How does a text prompt become a manufacturable mosaic?

The prompt goes to a Flux-based image model tuned for mosaic aesthetics, which returns design options. The chosen design is colour-quantised with K-Means clustering so it only uses colours you can actually supply, then converted to a tile grid with a count of tiles per colour that the workshop can lay out.

How are generated colours matched to real tiles?

Generated images contain millions of colours, but a physical tile catalogue has a limited set. K-Means clustering reduces the image to the size of your palette (typically 50 to 100 colours), then each cluster is matched to the closest tile colour using colour-difference (Delta-E) matching.

What size is the tile grid?

For Mozaico the design is converted into a 5 mm tile grid, with edge-aware logic so fine detail survives the conversion to discrete tiles. The grid size is a parameter: we set it to the tile size you manufacture in.

Can customers order the design themselves?

Yes. For Mozaico the pipeline is integrated with their Shopify backend: the customer enters the dimensions they want, sees the price, and adds the custom design to the cart. The system produces a bill of materials listing how many tiles of each SKU the piece needs.

Can it use our own tile catalogue?

That is the point of the colour-matching step. We analyse your physical tile collection, glass, marble or ceramic, and build the palette from the colours you stock, so nothing is generated that you cannot produce.

How long does a pilot take?

A proof of concept on your own tile catalogue and design styles typically takes 4–6 weeks. We agree up front what "manufacturable" means for your workshop and measure the output against it.

Generative AI for tile makers

Generative AI Mosaic & AI Tile Mural Art

Customers describe the artwork they want and get a mosaic design back in seconds, matched to the tiles you already stock.

Talk to an AI engineer Read the Mozaico case study

Designs from a plain-English prompt
Colours matched to your tile palette
Tile counts and orders, not just pictures

Mosaic of a striding tiger in orange, blue and black tiles on a pale tiled background — Generated designs. Each one started as a sentence a customer typed.

Mosaic of a red and gold fish swimming above flowers in a blue underwater scene — Generated designs. Each one started as a sentence a customer typed.

Why automate custom design

Custom mosaic work doesn't scale by hand

Every bespoke piece needs a designer to interpret a description and map it onto tiles that actually exist.

Four natural stone mosaic tile samples labelled Valentina, Travertino Gold, Rosso Laguna and Rosso Alicante — Every design has to resolve to tiles like these: the ranges you actually stock.

Days per concept
Sketching a design and revising it with the customer takes days.
Colours that don't exist
A render is useless if its colours are not in your catalogue.
Manual grid mapping
Mapping artwork onto a tile grid by hand is slow and error-prone.

Topic	Manual custom design	Generative design pipeline
Time to first design	Days of designer time per concept	Seconds from prompt to design options
Colour accuracy	Designer picks tiles by eye from the catalogue	Every colour snapped to a tile you stock
Layout	Grid mapped by hand, then checked	Tile grid generated with a count per tile colour
Ordering	Quote by email, then back and forth	Size, price and cart on the storefront

How it works

From prompt to bill of materials

Generation is only the first step. The stages after it are what make the output something a workshop can build.

custom-mosaic_sunset-sail

Designing

Design your own mosaic

A sailboat at sunset on a calm sea, warm coloursGenerate

Style: mosaic4 design optionsYour tile palette

1/4The customer describes the artwork they want in plain English, right on the store page.

Stage 1
Capture the customer's idea
The customer types what they want in plain English on your storefront, along with the size and the room it is for. No design skills, no brief, no waiting.
Prompt and size
Stage 2
Generate design options
Flux Dev turns the prompt into high-resolution artwork. Guidance scale and sampling are tuned, with mosaic-style prompt engineering, so results stay in a mosaic aesthetic rather than drifting into photo-realism.
Several design variationsFlux DevPrompt engineering
Stage 3
Quantise colours to your palette
K-Means clustering reduces millions of generated colours to the size of your physical palette, typically 50 to 100 colours, and each cluster is matched to the nearest real tile by colour difference.
Design in stocked tile colours onlyK-Means clusteringDelta-E matching
Stage 4
Convert to a tile grid
The quantised image becomes a precise tile grid, 5 mm for Mozaico, with edge-aware logic so outlines and fine detail survive the jump from pixels to discrete tiles.
Manufacturing layout5 mm gridEdge-aware conversion
Stage 5
Price, order and manufacture
The customer enters the dimensions, sees the price and adds the piece to the cart. The system creates the product in Shopify and a bill of materials with the tile count per SKU for the workshop.
Shopify order and bill of materialsShopify APIAutomated BOM

In production

The design-to-order flow we built for Mozaico

Mozaico sells custom mosaics worldwide. Their customers now go from a description to a ready-to-order tile layout on the store itself.

Step 1: the customer describes the piece

A prompt box sits on the storefront. The customer types an idea in plain English, presses generate, and the pipeline takes over. No designer is involved until the customer has something they like.

Runs inside the existing store
No design experience needed
Sizes and options captured up front

Read how the platform was built

Mozaico storefront page with a banner reading Generate Your Own Mosaic Design With AI, an idea input field and a Generate button — The generator on Mozaico's site: enter an idea, press generate.

Step 2: pick from generated designs

The model returns several variations of the same idea in a mosaic style. The customer picks the one they want and moves on, all in the same three-step flow on the site.

Multiple variations per prompt
Mosaic aesthetic, not generic AI art
Regenerate until the customer is happy

Mozaico design selection step showing four AI-generated motorcycle mosaic variations with the third selected — Four variations from one prompt, in a Dream it, Style it, Own it flow.

Step 3: size it, price it, order it

The customer enters the width and height they need and the price is calculated for that size. The design is added to the cart in Shopify, and the workshop gets a layout with the tiles it requires.

Price generated from the dimensions
Custom design added straight to the cart
Tile counts per colour for production

Try the live generator on Mozaico

Mozaico ordering step with width and height inputs, a calculated price and an Add to Cart button beside the chosen mosaic design — Dimensions in, price out, then straight to the cart.

Design showcase

Mosaic designs generated from prompts

Each of these started as a sentence. The tile texture is part of the generation, so what you see is close to what the workshop lays.

Prompt: “A striding tiger in orange and blue glass tiles”
Prompt: “A koi fish above flowers in deep blue water”
Prompt: “A portrait in profile on a warm two-tone background”
Prompt: “A canal between brick buildings, city skyline behind”
Prompt: “Riders with banners crossing a desert ridge”
Prompt: “A red classic sports car, side three-quarter view”

What we build around it

The parts that make it a product, not a demo

A generator on its own creates pictures. These are the pieces that turn it into orders your workshop can fill.

Your material library
We take your physical tile collection, glass, marble or ceramic, and turn it into a colour palette the pipeline matches against, so every design is producible.
Palette built from your catalogue
Style control
Prompt templates and model parameters keep output on-brand: classical, geometric or contemporary, rather than whatever the model feels like producing.
Consistent house style per collection
Grid and size logic
Tile size, panel dimensions and edge handling are parameters, so the same design can be produced at different scales without redrawing it.
Grid set to the tile size you manufacture
Bill of materials
Every approved design produces the tile count per SKU, so purchasing and production work from numbers instead of estimates.
Counts per tile colour and SKU
Storefront integration
Product creation, pricing by dimensions and cart handling in Shopify, so a custom design is an ordinary order for the rest of your systems.
Shopify product and checkout
Room previews
The same generative work sits behind our virtual staging pipelines, so a design can also be shown in a room before a tile is cut.
Optional visualisation step
Virtual staging case study

Integration and deployment

Built into your store and your workshop

The pipeline is a service your systems call, not a separate tool your team has to visit.

Shopify integration for product creation, pricing and checkout
An API your own storefront or ERP can call if you are not on Shopify
Generation on GPU infrastructure we size to your traffic
Your catalogue, prompts and designs stay yours

Custom AI solutions Model training and tuning Private deployment

API first
Generate, quantise and lay out as separate endpoints, so you can plug them into any flow.
E-commerce sync
Designs become real products with prices, images and order data attached.
Catalogue updates
Add or retire tiles and the palette updates; new designs use what you stock today.
Production handoff
Layouts and tile counts exported for the workshop in the format your team already uses.

Getting started

Prove it on your own tiles first

We start with your catalogue and a handful of designs your team would normally draw by hand.

Step 1: Scoping call
30 minutes
We look at your tile catalogue, the styles you sell and how orders reach production today.
- NDA on request
Step 2: Proof of concept
4–6 weeks
A working prompt-to-layout pipeline on your palette, checked against designs your team has produced by hand.
- Judged by your workshop
Step 3: Production
Ongoing
Integrated with your storefront and production handoff, monitored, and tuned as your catalogue changes.
- You own the IP

Not sure this is the first thing to automate? Start with an AI Opportunity Audit and we will map where AI pays off across your business.

FAQ

Questions, answered

What tile makers and manufacturers ask before a pilot.

Want the engineering detail? Read the Mozaico case study.

The prompt goes to a Flux-based image model tuned for mosaic aesthetics, which returns design options. The chosen design is colour-quantised with K-Means clustering so it only uses colours you can actually supply, then converted to a tile grid with a count of tiles per colour that the workshop can lay out.

Case studies

Book a strategy session

Talk to an AI engineer about your project

Tell us what you want to automate. The first call is a 30-minute working session with an engineer, not a sales pitch.

Send the form, it takes 2 minutes
We reply within 1 business day, under NDA if you need it
A 30-minute call to scope feasibility and next steps

Generative AI Mosaic & AI Tile Mural Art

Custom mosaic work doesn't scale by hand

Days per concept

Colours that don't exist

Manual grid mapping

From prompt to bill of materials

Capture the customer's idea

Generate design options

Quantise colours to your palette

Convert to a tile grid

Price, order and manufacture

The design-to-order flow we built for Mozaico

Step 1: the customer describes the piece

Step 2: pick from generated designs

Step 3: size it, price it, order it

Mosaic designs generated from prompts

The parts that make it a product, not a demo

Your material library

Style control

Grid and size logic

Bill of materials

Storefront integration

Room previews

Built into your store and your workshop

API first

E-commerce sync

Catalogue updates

Production handoff

Prove it on your own tiles first

Step 1: Scoping call

Step 2: Proof of concept

Step 3: Production