A paint by numbers generator does one job: it throws away almost all the color information in your photograph and keeps just enough that the result still looks like the subject. How well it chooses what to throw away is the entire difference between a template you can paint and one you cannot.
The five steps, in order
1. Resize. The image is scaled to the physical canvas at print resolution. Detail smaller than a brush tip cannot survive this and should not.
2. Color quantization. The image is reduced from millions of colors to a fixed palette, usually somewhere between 20 and 40. This is the step that decides how the finished painting will look. Smooth gradients become bands of flat color.
3. Region merging. Adjacent pixels of the same palette color become one region, and regions below a minimum size are absorbed into their neighbors. This is the step free tools most often skip or do badly, and it is the one that determines whether the template is paintable.
4. Outlining. Each region gets a thin boundary line, faint enough to disappear under paint.
5. Number placement. A number is positioned inside each region, at a size that fits. Regions too small for a legible number have to be merged away at step three, which is why steps three and five are really the same problem.
Why free generators disappoint
They almost all stop after quantization and outlining, and skip serious region merging. What comes out looks convincing on screen at full zoom and falls apart in practice.
- Thousands of tiny regions. Speckle across every area of gradual tone. Technically correct, physically unpaintable.
- Numbers that do not fit. Digits printed outside their region, or overlapping the boundary line, so you cannot tell which area they belong to.
- A palette you cannot buy. The generator outputs 32 specific colors. You then have to mix all 32 yourself and match them accurately, which is harder than painting the picture freehand would have been.
- No printing. You get a digital file. Getting it onto primed canvas at the right size, with lines faint enough to cover, is a separate problem and not a small one.
Our reviews of the specific tools go into which ones handle which step: the free options, the paid software, and PBNify.
What to feed it
The output is limited by the input more than by the software.
- High resolution. Detail that is not in the file cannot appear on the canvas. Original camera files beat anything saved off social media.
- Directional light. Side light gives a lit side and a shadow side, which is what the quantization has to work with. Flat light gives it nothing.
- Not backlit. A subject with the sun behind it reduces to a dark silhouette. This is the single most common cause of a poor result.
- A clear subject and a simple background. Busy backgrounds turn into a field of unrelated small shapes.
- Cropped to the canvas shape first. Otherwise the process crops for you, and it will not choose what you would have.
What a generator cannot give you
The physical half. Primed canvas printed at the right size with lines that vanish under paint, and a set of paints actually mixed to match the palette on that specific template. Those are the parts that make the difference between an interesting exercise and a painting you finish.
That is what a custom canvas made from your photo is: the same conversion, with the region merging judged properly, printed in the United States on pre-primed cotton at 12x16, 16x20 or 18x24, and shipped with paints mixed to the palette it generated.
If you want to understand what the outline stage is really doing, our note on where the sketch fits in covers it from the drawing side.
