How does PUMA use generative AI for campaign imagery and product content?
PUMA uses Google's Imagen models to generate localised backgrounds and handle routine image editing for its online shop. The case shows where generative imagery works for a global brand and what it depends on.
KEY TAKEAWAYS Summary by the editors
- PUMA uses Imagen 2 on Google Cloud's Vertex AI to generate product imagery with backgrounds tailored to the product, the customer and the region, according to a Google Cloud announcement of 24 September 2024.
- Google Cloud described PUMA as the first in its industry to use AI-generated imagery at scale in its online shop.
- PUMA also uses Imagen for routine editing tasks such as shadows, composition, colour accuracy, resolution and product positioning.
- Consumer Goods Technology, citing the Wall Street Journal, reported that PUMA's localised AI imagery improved click-through rates by 10% in India.
- PUMA's approach keeps the product itself real and uses generative AI for settings and editing, a lower-risk pattern than generating models or products.
PUMA uses generative AI to create localised backgrounds for product images and to automate routine image editing in its online shop, using Google's Imagen models on Vertex AI. The product stays the real product; AI changes the setting so that, for example, a shopper in Japan can see a trail running shoe near Mount Fuji. PUMA links the approach to higher click-through rates and faster campaign launches.
What did PUMA build with generative AI?
On 24 September 2024, Google Cloud announced an expanded partnership with PUMA to improve its digital shopping experience with generative AI. According to the announcement, PUMA uses Imagen 2 on Vertex AI to generate product imagery, including custom backgrounds tailored to the product, the customer and the region. Google Cloud described PUMA as the first in its industry to use AI imagery at scale in its online shop.
The example given was Japan: a shopper there might see a lifestyle shoe on the streets of Ginza in Tokyo, or a trail running shoe near Mount Fuji. Regional marketers can customise content for their own markets rather than relying on a single global image set. PUMA chief executive Arne Freundt said that "Google Cloud's generative AI has allowed us to create an immersive experience that is tailored to each consumer".
Which content tasks does PUMA automate?
Beyond backgrounds, PUMA uses Imagen for editing work that content teams previously did by hand. The announcement lists:
- Shadows and lighting consistency
- Composition and product positioning
- Colour accuracy
- Resolution improvements
These tasks matter commercially because they are repetitive and must be done for every product, colourway and market. Automating them frees retouchers and content editors for work that needs judgement, and shortens the time between a product arriving and being ready to sell online.
In a typical workflow of this kind, the brand photographs the product once under controlled conditions, separates it cleanly from its background, and then generates variants of the scene around it for different markets, channels and formats. The original product pixels are preserved, and a human editor reviews outputs before publication. Google Cloud referred to the system PUMA built as a "creative agent", which implies tooling that marketers use directly rather than a one-off campaign produced by an agency. That design spreads the benefit across many regional teams instead of a single launch.
| Use | Purpose | Prerequisite |
|---|---|---|
| Localised backgrounds | Make imagery relevant to regional shoppers | High-quality cut-out product images; regional creative guidance |
| Image editing (shadows, colour, composition) | Reduce manual retouching | Colour-accurate source photography; quality review |
| Personalised content | Tailor imagery to customer and context | Customer and regional data on the e-commerce platform |
| Search and recommendations | Improve product discovery and order value | Clean product data and catalogue structure |
What results has PUMA reported?
Published figures are limited and should be read carefully. Google Cloud's announcement said the imagery improves click-through rates and speeds time to market for campaigns, but gave no numbers. Consumer Goods Technology, citing the Wall Street Journal, reported that click-through rates improved by 10% in India where images were personalised by geography. The same report noted that Google Cloud had said in March 2024 that PUMA's average order value rose 19% following integrations of BigQuery and Google Analytics, which relates to data and analytics rather than generative imagery.
PUMA also said that after moving its e-commerce ecosystem, including PUMA.com, to Google Cloud in 2024, it saw higher average order value from personalised content and could make products shoppable as soon as they reached the warehouse. No figures were given for those effects. The company said it would explore Imagen 3 and extend Vertex AI Search for Retail to more subsidiaries.
Why has PUMA's approach drawn less criticism than other AI campaigns?
Several fashion brands faced strong criticism in 2025 for AI-generated campaign imagery, particularly where AI created people or replaced a photographed product. PUMA's documented use differs in two ways. First, the product remains the real item; AI changes the environment around it and performs edits a retoucher would otherwise make. Second, the purpose is relevance to the shopper (a local setting) rather than replacing a creative shoot. For a sportswear brand selling in more than 120 countries, localisation at scale is a problem that manual photography struggles to solve economically.
What are the risks of generative product imagery?
Even with real products, generated imagery carries risks that brands must manage:
- Product accuracy: any change to colour, shape or detail can mislead shoppers and drive returns, so colour fidelity needs checking against the physical sample.
- Cultural fit: automatically generated local scenes can include inappropriate or inaccurate settings; regional review is needed.
- Rights and provenance: generated backgrounds must not reproduce protected locations, artworks or brand marks.
- Disclosure: rules on labelling AI-generated content are tightening, including transparency duties under the EU AI Act from August 2026.
- Measurement: uplift claims need controlled tests by market, not before and after comparisons.
What can other brands learn from PUMA?
PUMA's case suggests a practical order of work. Put product data, photography and the e-commerce platform on a consistent footing first; PUMA's generative imagery sits on a platform migration and analytics integration. Then apply generative AI to high-volume, low-risk tasks such as backgrounds and retouching, give regional teams controlled freedom to localise, and measure by market. Brands that start with AI-generated models or products take on the most visible risks first, while PUMA's documented use starts where the value is clearest and the reputational exposure is lowest.
Frequently asked questions
Does PUMA use AI-generated images?
Yes. PUMA uses Google's Imagen 2 on Vertex AI to generate localised backgrounds for product images and to automate edits such as shadows, colour accuracy and composition, according to Google Cloud's September 2024 announcement.
Which AI does PUMA use for imagery?
PUMA uses Google Cloud's Imagen models on the Vertex AI platform. In 2024 it used Imagen 2 and said it would explore Imagen 3.
Did AI images improve PUMA's results?
Consumer Goods Technology, citing the Wall Street Journal, reported a 10% improvement in click-through rates in India from geographically personalised images. Google Cloud's announcement cited improved click-through rates and faster time to market without figures.
Is AI product imagery risky for brands?
It can be if products are altered or settings are inappropriate. Keeping the real product, checking colour accuracy, reviewing local scenes and labelling AI content where required reduces the risk.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- Google Cloud Press Corner: PUMA transforms product campaign development with Google Cloud's generative AI
- Consumer Goods Technology: Puma boosting CTRs with Google Cloud generative AI tools
- Retail Insight Network: PUMA partners with Google Cloud on generative AI
- EMARKETER: Puma, Walmart and Amazon among retailers turning to genAI for product listings
- EU Artificial Intelligence Act (artificialintelligenceact.eu): Article 50, transparency obligations
