All articles

Creative

How we use generative AI in ad production, and where we don’t

AI has changed how fast our studio works. It hasn’t changed who makes the decisions.

Creative studio29 July 20264 min read

We get asked about AI in almost every first meeting now. Sometimes the question is excited (“Can you make us 200 ads by Friday?”) and sometimes it’s nervous (“Is any of this going to look fake?”). Both are fair. So here’s a plain account of how our studio actually uses it.

Where it genuinely helps

Getting to more ideas, sooner

The early stage of any campaign is about range. Which angles could work? Which visual directions? What would the opening line of the video be? AI is very good at producing a wide spread of rough options quickly. Most of them are average, and that’s fine. A designer can scan forty rough directions in the time it used to take to sketch four, and the good ones jump out.

Scenes, backgrounds and settings

A lot of ad photography used to be about putting a product somewhere: on a bathroom shelf, on a beach, in a kitchen at golden hour. Image models can now generate those settings convincingly, and our designers composite the real product into them. It saves shoots that would have cost more than the campaign.

Adapting one idea to every placement

One concept might need to run as a vertical video, a square image, a carousel and six display banners. Resizing, recomposing and extending images to fit each format used to eat days of production time. Much of that is now assisted, which frees designers for the work that needs judgement.

First drafts of copy

Headline variations, alternative hooks, translations to work from. We treat all of it as raw material for a copywriter, never as finished copy. The first draft is rarely the one that ships, but it gives the writer something to push against.

Where a person always decides

There are parts of the process we keep firmly human, and not out of nostalgia.

Whether it’s on brand. Models produce images that are technically impressive and subtly wrong for the brand. The wrong kind of light, a mood that’s slightly too glossy, a setting the customer would never be in. Spotting that takes someone who knows the brand.

Anything that makes a claim. Health, finance, beauty results, prices, guarantees. If an ad says something that could be held against the client, a person writes it and a person checks it.

The product itself. AI is happy to invent details. Extra buttons, wrong labels, a logo that’s almost right. We use real product photography for the product and let AI handle the world around it.

Real people. We don’t generate likenesses of real people, and we’re careful with generated faces in general. Hands and teeth are still where a lot of AI images give themselves away.

What to test. Choosing which ideas go into market, and why, is a strategic decision. It depends on what we’ve learned from previous rounds, what the client’s sales team is hearing and what the competitors are doing. No tool makes that call for us.

What it changed for clients

Mostly, speed and volume. We can put more distinct concepts into a test, and refresh tired creative before performance drops rather than after. On platforms like TikTok, where ads wear out quickly, that matters a great deal.

It also made localisation cheaper. Adapting a campaign for five markets used to mean five rounds of production. Now the visual adaptation is quicker, and the time goes into getting the language and cultural details right. We wrote about that in our piece on localising ad creative.

What it didn’t change

A weak brief still produces weak ads, just more of them. AI doesn’t know who your customer is, what they’re worried about or why they chose a competitor last time. That still comes from research, from data and from talking to the people who sell the product.

It also didn’t remove the need for taste. If anything, it raised the bar. When anyone can generate a decent image in seconds, the difference comes from knowing which image to use, and which ones to throw away.

How the pieces fit

Our process runs in five stages: insight, concept, generate, refine, test. AI does a lot of the work in the middle two. People own the first, the fourth and the decisions in the fifth. Every asset is reviewed by a designer before it goes anywhere near an ad account.

That’s the short version of how our creative studio works. It’s faster than it was. It isn’t automatic, and we don’t think it should be.

Work with us

Curious what this looks like for your brand?

Send us a recent brief. We’ll show you how our studio would approach it, from first concepts to test-ready variants.

Keep reading

More from the blog.

All articles
Creative · 21 Sep 2026

Localising ad creative across five languages without losing what made it work

A direct translation keeps the words and loses the point. Here’s how we adapt ads for each market instead.

4 min readRead
Creative · 07 Sep 2026

How to build a creative testing framework that actually teaches you something

Running lots of ads isn’t the same as testing. A good framework tells you why something worked, not just that it did.

4 min readRead
Creative · 22 Aug 2026

A brand-safe workflow for AI-generated ad images

Generating an image takes seconds. Making sure it can run without embarrassing anyone takes a process.

4 min readRead