Luxury brands use AI for campaign imagery mainly as an extension layer: the hero campaign is still shot by photographers, and generative models produce the 10 to 20 matching assets each channel needs afterward. In 2025 that pattern held at Jil Sander, MCM, and Burberry, while Valentino's fully AI-led handbag video showed what happens when the extension tries to become the campaign. That split (extend the shoot, never replace it) is currently the sharpest line in luxury brand marketing.
How luxury brands actually used AI in 2025
The working pattern is extension, not replacement. Jill Asemota, whose Berlin studio Parallel Pictures produces AI imagery for MCM Worldwide and Peek & Cloppenburg, describes the typical brief: the campaign is already shot, and the brand needs 10 or 20 additional matching assets for social and digital [2]. The economics explain the volume: her studio's client data puts cost savings at up to 70% for AI e-commerce imagery and closer to 50% for campaign work [2].
The upside is not marginal. McKinsey estimates generative AI could add $150 billion, conservatively, and up to $275 billion to the apparel, fashion, and luxury sectors' operating profits within three to five years [1].
Four approaches, four risk levels
| Approach | 2025 example | What happened |
|---|---|---|
| Fully AI-led hero campaign | Valentino's DeVain handbag video (December 2025) | Hundreds of critical comments despite a clear AI label |
| AI-assisted concept imagery | Jil Sander (February 2025) | AI visuals folded into a broader conceptual campaign |
| AI campaign extensions | MCM Worldwide (spring 2025) | Surreal environments extending traditionally shot campaigns |
| AI-animated archive | Burberry (2025) | Movement added to 1980s campaign imagery; current product shots stayed photographic |
Cases as reported by Glossy, December 2025.
The risk gradient runs top to bottom. The further AI sits from the hero image, the more it saves and the less it threatens.
Why AI backfires faster in luxury
Valentino's December 2025 video was labeled as AI-generated and still drew hundreds of critical comments, "cheap" and "lazy" among the politer ones, accusing the house of prioritizing efficiency over artistry [2]. Transparency did not buy forgiveness.
Research published in the Journal of Advertising Research explains why. Across three experiments, when luxury ads used AI-generated images and disclosed it, consumers viewed the brand less favorably; the effect traced to a belief that AI ads require less effort, which reads as inauthentic [3]. A premium price is partly a claim about effort. Visible shortcuts contradict the claim.
The escape hatch is craft
The same research found a moderator: when AI-generated ads showed highly creative, unusual visuals beyond what traditional production could achieve, the negative effect shrank significantly [3]. Consumers do not punish the technology; they punish the impression that it replaced effort. AI as an obvious cost-cut fails in luxury. AI as a visible creative reach does not.
Lock the brand codes before you generate
A luxury image is recognizable before the logo appears, because the codes do the work: one light quality, one palette discipline, one way of framing the product, one register of set design. Generative tools will happily produce imagery that ignores all four, so the codes have to be written down before anyone opens a prompt box – as reference images and short rules a reviewer can check an asset against, not as a mood adjective.
Product fidelity is the hardest code. Hardware, stitching, and material color are exactly what generic generators redraw, and color drift is visible to any customer holding the product; the same discipline applies to brand color accuracy in product photography.
Three checks every generated asset passes
Review generated extensions the way a retoucher reviews a shot: the product matches the physical sample, the light behaves like the campaign's light, and nothing in the set breaks the brand's register. One person owns the veto. An asset that fails any check dies without discussion, because the archive of approved imagery is the brand, and one off-code asset in circulation lowers the bar for the next one.
A campaign-extension workflow that scales
- Shoot the hero campaign as usual. The photographic layer stays; it is what the extensions inherit from and what the brand's effort claim rests on.
- Codify the look from the selects. Pull final selects into a reference set that defines light, palette, and framing – the source of truth for every generated asset. Studio presets are the tooling version of this step: a saved scene applied identically across generations.
- Generate channel variants. The 4:5 crop for social, the 16:9 for the site hero, the colorway the shoot skipped, the quiet product still for retail screens.
- Run the three checks. Product fidelity against the sample, light against the selects, register against the codes. Kill what fails.
- Archive what ships. Approved extensions join the campaign assets, so next season's step 2 starts from a richer reference set.
Where the savings land
Steps 3 through 5 are where the 50% campaign-work savings reported by Parallel Pictures actually accrue – volume without extra shoot days [2].
Where Kive fits for luxury teams
Kive is built for the extension layer of this workflow. A trained product model keeps the same trench, chain, or jar rendering consistently instead of being redrawn per image, and a studio applies one saved look (light, set, grade) across every asset generated from it, which is the codes-first discipline in tool form. Extending an image's canvas covers the aspect-ratio spread a campaign needs, and generated assets land in the same searchable library as the campaign selects they extend.
What the setup costs
Training a brand style or product model sits on Kive's Pro plan, which starts at $100 per month with 5,000 credits; the Basic plan at $20 per month covers generation and editing without custom training. That pricing targets teams shipping campaigns quarterly, and it is the honest boundary: a house producing one fully art-directed campaign a year with no extension volume has little use for the machinery, and a brand chasing surreal AI-native stunts wants a different tool entirely. Kive optimizes for the same product photographed many times, which is what luxury consistency demands.
The luxury houses that navigated 2025 well share one trait: they used AI where customers were not being asked to admire it. Extensions, crops, colorways, and archive animation absorbed the volume; the hero image kept its photographer.
That division of labor is likely to hold even as the models improve, because the constraint was never image quality. It is what the imagery says about effort, and effort is the product. The brands that scale generative imagery without damage will be the ones whose generated assets are indistinguishable in discipline, not just polish, from the campaign they extend.
Studios with a luxury register
References
