The challenge
Audley Villages, one of the UK's leading luxury retirement village operators, had a vision for 2025: Audley Bespoke, a new interior design service for new owners. The idea was to help people visualise their future apartment before they moved in.
Every Audley apartment is delivered as a blank canvas. White walls, cream carpet, empty rooms. For prospective owners making one of the biggest decisions of their later life, imagining what their new home could look like was a leap of faith. Show apartments helped, but they only showed one vision. Printed mood boards offered options, but they felt flat and impersonal.
The growing number of people researching online before visiting faced an even bigger gap. They could see floor plans and empty room photographs, but there was nothing to bridge the distance between "bare magnolia walls" and "this feels like home."
When Audley approached us about the campaign, we suggested using AI to generate the visualisations. It would let every prospective owner see their style, their colours, their taste. Instantly.
Our approach
We built the Audley Bespoke Inspiration Creator: an interactive tool that lets users design their dream apartment and see it brought to life with AI-generated photorealistic visualisations.
The concept sounds simple. In practice, it required solving three distinct challenges:
The experience challenge: How do you guide someone through multiple choices — room, layout, furniture style, colour palette — without overwhelming them? Especially when your audience spans their 60s to 80s and expects something premium, not clinical.
The AI challenge: How do you generate over 1,200 photorealistic room images that look consistent, believable, and true to actual apartments, rather than the idealised spaces AI models tend to create?
The integration challenge: How do you turn design exploration into captured leads that feed directly into a sales pipeline, all while embedding seamlessly within an existing Drupal website?
Discovery: understanding what actually matters
Our discovery phase focused on what mattered most to Audley's audience. The insights shaped everything that followed.
Room selection matters. Different owners care about different rooms. Some want to see the living room first; others prioritise the bedroom or study. The tool needed to let users start where their interest lies.
Colour is deeply personal. Rather than offering named "themes," we used visual colour swatches based on Coat paint palettes. The 13 palettes were carefully curated to reflect the natural, muted tones that appeal to Audley's demographic — from soft greys and taupe through to warmer sand, terra, and green options.
Furnished vs. unfurnished. Early testing revealed people wanted to see both: the painted-but-empty room (to understand the colour scheme) and the fully furnished version. This led to a toggle that lets users flip between views.
Mobile-first, but not mobile-only. Most initial browsing happens on phones and tablets, but the tool also needed to shine on desktop when owners sit down for a more considered look.
The 8 furniture styles were chosen to span a broad range of tastes while remaining appropriate for the Audley brand: Contemporary Luxury, Classic Traditional, Scandi Minimalist, Mid-Century Modern, Industrial Chic, Art Deco Glamour, French Provincial, and Coastal Hamptons.
The AI challenge: getting machines to respect reality
The most technically demanding aspect of the project was prompt engineering: getting AI-generated images to look right while respecting the constraints of real apartments.
Here's the thing about generative AI: it has strong opinions about interior design, and they don't always match reality.
Floor replacement. The AI consistently wanted to replace cream carpet with hardwood or parquet flooring. Every prompt had to explicitly state "NEVER change the existing cream/beige carpet" as a non-negotiable constraint.
Ceiling embellishment. AI models love coving, cornices, crown moulding, and ceiling roses — none of which exist in actual Audley apartments. Multiple reinforcements were needed to keep ceilings plain.
Wall patterns. Without explicit instruction, the AI added stripes, geometric patterns, and decorative paint techniques. Each prompt explicitly forbade all of these.
Furniture blocking doors. A practical issue: the AI would place sofas in front of doors, making rooms look impractical for actual living.
We developed a two-step generation process. First, an "unfurnished" variant is created: the empty room with painted walls and curtains in the selected palette. Then, the furnished variant is generated using the unfurnished image as its base. This ensures colour consistency and gives the AI a stable foundation.
The result: 1,248 unique room visualisations across 6 room layouts, 8 furniture styles, and 13 colour palettes — each in furnished and unfurnished variants.
The business case: AI vs traditional photography
To appreciate what AI generation made possible, consider what the traditional approach would have required.
For 1,248 unique room visualisations across 624 distinct style/palette combinations, a conventional photography approach would need:
Traditional approach (per styled room):
- Furniture hire and room styling: £800–2,000
- Professional interior photography: £300–500
- Setup and breakdown time: 1–2 days
Scaled to 624 unique styled looks:
- £750K+ in styling costs alone
- £250K+ in photography costs
- 2–4 years of calendar time
- £1M+ total traditional cost
The AI approach required approximately 50 hours of prompt engineering, multiple regeneration cycles to refine quality, and API generation costs. That delivered the same 1,248 images in weeks rather than years, at a fraction of the cost.
More importantly, traditional photography would have locked in those 624 combinations permanently. With AI, adding a new colour palette or furniture style means generating new images — not re-staging and re-shooting entire rooms.
Building for the audience
Design decisions were guided by Audley's brand positioning: elegant, welcoming, reassuring, premium.
The quiz flow was designed to feel effortless. Each step presents a single choice as large, tappable cards. Room, then layout, then style, then palette, then preview, then gallery, then form. Users see a preview of their first design before entering the gallery, building anticipation.
Key UX details that made the difference:
- Progressive disclosure. One choice at a time, with clear visual hierarchy
- Favouriting and download. Users can heart their favourite images and download designs for offline reference — important when decisions involve family discussions
- Browser history integration. The back button works naturally throughout, a detail that matters enormously for usability but is often overlooked
- State persistence. Progress saves to localStorage, so users can return without starting over
- Accessible throughout. Proper ARIA labels, keyboard navigation, focus management, and semantic HTML
Drupal integration
The React app embeds within the existing Drupal site, with form submissions flowing through Drupal's Webform REST API. Leads appear in Drupal's webform submissions alongside all other enquiry forms, fitting naturally into the existing sales workflow.
A view-only mode allows Audley staff to see exactly what a customer selected: the same gallery, same liked images, via a URL with query parameters. Sales teams can walk into conversations already knowing an owner's taste.
The results
- 1,248 unique room visualisations
- <2s image load times
- 15+ GA4 custom events tracking user behaviour
- 320px+ fully responsive from mobile to desktop
More importantly, the analytics give Audley's marketing team unprecedented insight into customer preferences: which room types are most popular, which furniture styles resonate, which colour palettes are selected most frequently, and where users drop off. This data feeds back into both the tool and broader marketing strategy.
What's next
The Audley Bespoke tool was built for expansion:
- Village-specific imagery — using photographs from specific villages so owners can see their actual apartment visualised
- Real-time generation — as AI models become faster and cheaper, moving to on-demand generation for unlimited combinations
- Deeper CRM integration — passing design preferences directly into Salesforce so sales teams can personalise conversations from the first contact
Technology stack
React 19, TypeScript, Vite, Tailwind CSS, Google Gemini, Cloudflare R2, Drupal Webform, Netlify