About GetPaintByNumber

We make one thing, and we try to make it properly: a paint-by-number generator that is genuinely free, gives you real control over the output, and never asks you to hand over your photos.

Why this exists

Custom paint-by-number kits are a lovely gift, but the way they are usually sold is frustrating. You upload a photo of your dog to a shop you have never heard of, pay somewhere between $40 and $80, and wait — without ever seeing what the design will look like. If the conversion turns your dog's face into an unreadable smear of forty near-identical browns, you find out when the box arrives.

The conversion itself is not the expensive part. It is an image-processing problem that a modern browser can solve in a couple of seconds on the device already in your hand. The canvas, the paints and the shipping are what cost money. So we separated the two: the design is free and instant, and what you do with it afterwards — print it at home, transfer it to canvas, or take it to a print shop — is entirely up to you.

The short version: you should be able to see the design before you commit anything — money, an email address, or your family photos.

What we believe

How the algorithm works

A paint-by-number template is a colour-quantised version of a photograph: millions of shades reduced to a small set of paints, grouped into regions, each labelled with a number. Getting that to look good takes five steps.

1. Noise reduction A light separable blur removes JPEG artefacts and sensor grain. Skip this and a smooth cheek or a clear sky shatters into hundreds of one-pixel specks that no brush could ever follow.
2. Clustering in CIE Lab Pixels are converted from RGB into Lab, a perceptual colour space, then grouped with k-means++. This matters more than any other decision: clustering in Lab means two greens that look alike end up sharing a paint pot, which plain RGB clustering frequently gets wrong.
3. Region cleanup A majority filter smooths ragged boundaries, then every region below a minimum area is absorbed into the neighbour it shares the longest border with. This is the step that turns a noisy pixel map into shapes with clean, followable edges.
4. Number placement A chamfer distance transform finds the point furthest from any edge inside each region — its visual centre — and the digit is scaled to the space actually available. Regions too thin to hold a legible number are deliberately left blank rather than crowded.
5. Palette ordering Paints are numbered from lightest to darkest, because that is the order most painters work in, and because a darker colour straying over a boundary covers a lighter one cleanly.

All of it is plain JavaScript with no libraries, which is why the whole page weighs less than most single photographs and runs comfortably on a mid-range phone.

What we are honest about

No converter is magic, and there are limits worth knowing before you upload:

• Bad source photos make bad templates. Very dark, blurry, low-resolution or visually cluttered images produce noisy results. Cropping tightly to your subject fixes this more often than any slider will.
• Some subjects are genuinely hard. Fine hair, chain-link fences, dense foliage and busy group shots fragment into tiny regions. Lowering detail and raising smoothing helps, but a photo of six people at a distance will never make a great paint-by-number.
• It is a tool, not a service. We do not print, ship, or sell kits, and we do not hand-retouch designs the way a paid studio might. What you generate is what you get — which is exactly why you can see it before deciding.

Who it is for

Mostly people making a present: a pet portrait for a partner, a wedding photo for an anniversary, a childhood house for a parent. Also teachers building classroom activities, hobbyists who want a custom canvas without the kit price, care homes and therapists using painting as a calm, structured activity, and small makers producing their own templates. If you own the photo, the output is yours to use — including commercially.

Get in touch

The fastest way to improve the tool is to hear where it fell short. If a photo converted badly, if a feature is missing, or if something is broken on your device, please tell us — bug reports and awkward edge-case photos are genuinely welcome.