Dutch fashion group The Sting is using Google Cloud’s generative AI stack to automate product descriptions, generate model photography, and build a multi-brand content hub, all while keeping humans firmly in the loop.
At Google Cloud Live AI in Amsterdam, Techzine spoke with Martijn Schouwe, AI Manager at Distinct, and Martin du Prie from Xebia. Together, they are building AI-powered solutions for The Sting. The Sting is a fashion group that operates 145 stores across the Netherlands and Belgium. Like many enterprises, The Sting began its AI journey through experimentation. The first meaningful output was the automation of product descriptions. When a new garment arrives, the system captures an image, combines it with the brand’s tone-of-voice guidelines, and generates a product description. A content marketer reviews it before it goes live on the webshop.
This workflow reflects a principle that runs throughout The Sting’s AI approach: human oversight at every step of publication. Efficiency is gained, but editorial control is not surrendered.
The GenAI Photo Studio
The more ambitious initiative is The Sting’s GenAI Photo Studio, where AI-generated models wear actual garments in images intended for the webshop. Schouwe describes the goal as replicating the existing conventional photography process as closely as possible. This includes generating multiple shots per garment and requiring consent before any image goes live.
Getting there was not straightforward. Simply placing a garment on a surface and generating an image does not produce the quality needed for retail. The team had to work through how fabric drapes on a body. Think of how different materials, such as linen or leather, behave visually, and how to achieve consistency across thousands of SKUs. Schouwe notes that Google’s AI models already handle much of the material reasoning. This helps avoid the need to prompt for each image individually. A requirement for any solution that needs to scale.
Accuracy is a commercial imperative. If a generated image misrepresents a garment, think of wrong fabric texture, wrong fit, or wrong silhouette, customers who purchase based on that image are more likely to return the item. That outcome is precisely what the team is engineering against.
Also read: From pilot to production: what it really takes to run AI for real
The Content Hub: one platform, multiple brands
Du Prie from Xebia brings a production agency background to the project, and this shapes his contribution: a broader Content Hub built on Google Cloud. The premise is straightforward: brands need many different types of content, from Instagram statics to product detail page images to short-form video. Creating each type in a separate tool or workflow is inefficient. The content Hub could solve that.
The Content Hub provides a single UI/UX environment where brand teams can select the type of asset they need, generate it using the appropriate foundation model, and route it through an orchestration layer for review. Colleagues from legal or marketing can be assigned as reviewers, receive notifications, and manage their review queue from a dashboard.
Guardrails and grounding
Two technical elements make the Content Hub enterprise-ready. First, brand and visual guardrails ensure that generated content stays within each brand’s defined identity. Second, the model is grounded against real product data, a critical requirement for fashion. If the AI introduces a garment that doesn’t exist in inventory, customers receive something other than what they ordered. That drives up returns. Grounding closes that gap by anchoring generation to verified product assets.
Google’s foundation models at the core
The Content Hub draws on several of Google’s foundation models. Nano Banana handles image generation. Veo powers video creation. For short-form video content that requires a voiceover, Lyria is available for audio generation. Du Prie notes that Google productizes its own solutions but deliberately leaves space for industry-specific implementations. Something that works well for a fashion retailer operating across multiple brands in a specific regional market.
Scaling is the real challenge
Anyone with access to Nano Banana can generate a single compelling image. The challenge is generating 1,000 images of 1,000 different garments consistently, within brand guidelines, and connected to product and consumer data. And presenting it all through a UI that non-technical users can operate confidently. This is the problem the Content Hub is designed to solve, and it requires integration with existing e-commerce data infrastructure rather than standalone AI tooling.
The Sting also uses vectorized product descriptions from older garments as a reference layer for generating new ones, an application of vector search within the product description workflow, even if broader vector search capabilities live in their e-commerce platform rather than in the Google Cloud stack.
Also watch: How Mollie blends ML and AI in a regulated fintech
Social trend detection and the feedback loop
Looking ahead, Schouwe describes an ambition to detect trends on social platforms, and act on those. This can be done by generating relevant content in response, publishing continuously, measuring performance, and feeding results back into the next cycle. This closed-loop content engine is still on the roadmap, but the underlying components: trend detection, content generation, and performance measurement are being explored.
AI lets The Sting move faster and cheaper
This case is a good example of how a fashion company can move faster and cheaper with AI. The AI models are very good at creating on-brand product descriptions and can generate images and videos of the garments within minutes, whereas before they had to schedule a photographer and models to get all the photos taken. It saves time and money and drives innovation. It’s a good example of how AI can improve business processes and save on costs.