Case studies / FurnishVision
Generative AIClient project · Home retailJan – Jun 2026

FurnishVision

Shoppers photograph their room and see the rug, curtains and cushions in it before they buy.

Built for
A home furnishings retailer with 30 years in business, online and at exhibitions
Platform
Web shop and Android exhibition app
Status
Live web app, exhibition app prototype
The FurnishVision homepage invites shoppers to see rugs, curtains and cushions in their own room before buying.
1 dayto build the web platform
8products in a single render
~90sto transform a room photo
2platforms: web and Android
Overview

The short version.

FurnishVision answers the question every furnishing shopper asks: how will this look in my room? A shopper uploads a photo of their room, picks products, and gets a photorealistic image with those exact products placed in it. Renders can be layered, so a rug can go down first and curtains go up after.

For the retailer's exhibition stand we built an Android app where visitors photograph their own room on the spot and see it transformed while they wait.

The challenge

What was broken.

Online furnishing shoppers all ask the same question. The retailer's sales team answered it one WhatsApp message at a time, shoppers hesitated at checkout, and returns ate into margins. At exhibitions, visitors loved the fabrics and walked away without buying because they couldn't picture them at home.

  • Sales staff answered "how will it look?" by hand, one message at a time.
  • Shoppers abandoned carts they weren't confident about.
  • Returns from products that looked different at home.
How it came together

From idea to production.

  1. 23 Jan

    Proposal and v1

    Proposal, pricing and the first working web platform, all in one day.

  2. Jun 2026

    Better renders

    Moved to a newer image model with better realism and product fidelity, running as background jobs.

  3. 27–30 Jun

    Exhibition app

    An Android booth app with a guided visitor flow, and a queue-based render pipeline to handle the rush.

The solution

What we built.

The shopper takes or uploads a photo, chooses up to eight products, and FurnishVision composites them into the room at a believable scale, keeping the existing furniture intact. Each product's real dimensions guide its size, and each fabric's light behaviour is described honestly, so a blackout curtain lets only a soft glow through at the edges. The finished look goes straight to the cart.

How it works
  1. Step 1Shopper photographs the room
  2. Step 2Picks rugs, curtains and cushions
  3. Step 3AI places the products
  4. Step 4Before and after reveal
  5. Step 5Whole look added to cart
Key features

What we delivered.

Multi-product renders

Up to eight products placed into one room image.

Layered looks

Use a finished render as the base for the next product.

True-to-size placement

Real product dimensions guide the render, scaled against sofas, doors and ceilings in the photo.

Honest light

Sheers glow, blackouts block. Each fabric's real light behaviour shapes the image.

Exhibition mode

Staged progress messages during the wait, a sample room if a photo fails, and an automatic reset between visitors.

Shop and checkout

Catalogue, product pages, cart and checkout on web and mobile.

Inside the product

A closer look.

Where the AI does the work

Intelligent, and trustworthy.

Background rendering

High-quality renders take about 90 seconds, longer than a normal web request allows, so they run as background jobs the app checks on.

Keep the room as it is

The instructions to the image model stop it redrawing the customer's existing furniture.

Built to handle a crowd

The exhibition pipeline queues renders so a busy stand never overloads the system.

AI-made catalogue imagery

Product and banner images for the demo catalogue were themselves generated with an image model.

Built with

The stack.

Web
Next.js 16 shop with ImageKit image delivery
Mobile
React Native (Expo) Android app
AI
OpenAI image editing, earlier Gemini
Rendering
Background functions, then Cloudflare KV and Queues
Results

The numbers.

Build proof

1 day
web platform build
Verified
8
products per render
Verified
~90s
room render time
Verified
2
platforms shipped
Verified

Expected impact for users

+25%
add-to-cart on visualised products
Projected
−50%
"how will it look?" enquiries
Projected
−30%
returns
Projected
1 in 3
booth visitors leaving with an order
Projected

Verified from project repositories, test runs and release records. Projected modelled estimates of user impact, not measured client results.

Your product next

Got an idea? Let's ship it.

Tell us the workflow that's costing you customers or hours. We'll come back with a prototype plan, a timeline in weeks, and a fixed quote.