Back to Tools

Research / Tools / Perception

Afterimage, a color-memory study

See a color for a few seconds. Then it's gone, and you rebuild it from black — three channels, no reference, just the afterimage in your head. Seven rounds, scored in perceptual color distance. Every game feeds a small study of how designers actually remember color.

Name (optional — shown only if you opt into the leaderboard)

player …

Exposure time

If logging is on, each game stores: a random player ID kept only in this browser (no device fingerprinting, no cookies, no account), the name you enter (if any), the colors shown, the colors you mixed, timing, and scores recomputed on the server. Nothing else.

Before you start: turn brightness up, and switch off Night Shift, True Tone and blue-light filters — they shift every color on screen, and your score with it. And if you have a color-vision difference, that will show up here too: the scores measure the match, not the effort.

Mode

Recall
running avg —
Memorize
4
Target
Your mix

Watch the stage

Hold the color in your head. The mixer unlocks when it disappears.

Red
Green
Blue

Every round starts at black. No eyedroppers here.

—/100

ΔE00 —

Shown
HSL complement
Your mix
—/100

Average across seven rounds, CIEDE2000-scored.

Leaderboard

Best seven-round average per player, per mode, for players who opted in. Scores are recomputed on the server from the submitted colors. Daily is the fair board: same seven colors, same exposure, for everyone that day.

Loading…

#PlayerBest avgGames

* First logged game in this mode, and it scored over 90.

Method & what gets logged

Scoring. Your mix and the target are converted from sRGB to CIELAB (D65) and compared with CIEDE2000, the CIE's perceptual color-difference formula. The round score is 100 − 2·ΔE00, clamped to 0–100. A ΔE of 1 is roughly the smallest difference a trained eye can see side by side; scores in the 90s mean you were within "most people couldn't tell" territory — from memory, starting at black. Scores are recomputed on the server from the colors submitted, so the leaderboard runs on verified numbers rather than the browser's say-so.

Specimen mode. Each scene is grown, not drawn: three independent binary cellular automata (a 9-neighbor majority rule, four generations from random seeds) are combined into an eight-region label map, smoothed twice, and cel-shaded. One organically-shaped region is circled; its fill is your target. No two scenes repeat.

Complement mode. The target is the shown color's hue rotated exactly 180° in HSL, saturation and lightness held. You're scored against that computed complement, not the color you saw. Note that an HSL hue-wheel opposite is not the same color as a perceptual (CIELAB) opposite; this mode tests the wheel definition, deliberately.

Daily mode & target sampling. The day's seven targets are generated from the date by a seeded generator, identical for every player, at a fixed 4-second exposure. In every mode, a game's targets are sampled at least ~12 ΔE00 apart in CIELAB, so no two rounds in a game test near-duplicate colors and hue difficulty is spread evenly rather than clustered by HSL geometry.

The mask. Between memorizing and mixing there is a short neutral-gray interval. Staring at a saturated color leaves a literal retinal afterimage; the mask lets it fade so the score measures memory, not the burn on your retina.

Logging. Logging is optional and off the leaderboard by default. Each logged game records: a random player ID generated in your browser and kept in its local storage (no device fingerprinting, no cookie, no account), the optional first name — displayed publicly only if you opt into the leaderboard — mode, exposure time, and for each round the shown color, the target, your mix, ΔE00, score, and how long you took to mix. The point is population patterns — which hues drift warm in memory, whether 2-second exposures decay differently than 6, how far off "the exact opposite" lands. Aggregates may show up later as a visual essay.