The work. The thinking. The human part.115 case files, open for inspection

Field selectionMoncler × adidas Originals / 2023

Models and sculptural explorer figures in oversized puffer looks pose against a blue studio backdrop. — Moncler x adidas Originals; image distributed through adidas News media.

Fashion / From the archive

When the imagined
meets the made.

AI explorers, physical sculptures, human craft. A campaign that gets interesting when the image leaves the screen.

Inside The Art of Explorers

115Campaign & process files 2015—26A growing historical record 16Research chapters 08Working playbooks

01 / The editorial desk

Ideas worth a second look.

All news & opinion ↗

Editorial / 2026-09-15

The prompt budget is not the production budget.

More routes can mean more review. A practical way to price the decisions that turn generated material into an approved campaign.

Read the note ↗

News & opinion / 2026-09-09

Autodesk puts the maker back in frame.

A technology campaign chooses people as its subject. That does not make the tools irrelevant; it gives them a job.

Read the note ↗

News & opinion / 2026-09-05

In-house does not mean automatic.

Very’s autumn campaign makes a case for recognizable brand codes—and for being precise about what a production announcement actually proves.

Read the note ↗

02 / Under the surface

Don’t just see the ad.
See the decisions.

Original production visuals.
Three different jobs for AI.

Match footage of Serena Williams with yellow tracking boxes around the ball, torso and feet. — Nike / AKQA · Original source visual

Nike

A match made from movement.

A tracking overlay reveals the raw material behind the virtual Serena match.

The machine’s job: Machine-learning models analysed official tournament data and archival footage to simulate Serena Williams' 1999 and 2017 playing styles.
The human decision: Nike and AKQA framed the anniversary story, adapted the research technique, produced the broadcast and interpreted the sporting data.

Inspect the full case ↗

A shelter photograph of a black-and-white dog beside four yellow ad layouts showing the dog in different poses. — Pedigree / Colenso BBDO / Nexus Studios · Original source visual

Pedigree

A real dog. Many formats.

A shelter photograph becomes a family of ad layouts. The dog’s identity is the constraint.

The machine’s job: Machine learning adapted shelter photographs into advertising imagery of the same dogs.
The human decision: Pedigree, Colenso BBDO and Nexus Studios developed the campaign and its image-production system; shelters supplied the real subjects.

Inspect the full case ↗

Lipton reference image showing a woman on a motorcycle and the production prompt beside it. — Lipton / Critical Mass / Adobe · Original source visual

Lipton Iced Tea

A prompt is a production choice.

Lighting, framing and colour move from the written brief into a visual reference.

The machine’s job: Adobe reports that the team used Firefly Structure Reference and Style Reference to generate more than 100 storyboards, 2,000 AI-driven images and 500 prompts, including casting and prop references.
The human decision: Critical Mass, Lipton LATAM and the production team set the visual language, selected directions, cast real models and made the final campaign.

Inspect the full case ↗

03 / New to the archive

Different years. Still useful.

Explore recent additions ↗

A model in a colourful patterned two-piece outfit stands before sunlit sand-coloured buildings. — Mango Teen; image published in Mango Fashion Group's pressroom.

2024 / Mango Teen

fashion imagery generated from the real garment

Mango used generative AI to create the campaign for its limited-edition Sunset Dream collection, while grounding the imagery in photographs of the actual garments and retaining art-team selection and retouching.

Open case file ↗

A woman in a call-centre office looks at a monitor showing Daisy, O2's AI granny, talking on the phone beside a cat. — O2 / Virgin Media O2; image published in the Virgin Media O2 newsroom.

2024 / O2 / Virgin Media O2

an AI character used to occupy the scammer, not the customer

O2 introduced Daisy, an AI 'granny' designed to answer suspected scam calls with real-time, deliberately meandering conversation as part of its Swerve the Scammers campaign.

Open case file ↗

Two characters face each other through glass beneath the Morgan title in the official trailer thumbnail. — 20th Century Fox / 20th Century Studios / IBM; official YouTube thumbnail.

2016 / 20th Century Fox / IBM

Watson found the moments; an editor made the trailer

The 2016 Morgan trailer used machine analysis to shortlist footage, then a human filmmaker assembled the promotional edit. It is an early case of AI helping a marketing team choose from existing material.

Open case file ↗

The practical desk / 20 new guides

From good work
to better decisions.

Explore the practical desk ↗

01 / Creative craft

Make the idea survive the output.

What must stay true when the image changes?

02 / Production operations

Plan for the work after generation.

Who owns the next decision—and what will they receive?

03 / Evidence & measurement

Know what the result can actually tell you.

Compared with what, for whom, and under which conditions?

04 / Responsible release

Make accountability part of the delivery.

Can the team explain, correct and withdraw what people receive?

Guide / 3 min read

Product fidelity review: inspect the thing being sold

A review method for stopping attractive AI imagery from changing the product, the subject or the offer.

Open the guide ↗

Two stylised Serena Williams tennis figures face each other beneath the words Game Set Serena. — Campaign reference for this guide. Nike / AKQA.

Guide / 3 min read

Read a campaign performance claim without borrowing its confidence

A reported lift can be useful evidence, but only after you identify the outcome, denominator, comparison and owner of the claim.

Open the guide ↗

A woman in a white top and frayed white shorts stands on a rooftop terrace above a city, with a Digital Twin label at lower right. — Campaign reference for this guide. H&M; image published by H&M Group.

Guide / 3 min read

A digital likeness needs an approval system, not a one-time yes

H&M’s digital-twin release shows why a participant’s involvement, visible labeling and review process matter more than a vague claim that a likeness was ‘cleared.’

Open the guide ↗

04 / The practical desk

Bring a better
question to
the next brief.

The finished work is only the beginning. Get into the process, the budget and the evidence.

Open the research library ↗

The production plan

Eight workflows, from social to brand film.

The real budget

Selection, finishing, review and delivery.

The evidence test

What changed—and what can we prove?

The release review

Consent, provenance and the claims in the work.

Ideas worth a second look.

Don’t just see the ad.See the decisions.

A match made from movement.

A real dog. Many formats.

A prompt is a production choice.

Different years. Still useful.

From good workto better decisions.

Make the idea survive the output.

Plan for the work after generation.

Know what the result can actually tell you.

Make accountability part of the delivery.

Bring a betterquestion tothe next brief.

The production plan

The real budget

The evidence test

The release review

One useful note for the next brief.

Don’t just see the ad.
See the decisions.

From good work
to better decisions.

Bring a better
question to
the next brief.