How six cube colours become thousands of shades
A Rubik's Cube offers exactly six sticker colours. A photograph contains millions. Every mosaic is therefore an act of controlled lying: the generator throws away almost all the colour information in your image and relies on your eye to reconstruct it from a distance. Understanding how that works is the difference between a mosaic that reads instantly and one that looks like coloured noise.
The palette you actually have
Highlights, skin highlights, sky, paper
Warm mid-tones, sunlight, blond hair, sand
Skin tones, sunsets, brick, wood
Deep warm shadows, lips, brand accents
Foliage, cool mid-tones, muted shadow
Sky, water, dark shadows, near-black
Notice what is missing: black, brown, grey, pink, purple and any pastel. Notice also the luminance spread. Ranked from bright to dark, the six colours run white, yellow, orange, green, red, blue — with white and yellow clustered at the top and nothing genuinely dark except blue. This lopsided distribution is why cube mosaics tend to look washed out unless you deliberately increase contrast in the source image.
Why RGB distance produces bad matches
The naive approach is to treat each colour as a point in RGB space and pick the nearest one by straight-line distance. It fails badly, because RGB distance does not match human perception. A mid-grey is nearly equidistant from all six colours in RGB, so tiny changes in the source pixel flip it between orange and green — producing speckle in what should be a smooth area. Worse, warm skin tones frequently round to green, which is instantly visible as a mistake on a portrait.
The generator instead converts pixels to CIE L*a*b*, a space designed so that equal numeric distances correspond to roughly equal perceived differences, and measures with the CIEDE2000formula. That formula weights lightness, chroma and hue separately and applies corrections in the blue-purple region where earlier formulas were least accurate. In practice it means shadows go blue, highlights go white, and skin stays warm.
Dithering: trading spatial resolution for colour resolution
Even with perfect matching, a six-colour image bands heavily — smooth gradients turn into hard stripes. Dithering fixes this by refusing to discard the rounding error. When a pixel is forced from its true colour to the nearest cube colour, the difference is pushed into neighbouring pixels that have not been processed yet. Floyd–Steinberg distributes it in fixed fractions: 7/16 to the pixel on the right, 3/16 down-left, 5/16 directly below and 1/16 down-right.
The visible result is that a region which should be pink becomes a fine scatter of red and white stickers. Stand two metres back and your visual system averages them into pink. This is exactly how newspapers printed photographs with a single ink colour for a century.
The trade-off is real: dithering converts colour error into spatial noise. On a small mosaic — under 20 cubes wide — there are not enough stickers for the eye to average, and dithering just looks dirty. On logos and flat illustration it destroys clean edges. Rules of thumb:
- Turn dithering on for photographs, portraits, and anything with gradients, at 25 cubes wide or larger.
- Turn dithering off for logos, text, flat vector art, and any mosaic under 20 cubes wide.
Blend recipes
These are the mixes dithering will produce automatically, but knowing them helps you predict whether an image will work before you buy cubes.
| Target | Sticker mix | Notes |
|---|---|---|
| Skin tone | Orange + white, ~2:1 | Add a little yellow for warmer light, red for shadow sides. |
| Brown | Red + green, ~1:1 | The most reliable brown available in six colours. Works for hair and wood. |
| Pink | Red + white, ~1:2 | Lighter ratios read as blush; heavier red reads as salmon. |
| Grey | White + blue, ~3:1 | Cool grey. Use white + green for a warmer, slightly olive grey. |
| Teal | Blue + green, ~1:1 | Convincing at viewing distances above two metres. |
| Purple | Blue + red, ~2:1 | The weakest blend. Avoid purple as a subject's dominant colour. |
| Near-black | Solid blue | There is no black sticker. Blue is the darkest of the six in luminance. |
Preparing an image before you upload it
- Crop tight. Empty background wastes cubes. A face should fill most of the frame.
- Raise contrast. Push highlights brighter and shadows darker than looks natural on screen. Six colours compress tonal range, so you need to start wider.
- Simplify the background. A single flat colour behind the subject reads far better than a busy room.
- Check the luminance histogram. If the image is entirely mid-tone, the mosaic will be mush regardless of settings.
- Try monochrome. High-contrast black-and-white sources map to white / green / blue and are often the most recognisable result of all.
The high-fidelity option
The generator's high-fidelity mode changes the weighting so that luminance is preserved more aggressively than hue. A pixel that is the right brightness but the wrong colour reads better at distance than the reverse, because human vision resolves brightness detail at far higher spatial frequency than colour detail. Use it on portraits and any image where the subject is defined by light and shade rather than by colour. Leave it off for brand colours, where hue accuracy is the whole point.
Ready to see the theory in practice? Try it on the generator, compare finished results in the gallery, or plan cube counts and cost with the mosaic planner.