Client work · MENA retail
HomesRus had no shortage of ideas. They had a reformatting problem, and it was eating the quarter.
Client: HomesRus, MENA retailFormats: 9:16, 6:5Published 11 September 2026
This is the part most AI creative pitches get backwards. The hard part of a retail catalogue campaign is not producing an ad. A good team produces the ad in a day, and the client signs it off.
The hard part is the next four weeks: the same approved ad, rebuilt for every product in the catalogue, then rebuilt again for every placement, then checked because somewhere in asset 240 the logo drifted and the headline lost a line. That work is mechanical, it is expensive, and it is where the studio time actually goes.
HomesRus had exactly this shape of problem. A furniture catalogue, a paid social plan in 9:16, a programmatic plan in 6:5, and a production pipeline that priced every new variant like a new design.
Logo lockups, palette by exact value, type hierarchy, clear space and safe zones went in before anything was generated. Everything produced afterwards inherits them, which is what makes 530 assets approvable instead of 530 review items.
Not a brief, not a prompt, the actual signed-off composition. The whole campaign is derived from it, so the expensive part of the review happens once.
Background, product, headline, subhead, price, CTA, logo and legal became addressable layers with rules about how each behaves when the product or the frame changes. The product layer is the one that swaps per SKU. Nothing else moves unless the frame forces it to.
Every SKU against every placement, recomposed for each frame rather than cropped into it. The system flagged the handful of cells where a rule could not be met, a headline too long for a narrow frame at minimum type size, and a human resolved those. Nobody touched the rest.
The approved master
One sign-off, then the product layer moves.
The sale panel, the date range, the payment badges, the delivery promise and the brand furniture are all fixed by the kit. Per SKU, only the product photography and its staging change, which is why the set below reads as one campaign rather than eight.
More than 530 finished ads in under four weeks, across two placement families, with the product swapped per SKU and the brand identical on every one of them. Production time per placement set moved from weeks to minutes, and the review load stayed where it belongs, on the master.
The number that matters to a marketing director is not 530. It is zero: the number of assets that had to go back through brand approval after the first one was signed off.
Any advertiser with a product range and more than one placement has the same maths. Products times placements times sizes times languages is a number that grows faster than a studio can staff for, and it is the reason catalogue campaigns ship late and thin.
The fix is not a faster designer or a better prompt. It is making the approved master the only thing anyone has to approve, and making everything downstream derive from it. That is what creative automation means when it is done properly, and it is the same mechanism that produces mirrored Arabic versions for GCC markets in the same pass.
Lock the brand kit, take the approved master as the source, split it into addressable layers, then fan it across the matrix of products, placements and sizes. Each output is recomposed for its own frame rather than cropped from a single artboard, and the system flags only the cells where a layout rule cannot be satisfied. For HomesRus that produced more than 530 finished ads in under four weeks from one approved concept.
Reformatting an approved concept across a product range and two placement families is typically weeks of studio time, priced per variant, because each new size is treated as a new design. The HomesRus set ran end to end in under four weeks including review, with production time per placement set measured in minutes rather than days.
It does when the brand kit is applied as a constraint at composition time rather than as a prompt. Logo lockups are placed and never generated, palette values are pinned exactly, and type hierarchy and clear space are enforced per frame. That is what deterministic brand memory is for, and it is why no asset in this set needed a second approval.
9:16 for paid social placements and 6:5 for programmatic display. Both were recomposed from the same approved master, so the composition holds in each frame rather than being cropped from a square original.
Yes. Right-to-left output is part of the same derivation: the composition mirrors, the Arabic typeface comes from the brand kit, the copy is retypeset at the correct line height and safe zones are recalculated per placement. The detail is in RTL ad localisation.
We will send back the placement matrix, brand-locked, in English and Arabic.