Case study: a 16 × 24 Mona Lisa cube mosaic

Why this subject is difficult

Cube mosaics carry no tonal gradient inside a single sticker. Each 3 × 3 face contributes nine flat colour samples and nothing in between, so the illusion of a curved cheek has to be built out of neighbouring stickers that disagree with each other. The original painting's transitions span dozens of intermediate shades over a few centimetres of canvas; at 16 cubes wide, that same cheek has roughly 48 stickers to work with across the whole width of the face.

The backdrop is the second problem. The painting's browns and olive greens land almost exactly between the cube's green, orange and blue. Without dithering the quantiser picks one winner and the whole background flattens into a single slab, which destroys the depth that makes the portrait read as a portrait.

Sizing decision: why 16 × 24 and not 12 × 18

The first draft was 12 × 18 (216 cubes). At that resolution the eyes occupy four stickers each and the mouth two, and the face stopped being recognisable from across a room. Moving up one step to 16 × 24 raised the sticker grid to 48 × 72 and put roughly nine stickers into each eye — enough for a pupil, an iris edge and a highlight.

The step after that, 20 × 30 (600 cubes), was rejected on cost rather than quality. It looked better, but it added 216 cubes and about six hours of build time for a gain that was only visible from under a metre away. For a portrait hung at eye level in a hallway, 16 × 24 was the point where extra resolution stopped paying for itself.

Palette handling

Skin tone is the entire job. In the six-colour palette the warm range is orange, yellow and white, and the ratio between them carries the shading: predominantly orange for shadowed jaw and neck, orange with scattered yellow for the mid-tones on the cheek, and yellow with white for the lit forehead and the bridge of the nose.

Floyd–Steinberg dithering was left on for this build. Because the error from each sticker is pushed into its neighbours, the cheek becomes a stippled mix of orange and yellow that the eye averages into a single warm tone at viewing distance — the same trick a newspaper halftone uses. Turning dithering off produced visible banding across the forehead in the preview, which is the fastest way to tell a portrait needs it.

The backdrop was handled by accepting blue as a stand-in for the darkest browns. It is not accurate, but it is consistently darker than every skin sticker, and consistency of value matters far more than hue accuracy at this scale.

Build sequence and what went wrong

The wall was split into six panels of 8 × 8 cubes and each panel finished completely before the next was started. Two mistakes cost real time. First, a full row was placed one cell to the left near the shoulder, which was only noticed two rows later — laying finished cubes into a tray gridded like the mosaic prevented it happening again. Second, several original stickers stretched during peeling on the first fifty cubes because they were pulled straight up rather than rolled off from a corner.

Ordering 5% spare cubes covered both problems. Of 384 cubes, eleven ended up with a stretched or torn sticker that would have been visible on the finished wall.

Finished result

At about 27 kg the panel set hangs on a plywood cell grid fixed into studs, not on picture hooks. Viewed from three metres the portrait reads immediately; from one metre it dissolves into coloured squares, which is the effect that makes cube mosaics worth building in the first place.