Published September 4, 2023 · By CampaignsLive · Case Study
In March 2023, Levi’s announced a partnership with Lalaland.ai, an Amsterdam-based startup, to generate AI models for the brand’s e-commerce product imagery. The framing in the announcement was about diversity: AI-generated models would let Levi’s show its products on a wider range of body types, ages, and skin tones than the typical e-commerce shoot would deliver. The internal logic was that broader representation was the goal and AI generation was the most cost-effective path to it.
The backlash was immediate. It was also, in retrospect, less about the technology than about the framing.
What the criticism actually said
The argument against the campaign, in the more measured corners of the response, was not that AI-generated models were inherently wrong. It was that using AI as a diversity strategy was a category error. Diversity in advertising is, fundamentally, about representation — about hiring the real people the work is meant to reach, paying them, putting them in front of the camera, and crediting them. Generating synthetic models with a wider range of skin tones did not produce more representation; it produced more variety in a synthetic image. The actual people in the categories Levi’s was claiming to represent were not on set.
The economic argument compounded this. AI-generated models are cheaper than real models. A brand that frames the move as a diversity win, while pocketing the cost savings and removing the work from a real workforce, is not having a conversation about representation. It is having a conversation about cost, wearing a representation costume.
The critics’ position was not anti-AI. It was: be honest about why you are doing this. If the move is about cost, say so. If it is about scale, say so. Do not pin it on representation.
What Levi’s could have said instead
A coherent version of the announcement might have read like this: AI-generated imagery lets us show the product on more body types, in more contexts, at lower cost than a multi-week shoot. This is a production tool. The real models we work with on brand campaigns are not affected. Our editorial and commercial talent partners remain in place. This is e-commerce thumbnail generation, not campaign creative.
That framing would have been honest, defensible, and almost certainly less controversial. It also would not have generated the press attention that the original framing did, which is part of why the original framing was chosen.
The deeper question
Beneath the framing problem, there was a real question that the campaign exposed and that the industry has been working through ever since: where does generative AI belong in the production stack, and where does it not?
For e-commerce product imagery — thumbnails, color variants, environmental contexts — the case for generative tooling is strong. The work is high-volume, low-equity, repetitive, and economically painful to produce traditionally. A single product launch can require hundreds of variant images, each of which is a small line item in a large total. Generation makes this economical at a scale that traditional shoot production cannot match.
For brand campaign creative — the work that defines the brand’s visual identity, runs in paid placements, shapes the equity — the case is more nuanced. The talent in front of the camera is part of the brand story. The casting decisions communicate values. The work has weight beyond its production cost.
Levi’s announcement collapsed these two categories. The use case was the first one. The narrative around it implied the second.
What the industry learned
The episode taught several things that the next two years of brand-AI conversation have been working out.
The first is that disclosure norms matter. Brands that disclose AI use, frame it honestly, and locate it appropriately in the production stack fare better than brands that lead with a more flattering narrative.
The second is that the economic argument cannot be hidden. Whatever the framing in the announcement, the cost savings will surface in industry coverage. A brand that lets the cost argument be made by critics, rather than making it themselves, ends up with the worse version of the story.
The third — least discussed at the time — is that the labor question is real. AI imagery removes work from people whose livelihoods depend on that work. The right response to this is not to pretend it does not happen; it is to think clearly about which categories of work AI should replace, which it should augment, and which it should leave alone. Levi’s announcement did not appear to have done that thinking.
Where this sits in the 2023-2026 timeline
The Levi’s-Lalaland campaign sits in the same period as Coca-Cola’s “Create Real Magic” and a wave of other brand-AI experiments. Coca-Cola’s version was structurally sound: a constrained, curated, brand-anchored co-creation platform with clear permissions architecture. Levi’s was structurally sound at the use-case level (e-commerce thumbnails) but framed incoherently (as a diversity initiative).
The two cases, read together, are a useful diagnostic. The brands that have done well with AI imagery since have been the ones that locate the work clearly in the production stack, frame the decision honestly, and let the technology be a tool rather than a brand-story prop. The brands that have struggled have been the ones that tried to make the technology itself the story, on terms the audience could see through.
For more on how brands have navigated the IP and labor questions since, see Why Your Brand Should Own Its AI Creative. For the broader argument about training data and output character, see AI Creative vs. AI Slop.