Portrait tutorial: making a face readable in six colours

    Portraits are the most requested cube mosaic and the most commonly abandoned. The failure is almost never the generator — it is the source photograph. Six sticker colours give you roughly four usable brightness steps, so a face survives only if its light and dark areas are already separated before quantisation. This tutorial walks the decisions in the order they actually matter.

    Step 1: choose a photograph with directional light

    Flat frontal flash is the worst input. It removes the shadow under the cheekbone, the shadow beside the nose and the shadow under the lip — the three shapes that make a face a face rather than an oval. A photo lit from roughly 45° to one side, with a visibly darker side of the face, will quantise into a readable portrait at half the resolution of a flat one.

    Reject a candidate photo if any of these are true:

    Step 2: crop tight, then tighter

    Every cube you spend on empty background is a cube not spent on the eyes. Crop to head and shoulders, with the eyes about a third of the way down the frame. On a 32-cube-wide mosaic, that crop typically gives the head around 24 cubes of width — roughly 72 stickers — which is enough for two distinct eyes, a nose shadow and a mouth.

    Crop to your final aspect ratio yourself rather than letting the tool decide. A portrait at 2:3 (for example 24 × 36 cubes, 864 cubes total) wastes far fewer cubes than a square crop of the same head size.

    Step 3: raise contrast and flatten the background

    In any photo editor, before uploading:

    1. Increase contrast until the shadow side of the face is clearly darker than the lit side, even if it looks slightly harsh on screen. Quantisation always reduces apparent contrast.
    2. Darken or lighten the background to a single flat tone that differs in brightness from the hair. A background that quantises to one colour also makes the build sheet dramatically faster, because hundreds of cubes share one identical face.
    3. Brighten the eye whites and the catchlight slightly. These are tiny areas that carry most of the recognition, and they are the first thing dithering destroys.
    4. Convert to greyscale, look, then undo. If the face is unreadable in grey, it will be unreadable in six colours — colour is not going to save it.

    Step 4: run a resolution ladder before buying anything

    Generate the same cropped photo at several widths and compare on screen at arm's length. This costs nothing and prevents the most expensive mistake in the hobby. Typical readability for a single tightly cropped face:

    Mosaic sizeCubesWhat reads
    10 × 10100Silhouette only. You can tell it is a head; you cannot tell whose.
    16 × 16256Hair mass and skin block read. Eyes are single dark smudges.
    25 × 25625Eyes, nose shadow and mouth separate. Recognisable to people who know the subject.
    32 × 321,024Recognisable to strangers from a photo. The practical sweet spot.
    48 × 482,304Expression, teeth, iris highlight. Installation-scale effort.

    These are for one face filling the crop. Add a second person and you need roughly the same cube width per face again, so a two-person portrait at the quality of a 32-wide single portrait needs about 64 cubes of width — four times the cubes, not twice.

    Step 5: decide on dithering deliberately

    Dithering scatters quantisation error into neighbouring tiles, which creates the illusion of intermediate tones — the closest this palette gets to skin. On portraits above about 25 cubes wide it usually helps noticeably. Below that, each tile is such a large part of the face that dithering reads as random speckle in the middle of a cheek.

    The other reason to turn it off: dithered mosaics have far fewer repeated cube faces, so the build sheet has more unique sequences and the build takes longer. If you are working to a deadline, an undithered mosaic at a slightly higher resolution is often the better trade. The mechanics are covered in detail in the colour guide.

    Step 6: check it the way viewers will

    Cube mosaics are designed for distance. Before committing, view the generated preview from three to four metres away, or shrink it on screen until it is about the size of a playing card. If the face resolves at that size, it will resolve on the wall. If it only works when you are nose-to-screen, increase resolution or improve the crop — do not start buying cubes.

    One last practical check: squint. Squinting approximates low-frequency vision surprisingly well and instantly reveals whether the eye sockets and mouth are separated or have merged into one dark mass.

    Common portrait failures and the fix

    Next steps

    When your resolution ladder gives you a size you are happy with, work out cube count and budget in the mosaic planner, order stock using the buying guide, and then generate the final build sheet.