Capability 04

Generating is free. Choosing is the job.

Human-directed generative production, built on a brand model you own, with the rights chain intact.

The shift

The cost moved. It didn’t disappear.

The efficiency is real and the largest advertisers in the world are publishing it themselves: production timelines cut from seasons to weeks, costs down by a third to a half or more, concept to testable asset inside a single working day.

But look at how one of those films actually gets made. Thousands of options generated. A handful used. The cost didn’t vanish — it moved out of the shoot and into curation, judgement and quality control. The unit of labour changed.

Which is why we don’t lead on speed. Speed is now table stakes, it’s being commoditised in public, and your procurement team has read the same announcements you have — they will take the saving at the next renewal. What holds its value is the thing all the measurement says generated work is worst at: distinctiveness. That’s a system to build, not an hour to bill.

One more thing worth saying plainly. The most talked-about generated campaign of recent years scored at the very top of the largest creative testing panel in the market — and also drew real public criticism. Both are true at once. Volume is not craft, and audiences can tell the difference between a brand using a new tool and a brand cutting a corner.

Client challenges

SOUND FAMILIAR?
Challenge one

“We make five times more content than last year and it all looks the same.”

Brand codes written as enforceable constraints, plus a feedback loop where every approval and rejection teaches the system. Volume without codes regresses to the average, reliably and quickly.

Brand model Distinctiveness Constraint design
Challenge two

“We got faster and the bill didn’t move.”

Then you bought labour, not a system. We price on what gets delivered, not on how many people were on it — and we’ll show you cost per asset and cycle time before and after, so it isn’t a matter of opinion.

Commercial model Cost per asset Cycle time
Challenge three

“Legal has stopped three campaigns because nobody can prove where the assets came from.”

Model access on commercially indemnified terms, credentials attached to generated assets, a logged chain of how each thing was made, consent for any likeness used, and disclosure classification built into the workflow rather than remembered at the end. That’s the spine, and it’s included.

Provenance Rights Indemnity Disclosure
Challenge four

“Our regional content is translated, not localised, and it shows.”

Most of India’s internet consumes content in a language other than English, and audiences outside the metros now spot the difference between a translation and something written for them. We transcreate with native review. We don’t run copy through a language model and call it localisation.

Vernacular Transcreation Cultural review
Challenge five

“We can’t find the asset we made six weeks ago, so we make it again.”

Asset operations: taxonomy, metadata, versioning, retirement, and machine-readable descriptions so your own content is findable by the systems that recommend you. Deeply unglamorous. Decides whether any of this compounds or just accumulates.

Asset ops Metadata Versioning

What we build

Offer 01 Two to four weeks, per project

Upstream Pre-visualisation

The highest-leverage thing on this page and the least obvious. Explore twenty to fifty creative directions before a production budget is committed, instead of two or three afterwards. It reduces the cost of being wrong, which is the real cost in production.

Offer 02 Six to ten weeks, then ongoing

Brand Model & Codes Engineering

Train and govern the generative system on your brand. Reference sets, constraint definitions, rejection criteria — and critically, the feedback loop, where every approval and annotation becomes signal. That accumulating judgement is the durable asset, not the model. It’s yours, and it’s in the contract.

Offer 03 Per project, or monthly capacity

Production Pods

Human-directed generative production: brand film, photoreal product imagery, campaign systems, social-first variants, presenters. Small senior teams. Sold as an outcome with a date on it, never as a headcount.

Offer 04 Per campaign, or ongoing

Localisation & Vernacular at Scale

Transcreation with native cultural review across the languages your customers actually think in, including for voice.

Offer 05 Build, then ongoing

Asset Operations

The layer that decides whether the rest compounds. Taxonomy, metadata, versioning, retirement, and machine-readable descriptions so your library is retrievable by the systems that matter.

Offer 06 Six to eight weeks

Content System Design

The consulting-grade version. Planning model, production model, governance model, cycle-time targets, cost-per-asset baselining, and a clear line around what stays human.

Where the humans are

Machines generate. Humans choose.

This line was on our site before any of the rest of this existed, and it’s the reason the rest works. Here’s what it means in practice, so it isn’t just a nice sentence.

Strategy and the idea

Never generated. The work that wins is still the work someone thought of.

Creative direction and selection

When thousands of options exist for every one that ships, selection is the craft. It’s the most senior job in the studio, not the most junior.

Brand-code judgement

What is and isn’t us. Documented by people, then enforced on the machine.

Quality and rights control

Every asset that ships has a person’s name against the decision to ship it. That record is auditable, and for some clients it’s the reason we get hired.

The client relationship

Senior people on the actual work. This is structurally easier for a small studio than for a large one, and it’s most of what independent shops are winning on right now.

FAQ

Frequently asked questions.

Whichever suits the job, on commercially indemnified terms — that part isn’t negotiable. The landscape shifts every few months, and the market gets reminded of it regularly when a prominent product is withdrawn. We sell the system and rent the models.

That’s what the brand model is for. Reference sets, constraints, rejection criteria, and a loop that tightens over time. Expect the first weeks to be about calibration.

It can. But the measured risk isn’t that people spot the AI — most don’t. It’s that generated work is less distinctive and therefore less memorable, and that hiding it costs more trust than disclosing it. Advertisers consistently overestimate how comfortable audiences are with this, while those same audiences say clear disclosure would make no difference or a positive one to whether they’d buy. So we disclose, and we spend the effort on distinctiveness.

Yes, and sometimes we’ll recommend it. The pre-visualisation work often makes the case for shooting something properly — it just makes that case before the budget is committed rather than after.

See it
before you pay for it.

Two to four weeks of directions explored, on-code, before a production budget is committed.