The landmark AI advertising campaigns are Coca-Cola’s Create Real Magic, Nutella Unica, Sephora’s Virtual Artist, Virgin Voyages’ Jen AI, Nike’s Never Done Evolving, Cadbury’s Not Just a Cadbury Ad, and Heinz A.I. Ketchup. Each one won on a single AI mechanic rather than a bigger budget: personalization at scale, the customer inside the creative, or a synthetic spokesperson. The brand’s own product stayed the constant.
Every roundup of these campaigns lists the brands and stops there. Coca-Cola, Nike, Heinz, a Cannes trophy, done. That framing hides the useful part, because the brand names are the part you cannot copy and the mechanic is the part you can.
So this piece groups seven landmark campaigns by the AI mechanic each one ran, verified against the sources, and then shows which of those mechanics scales down to a brand with one product photo and no celebrity on retainer. If you run creative for a marketing team, an ecommerce brand, or an agency, the second half is the part written for you.
Key Takeaways
- The budget was never the mechanic. Every campaign below leaned on one cheap, repeatable AI move. The expensive parts were the idea and the rights, not the compute.
- Personalization at scale turns one asset into millions. Nutella Unica’s algorithm generated 7 million unique jar designs, and all 7 million sold out in Italy in a month (dezeen.com, June 2017).
- The customer can become the creative. Sephora’s Virtual Artist let shoppers try makeup on their own face, and Cadbury recreated Shah Rukh Khan’s face and voice so local shops could name themselves inside his ad (lbbonline.com, October 2021).
- A synthetic spokesperson scales one face. Virgin Voyages licensed Jennifer Lopez’s AI likeness to generate personalized invite videos, with a deliberate visible tell every time it spoke (adweek.com, July 2023).
- AI can be the idea, not just the tool. Nike simulated a match between 1999 Serena and 2017 Serena and won a Cannes Grand Prix; Heinz let DALL-E 2 prove that “ketchup” looks like Heinz (adweek.com, June 2023; musebyclios.com, July 2022).
- The copyable pattern is the constant. In all seven, the product stayed fixed and AI varied everything around it. That is the exact move a brand runs from one product photo.
What the best AI advertising campaigns have in common
Strip out the celebrity deals and the agency budgets and every campaign on this list runs one of four mechanics. Personalization at scale, the customer inside the creative, a synthetic spokesperson, or AI as the idea itself. None of them is about a model being clever. Each is about pointing generative tools at a job that used to need a photoshoot, a print run, or a thousand separate edits.
| Campaign | Year | What the AI did | The mechanic |
|---|---|---|---|
| Nutella Unica | 2017 | Generated 7 million unique jar labels | Personalization at scale |
| Coca-Cola Create Real Magic | 2023 | Let the public remix brand assets with GPT-4 and DALL-E | Personalization at scale |
| Sephora Virtual Artist | 2016 | AR try-on of thousands of makeup shades | Customer in the creative |
| Cadbury Not Just a Cadbury Ad | 2021 | Recreated a star’s face and voice per local shop | Customer in the creative |
| Virgin Voyages Jen AI | 2023 | Generated personalized invite videos from a licensed likeness | Synthetic spokesperson |
| Nike Never Done Evolving | 2022 | Simulated a match between two eras of one athlete | AI as the idea |
| Heinz A.I. Ketchup | 2022 | Turned image prompts for “ketchup” into brand proof | AI as the idea |
Read the mechanic column, not the brand column. That is where the reusable idea lives.
Personalization at scale: one asset, millions of variants
The first mechanic takes a single brand asset and produces a version of it for every customer, package, or placement. The old way needed a print run or a design team. AI drops the marginal cost of the next variant to near zero.
Nutella Unica is the cleanest example, and it predates the generative-AI wave. In 2017 Ferrero worked with Ogilvy Italia on an algorithm fed a library of brand-approved patterns and colors. The software combined them into 7 million unique jar labels, each stamped with its own code, sold across Italy that February. All 7 million sold out in a month (dezeen.com, June 2017; futurism.com, June 2017). One design system, seven million outputs, zero repeated jars.
Coca-Cola’s Create Real Magic ran the same mechanic in the other direction, letting the public do the generating. In March 2023 Coca-Cola, OpenAI, and Bain built a platform combining GPT-4 and DALL-E, and invited digital creators across 17 countries to make artwork from Coca-Cola’s archive assets, the contour bottle, the script logo, the polar bears. Winning work ran on the billboards in Times Square and Piccadilly Circus, and 30 creators were flown to Atlanta for a workshop (thedrum.com, March 2023; coca-colacompany.com, March 2023). The brand asset was the seed, and thousands of variations grew from it.
The lesson for a smaller brand is direct. You do not need seven million jars. You need the tenth, twentieth, and fortieth version of a product shot, resized for each placement, without booking the shoot again.
The customer becomes the creative
The second mechanic puts the buyer, or something the buyer owns, inside the ad. It is the one that reliably lifts conversion, because the person seeing the creative is looking at themselves.
Sephora’s Virtual Artist launched in 2016, built for Sephora by ModiFace. It maps a shopper’s face through the phone camera and renders makeup on it in real time, thousands of eyeshadow, lip, and lash shades tried on without a cotton pad. Sephora reported that customers had virtually tried on hundreds of millions of shade combinations (prnewswire.com, March 2017). The product stayed exact. The model was the customer.
Cadbury’s Not Just a Cadbury Ad put the customer’s business into the ad instead. For Diwali 2021, Mondelez, Ogilvy, and Wavemaker used Rephrase.ai to recreate Shah Rukh Khan’s face and voice so he could name individual local shops. Small retailers generated their own free version, with Khan saying their store name, lips synced to each one. The agency Wavemaker reported more than 2,500 participating businesses, and the campaign won a Titanium Lion at Cannes in 2022 (wavemakerglobal.com; mediabrief.com, June 2022). One shoot, thousands of personalized cuts.
Try-on and personalization are not a beauty-only trick. Putting a garment on a model, or a product into a customer’s context, is the same move applied to your catalog. Our guide to on-model shots from one product photo covers the ecommerce version of it.
The synthetic spokesperson
The third mechanic licenses one face and delivers it a million times over, personalized. It is also the mechanic with the sharpest ethics, and the best example handled them well.
Virgin Voyages’ Jen AI launched in 2023 with VMLY&R. Jennifer Lopez licensed her likeness for a tool that combined generative voice and generative video, so users could enter details about a celebration and receive a personalized invite video from “Jen AI” (adweek.com, July 2023; cnbc.com, July 2023). Two details matter for anyone copying it. The likeness was licensed, not scraped. And the team added a deliberate, visible effect every time the AI made Lopez say something she never recorded, so no viewer could mistake it for real footage.
That disclosure choice is the part to steal, not the celebrity. A synthetic presenter only works when the audience is told it is synthetic. The alternative is the backlash that follows the campaigns that hid it.
When AI is the idea, not just the tool
The fourth mechanic is the rarest. Here the AI method is the story itself, and it only lands when the product stays the anchor.
Nike’s Never Done Evolving marked the brand’s 50th anniversary in August 2022 with AKQA. Building on Stanford’s vid2player research, the team modeled two eras of Serena Williams from archive footage, her 1999 US Open self and her 2017 Australian Open self, and simulated a match between them across thousands of games. It launched two days after Williams announced her retirement, and it won the Grand Prix for Digital Craft at Cannes in 2023 (akqa.com, 2022; adweek.com, June 2023). The AI was not decoration. The simulation was the ad.
Heinz A.I. Ketchup did it for a fraction of the effort. In 2022 the agency Rethink fed DALL-E 2 prompts like “ketchup in space” and “ketchup impressionism” and found the outputs kept looking like a Heinz bottle. Heinz turned that into the campaign, then crowdsourced more prompts from fans (musebyclios.com, July 2022; rethinkideas.com, 2022). The idea only works because the product is that recognizable. AI was the mechanism; the Heinz bottle was the message.
Both campaigns prove the same rule the other five do. The product is the constant, and AI varies the world around it. When a campaign points AI at the product itself and tries to replace it, the result is the generic AI look that audiences reject.
What none of these campaigns needed
It is easy to read this list as an argument for a big budget. Coca-Cola had OpenAI on the build. Virgin Voyages had a global pop star. Nike had AKQA and a Stanford paper. That is the part you cannot reproduce, and it is not the part that made the mechanic work.
Every one of these campaigns kept the brand’s product fixed and used AI to multiply the creative around it. Nutella’s jar, Coca-Cola’s bottle, the shopper’s own face at Sephora, the local store name for Cadbury, Lopez’s licensed likeness, Serena, the Heinz bottle. The compute was cheap. The rights and the concept were the line items with the commas in them.
Which is why the honest benchmark for most brands is not the hero film. It is the repetition underneath it: the next product shot, the same ad resized for four placements, the next drop needing what the last drop needed. For the video-spot version of this, and the real cost behind the famous numbers, see how the most-cited AI commercials were actually made. For the operational side, scaling AI ad creative without burning budget covers the loop.
Almost nine in ten advertisers now use or plan to use generative AI to build video ads, and around a third of digital video ads are already made or enhanced with it (iab.com, July 2025). The mechanic is becoming the default. The brands above just ran it first, and loudly.
How to run the same play on one product photo
The mechanic that scales to a brand without a celebrity is the one all seven share: start from your actual product, then let AI produce the campaign around it. That is exactly what DesignerBox is built to do. You upload one product photo, and it turns into a full campaign, product stills, on-model shots, video ads, and social creative, all on brand.
The four mechanics map straight onto product work:
- Personalization at scale becomes variants per placement and per product, generated from one source photo instead of one shoot.
- Customer in the creative becomes on-model and try-on, putting your product on a model or in a scene without booking either.
- Synthetic spokesperson becomes an avatar or UGC-style presenter for a video ad, built from a brief.
- AI as the idea becomes the Product Ad Generator, turning a product shot into an ad concept without a designer.
Every top image and video model sits behind one subscription instead of six, so the model is a per-shot choice, not a per-tool bill. Save the campaign that worked as a workflow, and it reruns for the next product instead of being rebuilt.
Two honest limits before you budget it. Video is by far the most expensive operation, priced per second of output, so a single high-end clip with audio can run 6,400 credits, more than a full month on the mid plan, while an image edit costs 5 credits.
And AI video needs the Premium plan or higher, while the commercial license starts at Pro. There is a free plan if you want to see the output on your own product before you pick a tier. The full breakdown is on pricing and the model catalog.
FAQ
What is an AI advertising campaign?
An AI advertising campaign is any marketing campaign where generative or machine-learning tools do meaningful creative work, from generating the visuals to personalizing them at scale. In practice it means one of four mechanics: producing millions of variants of one asset, putting the customer inside the creative, delivering a synthetic spokesperson, or making the AI method the idea of the ad itself.
What was the first major brand AI advertising campaign?
Nutella Unica in 2017 is one of the earliest at scale, with an algorithm that designed 7 million unique jars for Ferrero (dezeen.com, June 2017), and Sephora’s ModiFace-powered Virtual Artist launched a year earlier in 2016. The generative-AI era arrived later: Heinz A.I. Ketchup in 2022 was among the first campaigns built on a text-to-image model, DALL-E 2 (musebyclios.com, July 2022).
Which brands have used AI in their advertising?
Coca-Cola, Nutella, Sephora, Nike, Heinz, Cadbury, and Virgin Voyages all ran landmark AI campaigns, verified above against trade press and agency sources. Adoption is now mainstream rather than experimental: the IAB reports that close to nine in ten advertisers use or plan to use generative AI for video ad creative (iab.com, July 2025).
Do AI advertising campaigns perform better?
Some do, and the ones that do share a trait: AI is pointed at variety and personalization, not at replacing the thing the audience came for. Sephora’s try-on and Cadbury’s per-store personalization drove engagement because the viewer saw themselves or their own business. Campaigns that use AI to replace human performance or warmth tend to draw backlash instead. The mechanic matters more than the model.
How much does an AI advertising campaign cost?
For the landmark campaigns, most of the cost was the agency, the concept, and the talent rights, not the AI. For a brand running the same mechanics today, the cost is credits. An image generation or edit is a few credits, while video is priced per second and is the expensive line. We break the numbers down in AI ad generator pricing.
Can a small brand run an AI advertising campaign?
Yes, and the realistic version is not a hero film. The repeatable win is product-level creative at volume: on-model shots, ad variants per placement, and the same campaign rerun for the next drop, all generated from one product photo. Start from the mechanic the big campaigns share, keeping the product constant and letting AI vary everything around it, rather than trying to reproduce the budget.
Every campaign above was verified against at least two sources as of August 2026, including Dezeen and Futurism (Nutella Unica), The Drum and the Coca-Cola Company (Create Real Magic), PR Newswire and Next Reality (Sephora Virtual Artist), LBBOnline and Wavemaker (Cadbury), Adweek and CNBC (Virgin Voyages Jen AI), AKQA and Adweek (Nike Never Done Evolving), Muse by Clios and Rethink (Heinz A.I. Ketchup), and the IAB 2025 Video Ad Spend & Strategy report. Widely repeated earned-impression and ROI figures are deliberately left out where they trace only to agency award-entry copy. Individual results vary.
