a folder somebody was careful with.
What the film shows next is not a second dataset — it is each figure counting up from zero to the number that dataset gives it, and a mark landing on every card to say the value came from the data rather than from a designer. That is a provenance gesture, not a re-render, and it is worth being exact about because the re-render is the thing being sold. The evidence for THAT is in the six plates side by side: measure the brand mark against the same mark on the next card and it does not move a pixel, on any of them. A layout that survives that test is a build you can run again next quarter against a newer source and land in the same place. It is the same machinery behind the data-driven work, where the dataset is the brief, and it is why that specialty and this one keep meeting: an ad that fans out across markets is a data problem wearing a marketing brief.
Now the number this page will not give you, because it does not have it to give. The largest set published on this site is the six you can watch. The film says a larger figure out loud on its closing card, and that is the film talking — a claim about what the build is designed to do, not a delivery this studio can point at and invite you to count. What is actually checkable from where you are sitting is the mechanism: the layers are separated, the layout is fixed, the data is an input. Whether that has been run a hundred times for somebody is a different question, and the honest answer to it is the one at the top of this page.
One limitation, while the film is in front of you. The source credit sits beneath the full six-card composition rather than inside each card. Crop a single plate out to post on its own and the attribution does not travel with it — which, on a piece whose whole argument is fanning out to many markets, is a real fault rather than a detail. The honest fix is per-plate attribution in the composition, not a note on a web page. Another film in this library does it correctly — its licence credits are burned into the plate, and this round checked that per frame rather than by sampling it: all 1,508 frames carry them.
Lookbook (Concept)
Look-library range — one frame rendered through 24 of 516 catalogued looks. Delivered 16:9, running 33.8 seconds — the question that comes BEFORE a variant set exists. Every card in it is the same photograph, so the only variable on screen is the treatment rather than the subject. The full craft note, including what its own records do not cover.
Deciding how an ad should look, before anybody spends on it
The most expensive question in a campaign is asked at the beginning and answered badly: what should this feel like. The usual answer is a mood board — other people's finished work, arranged on a slide, which everybody in the room reads slightly differently and nobody can execute from. The film above this section replaces it with something orderable. Under each card is the recipe that produced it — the treatments in the order they were applied, each with a number on it, and a line naming the categories that look was built for. The result is that feedback stops being taste and starts being an instruction: you can name a card, name the treatment on it and name the direction to move it in, and what comes back is the thing you asked for rather than somebody's reading of it.
The interstitial is what makes the rest of it mean anything. Midway through, the film stops and shows the plate with nothing applied at all, captioned as the source. Without that, a look reel is a run of attractive frames and a viewer cannot tell a grade from a different photograph — which is the quiet trick in most of them. With it, the base is on screen and every card afterwards is being measured against something you have actually been shown. It is the same instinct as the identity film that prints its own colour values on top of the swatch: publish the number, and the work becomes checkable instead of merely impressive.
Name a look for what it does, not for whose film it came from
Colour suites have always named grades after pictures people admire. It is completely ordinary shop-floor language and everyone in the trade reads it instantly. It is also, on a public catalogue that a stranger reads as a product list, a different thing: those are other companies' names, attached to work that is not theirs, in a document that looks like an order form. This studio's own look film carries some of that inherited vocabulary on its cards, and rather than quietly leaning on it, the position is written down here — the naming is being revised, and no such name appears anywhere in this studio's own copy.
The practical case is stronger than the cautious one, which is why this is craft advice and not a disclaimer. A look described by what it does — warm highlights, heavy grain, slightly crushed blacks, letterboxed — can be built, checked against what comes back, and handed to a different person next year with the same result. A look named after somebody else's picture cannot: it means a different thing to every person in the room, and it carries an argument about somebody else's property into a conversation that should be about yours. Brief the property you own.
No frame is generated, and one voice on this page is
The platforms that answer this question loudest answer it with generation: a model writes the voice, names your city in the opening seconds, dubs it into six languages and lip-syncs the result. The picture here is made the other way round. Every map, every rule, every plate and every letterform is a component with a value passed into it, so the film has no opinions of its own — build it twice and you get the same frames, which is a duller sentence than "generated" and the only one that lets a hundredth variant be trusted sight unseen.
Now the machine-made part, in the same breath rather than a paragraph later. A denial that reaches further than the truth costs exactly what an exaggeration costs, and it is harder to notice, because it arrives dressed as scruple. 2 of the films on this page are narrated, and those voices are text-to-speech renders — an open-weights speech model, run locally, reading the script that is published verbatim on that piece's own page. No voice is cloned from a real person and no performance is imitated. The other film has no narration at all — a music bed and nothing over it. Using a model at all is a decision about self-initiated work with no budget and nobody to book, not a house style: hand over a recorded read and it goes into the build as an input like any other, with the picture untouched. The explainer page says the same thing at more length, because on that format the voice is the product.
What ad creative cannot do, and what to buy instead
Ad creative cannot introduce you. It is seen briefly, repeatedly and out of context, by somebody who did not ask for it, so it holds one idea and one response — and if nobody knows who you are yet, every variant is spending its one idea on the wrong thing. That is a brand film's job, bought on register rather than on response. It cannot teach either: if the product has to be understood before anybody will act, the format is an explainer, built around a script, and the ads come afterwards assuming it.
One hard limit, worth knowing before a call rather than during one: nothing here was filmed, and this studio does not own a camera. If the ad needs your product in a real pair of hands, your storefront, or anybody's face, that footage has to come from somebody who shoots — and everything this page is about can then be built around it. The titles, the figures, the localized variants and the brand layer do not care whether the middle of the frame is drawn or photographed.
And the failure specific to this format: a variant set is only as good as the source it fans out from. Rendering a hundred plates off a spreadsheet nobody has checked does not produce one mistake, it produces a hundred confident ones, each of them typeset beautifully and each of them wrong in public. The fan-out is the cheap part. Getting the source right is the part worth arguing about, and it is why a regulated buyer's version of this work starts with where the figures come from rather than with what it should look like.
How an ad-creative project runs
It starts with a list rather than a brief: what has to be identical in every cut, and what changes between them. That one page decides the composition, the runtime, where the type can live and how far the set can be extended later without re-opening anything. The common failure is putting something in the fixed layer that turns out to be variable, and discovering it on cut sixty. Send the list of what changes and what must not — nobody triages it first.
You keep the project, and on this format that is not the usual boilerplate about ownership: a variant set exists in order to be re-run. New markets, new figures, a line that has to change — whoever holds the only openable copy of the build holds the pricing on every one of those. So the handover is the build itself rather than the exports from it — the project, the data bindings and the render script, which is the difference between owning a hundred files and owning the thing that made them. Running it again needs the same toolchain this studio uses, and that is worth knowing before you assume it is a button. The same person writes, draws and renders every piece here, so a question about the fixed layer reaches whoever decided it. The field notes go through how a build is put together. And if you want to see a locked layer behaving under pressure before you trust one with your own mark, this studio runs its own brand off exactly that arrangement — the mark, the type scale and the colour on this site are read from one file, so the argument on this page is one we are already living with rather than one we are only selling.
What to ask before you commission ad creative
Has any of this ad creative been made for a real client?
No. Both pieces on this page are self-initiated studio work, made to build the format end to end rather than for a real brand, and both are labelled “(Concept)” everywhere they appear. Neither of them has ever run as a paid advertisement. That is worth saying in the first answer rather than leaving a portfolio grid to imply otherwise: this studio has one real client and their work is a mortgage promo, not ad creative. What these two pieces are good for is judging the craft and the method — whether the frames are built properly, whether the claims on them are checkable — and they are not evidence that anybody has asked for this particular format yet.
What is video ad creative, and how is it different from a promo or a brand film?
Video ad creative is the set of short films made to be placed as paid advertising — usually many of them at once, cut per market, per audience or per offer, and judged on what people do after seeing one. The difference from the neighbouring formats is what each is bought for rather than how each is made. A brand film is bought once and lived with for years; it answers to how a company feels. An explainer is watched deliberately by somebody who already wants to understand the product. A promo is made for one campaign and one place. Ad creative is the only one of those that is plural by default, which is why the interesting engineering question is not what the first cut looks like but whether the fortieth was built the same way.
How do you make a hundred versions of one ad without them drifting apart?
By making one composition and changing only the data underneath it. The film at the top of this page holds six city plates on screen at once, and across all six the brand mark, the dividing rule, the label positions and the type sit at identical coordinates — what differs is the place: its name, its map and its figure. The film names its own layers in a panel: a basemap, a dataset, and a brand layer it calls locked, under the line “two layers change per city · one never does”. Because the render is deterministic, the hundredth plate is built exactly the way the first one was, rather than being assembled again by a person at the end of a long afternoon. Drift is not a discipline problem in this method; it is structurally unavailable, because there is only ever one layout.
Can you shoot my product, my premises or my team for the ad?
No — this studio does not shoot, and it is better to know that before you spend a call on it. Everything in these films is drawn and rendered from code: type, rules, vector shapes, maps built from published data, and photographic plates that are licensed rather than captured. If the ad needs your real product in somebody’s hands, your storefront or your staff on camera, the honest answer is that a person with a camera should film that. What can still be built is everything around it — the titles, the data, the localised variants, the end frame, the brand layer that has to be identical in every cut — composited around footage once that footage exists.
Both of these are widescreen. What about vertical feeds?
Correct, and it is a real gap in this particular pair rather than a limitation of the studio: both ad-creative pieces publish a landscape cut and nothing else. The vertical craft on this site is genuinely there, it is just filed elsewhere — there is a vertical data drop built as a stack for the feed, and a vertical promo for a local business built the same way, both composed vertical rather than cropped down from a wide layout. That distinction matters more here than anywhere: cropping a landscape ad into a feed keeps the middle and throws away the edges, and on this kind of work the edges are where the source line, the qualifier and the call to action were sitting. A vertical variant of either piece would be a new build, not an export.
Is any of this generated by an AI model?
No frame is. Nothing in the picture is synthesised — not the maps, not the plates, not a letterform — and every frame is a component rendered from code, so building it twice returns the same film. Now the machine-made part, stated in the same breath rather than a paragraph later: the first film is narrated by a text-to-speech model, open-weights and run locally, reading the script published word for word on that piece’s own page. Nobody’s voice was copied to make it and no performance was imitated. The second film is not narrated at all; it carries a music bed. Hand this studio a recorded read and the build takes it as an input like any other, and the picture is untouched.
How should we brief the look of an ad?
In words that describe what the look does, not in the name of somebody else’s picture. “Warm, soft highlights, heavy grain, slightly crushed blacks” can be built, checked against what comes back, and reproduced by somebody else a year later. “Make it feel like that film everyone liked” cannot — it means something different to every person in the room, and it names a property that is not ours or yours. The look film on this page is built to make that briefing concrete rather than atmospheric: every card prints the treatments applied to it with a strength value beside each one, so a note like “that one, but push the grain” describes an edit that can be executed exactly. Partway through, the film drops back to the untouched plate captioned “SAME PIXELS.”, which is the only way to tell a grade from a different photograph.
When is ad creative the wrong thing to buy?
It is the wrong buy when the problem is that nobody understands the product yet, or that nobody knows who you are. Ad creative is built to be seen briefly, repeatedly and out of context, so it can carry one idea and one response — and if the idea needs explaining first, the money is better spent on something that explains, then on ads that assume it. It is also the wrong buy when the variants are the point but the underlying data is not ready: a set of localised cuts is only as good as the source it fans out from, and rendering a hundred correct-looking plates off a spreadsheet nobody has checked produces a hundred confident errors instead of one. Fix the source first; the fan-out is the cheap part.
See one of your variants before you commit to the set
Send the thing that changes — a list of markets, a product range, a spreadsheet of figures. When there is room, a watermarked cut gets built from it, so what you are judging is your own data in the composition rather than somebody else's showreel. No paperwork, and nobody has to sit on a call to see it.
Written by Baljeet Aulakh, who builds every piece in this portfolio. Last updated . No AI-generated frames.