A dark theme for Omarchy. Solitude's geometry, measured and regularised, with one brass voice.
A vernier is the second scale on a caliper, the one that reads finer than the eye can. Nothing here was picked by eye either.
omarchy theme install https://github.com/thisisgm/omarchy-vernier-theme.gitVernier is not inspired by Solitude, it is derived from it. Every constant in
colors.toml was measured off /usr/share/omarchy/themes/solitude/colors.toml
rather than chosen:
- the surface hue, OKLab 236.68, is the chroma-weighted mean hue of Solitude's twelve-colour slate ladder;
- the chroma each surface and ink carries comes from a least-squares fit of
Solitude's own chroma against lightness,
C(L) = 0.0660 * L^1.286 * (1-L)^1.117, residual rms 0.00425. The ring does not use it: five of the six base ring colours are solved at one chroma, red carries more by design, and the accent has its own, all stated below; - the accent hue, 70.9, is the chroma-weighted mean hue of Solitude's two
warm inks,
#cbc2beand#c9c2b4. Solitude has a warm direction already; it just never spends it on anything; darker_backgroundis byte-identical to Solitude's.
Solitude is a beautiful theme with six measurable faults. Vernier keeps the shape
and fixes them. The Accent against its nearest ring colour row is not one of
them: it is a constraint Solitude never had and Vernier adds.
| Solitude | Vernier | |
|---|---|---|
| Surface ladder, OKLab steps | 0.019, 0.022, 0.051 | even by construction at 0.0307; 8-bit quantisation lands them at 0.0320, 0.0309 and 0.0303 |
| Emphasis ink vs body ink | bright_foreground sits 0.098 darker than foreground |
ordered, emphasis is the lightest rung |
| Ring colours clearing WCAG AA on all four surfaces | 6 of 12 | 12 of 12 |
| Tightest pair meaning different things, worst case over normal, protan, deutan and tritan vision. A colour and its own bright variant are one meaning, so that pair is excluded | 0.0222, red against bright_blue |
0.0469 |
| A colour and its bright variant sharing one hex | cyan and bright_cyan are both #707070 |
all twelve distinct |
| Accent against its nearest ring colour | not a constraint | 0.0534 |
A diff's + against its - |
inverted, bright_green sits 0.290 darker than bright_red |
+ leads by 0.0685 for every viewer |
Foreground over background measures 11.06:1.
- One material. Every surface and every ink sits on the same hue and carries the chroma its lightness earns. Paper and ink are the same stone.
- One voice. Brass
#b38956is the only colour allowed to be loud, and it means exactly one thing: this is what has focus. sRGB tops out at C 0.142 on this hue at this lightness, where the blue channel is down to 1 of 255 and the colour has gone to mustard. It is held at C 0.085, where blue still reads 86, which is what keeps it brass. - Everything clears AA. All twelve ring colours clear 4.5:1 against all four surfaces. That requirement is what costs the ring its lightness range, and it is why the ring is solved rather than picked.
Solitude's ring is deliberately near-neutral, and measurement says a fully neutral ring cannot carry twelve meanings. Clearing AA against the lightest surface puts a floor at OKLab lightness 0.616, and the twelve end up between 0.659 and 0.885, a span of 0.23 to hold twelve meanings. Under dichromacy the red-green axis is gone, so hue can only help along blue-yellow.
Chroma is the other lever, and it is spent sparingly. Five of the six base
colours are solved at C 0.105, which is where a chroma sweep stopped
buying separation, and 8-bit quantisation lands them at 0.105 or 0.106; red
sits at 0.126, the 20 percent exemption below; and the brights land between
0.086 and 0.107 once eased and gamut-clipped.
The accent is solved as a thirteenth meaning, not left out of it. Focus is a
distinct thing from an error or a success, so #b38956 sits in the same
separation cost as the twelve. Leaving it out of that cost is not hypothetical:
an earlier build did exactly that and put the accent 0.0239 from red,
tighter than anything inside the ring. Constrained, it sits at 0.0534.
Only the six base ring colours are solved that way. Each bright variant is
derived from its base at a lightness step solved per family, 0.108 to 0.119,
with chroma eased 5 percent and then clipped to the sRGB gamut. The clip is what
it sounds like: bright_red and bright_blue sit on the gamut boundary and lose
15 and 18 percent of their base chroma rather than 5. Every family still
separates by lightness, which is the property the step exists for.
Solving all twelve independently scored better on paper and produced green and
bright_green at the same hex, which is the exact defect this theme exists to
fix.
Red is the one exemption: it carries 20 percent more chroma, because an error that fails to alarm has failed.
| name | hex | OKLab L |
|---|---|---|
darker_background |
#080a0b |
0.143 |
dark_background |
#0e1112 |
0.175 |
background |
#14181a |
0.206 |
lighter_background |
#1b1f21 |
0.236 |
One more colour sits on the same hue without a colors.toml key, because Omarchy
has no key for it. #626a6f at lightness 0.519 is the neutral border
shell.hyprland.toml gives to menus, tooltips and popups. It is not a step on
the ladder above; it is simply the darkest colour on that hue which still clears
3:1 against the raised card it outlines, measured 3.02:1.
| name | hex | OKLab L | worst contrast on any surface |
|---|---|---|---|
bright_foreground |
#dde0e2 |
0.905 | 12.52:1 |
foreground |
#c8ccd0 |
0.843 | 10.28:1 |
light_foreground |
#b3b9bd |
0.782 | 8.38:1 |
dark_foreground |
#9fa6ab |
0.721 | 6.73:1 |
muted |
#868d93 |
0.640 | 4.94:1 |
| name | hex | OKLab L | C | hue | worst contrast on any surface |
|---|---|---|---|---|---|
red |
#e37d7d |
0.702 | 0.126 | 20.8 | 5.93:1 |
yellow |
#b8ae5c |
0.742 | 0.105 | 102.4 | 7.31:1 |
green |
#82c68e |
0.766 | 0.106 | 149.0 | 8.24:1 |
cyan |
#27a6a2 |
0.659 | 0.105 | 191.7 | 5.58:1 |
blue |
#80a4e6 |
0.718 | 0.105 | 262.1 | 6.62:1 |
magenta |
#ca8bc2 |
0.717 | 0.105 | 331.1 | 6.31:1 |
bright_red |
#ffa5a3 |
0.811 | 0.107 | 21.1 | 8.83:1 |
bright_yellow |
#dbd183 |
0.851 | 0.099 | 101.9 | 10.64:1 |
bright_green |
#abecb6 |
0.885 | 0.099 | 149.2 | 12.18:1 |
bright_cyan |
#5ecbc6 |
0.777 | 0.100 | 191.2 | 8.57:1 |
bright_blue |
#a8c7ff |
0.827 | 0.086 | 262.1 | 9.71:1 |
bright_magenta |
#ecaee3 |
0.825 | 0.100 | 331.5 | 9.28:1 |
Five wallpapers, one layout, five ramps, all rendered natively at 6016 x 3384
(16:9 6K). The layout is a fanned stack of rounded swatches, measured off a
reference render rather than eyeballed: card width 0.125 of the frame, step
0.0643, so cards overlap 49 percent; card height 0.7122; corner radius half the
card width, which makes each card a stadium. Ten cards, except 4-palette, which
carries twelve because the ring has twelve colours.
Flat fills, no shadow, no gradient, and no dither, because a frame of flat fills has no gradient to band. Shapes are composited by coverage in linear light, so an edge never picks up the dark fringe that blending encoded sRGB produces.
The flat-fill claim you can confirm on the shipped PNGs without trusting me.
Every master is 6016 x 3384, and its eleven most common colours account for at
least 99.79 percent of its pixels. 4-palette carries twelve cards plus a
ground, so for that one it is the top thirteen. What is left is the one-pixel
anti-aliased seam along every shape edge, against the ground or against whichever
card lies beneath.
This needs a colour histogram rather than a count of distinct colours. A distinct count returns between 859 and 3985 values depending on the master, because those seams take intermediate values the flat fills never use, and a reader who ran that would think the claim had failed. Pass the file and its fill count:
import numpy as np
from PIL import Image
def flat_share(path, fills):
a = np.asarray(Image.open(path).convert("RGB")).reshape(-1, 3)
_, counts = np.unique(a, axis=0, return_counts=True)
return round(np.sort(counts)[-fills:].sum() / len(a), 5)
print(flat_share("backgrounds/1-surfaces.png", 11)) # 0.99833
print(flat_share("backgrounds/4-palette.png", 13)) # 0.99798That each master is also byte-identical to what its source renders is not something this repo lets you check, because the renderer is not published. Take that one as an author's claim.
The paper ladder. Ten stops from the canvas surface up to comment grey, on the deepest surface. The quietest of the five. Mean relative luminance 0.0565.
The ink ladder. Comment grey up to emphasis ink, a close-valued study where the whole ramp spans 0.265 in lightness. Mean relative luminance 0.2456.
Brass. Ten stops on the accent hue, each holding the same fraction of the sRGB gamut, so saturation reads constant while lightness does the work. Mean relative luminance 0.1548.
The ring. All twelve ANSI colours in hue order, each family's bright leading its base, which doubles as a legibility check on the set. Mean relative luminance 0.2765.
The same surface ladder, dark leading, on content ink. The only light ground in the set, and the brightest of the five. Mean relative luminance 0.3519.
Terminal colours render at Omarchy's stock window opacity, 0.985 focused and
0.96 unfocused, so on-screen values sit slightly off the literal hex. That is a
compositor setting shared by every Omarchy theme, not a property of this one.
Omarchy applies it as a window rule rather than a decoration variable, and
~/.config/hypr/hyprland.lua requires only monitors, input, bindings,
looknfeel and autostart, so a windows.lua of your own is never read. Append
the rule to ~/.config/hypr/looknfeel.lua, which is loaded, then hyprctl reload:
o.window({ tag = "default-opacity" }, { opacity = "1.0 1.0" })Vernier is not affiliated with or endorsed by the Solitude theme or its author. It is a derived work in the sense that its constants were measured from Solitude's published palette.
If this saved you an afternoon, you can buy me a coffee.
MIT. See LICENSE.