Stream: green (sourced greens, foliage, meadows)
The user noticed that last night's paintings all share one palette and one
mood, while Friedrich "made many paintings that were very green." The
palette had no green tube, and growth grew only bare skeletons. This
stream adds the greens he is documented to have used, leaves on grown
trees and a ground-cover primitive for meadows, plus a study that uses them.
1. Research (notes/research/friedrich_materials.md §9)
The key new source is the Dresden technical study of 14 paintings, Mäder et al., METALLA Sonderheft 13 (2025), p.102 [MÄD]. I checked the quotes against the text myself:
- Given how sparse his palette was, his greens are an "üppige Vielfalt" (a lush variety).
- The true green pigments were "kupferhaltige Pigmente sowie Grüne Erde" (copper greens and green earth).
- He mixed greens from blue and yellow, "sehr häufig" (very often) with Prussian blue and also with smalt and cobalt. The yellows were yellow ochre, Naples yellow and chrome yellow.
- Green passages often stack "bis zu vier" (up to four) layers of different greens.
- Rinmann's green is rare: three paintings, c.1819–23.
- One painting (1808) has copper plus arsenic, a degraded copper-arsenite green. My inference: it must be Scheele's green, since Schweinfurt green dates from 1814.
- The London Winter Landscape grass has no green pigment at all: smalt, Naples yellow, bone black, ochre and possibly Prussian blue [NG p.56]. §4 had left out the smalt; I've fixed that.
Dates come from the NGA Artists' Pigments volumes and Field 1835. Viridian, chromium oxide and emerald green are too late or too doubtful for him. No green analyses were found for Hill and Ploughed Field, Meadows near Greifswald, Morning in the Riesengebirge or Chalk Cliffs; that is a gap, stated in §9.
2. Palette (crates/paint/src/palette.rs)
New tubes, with masstone, hiding, stiffness and strength as documented approximations (sources in the doc comments):
| tube | masstone | hiding | stiff | strength | basis |
|---|---|---|---|---|---|
| Prussian blue | #172440 | 0.35 | 0.45 | 3.0 | transparent, very strong [AP3 pp.196–197]; strength tempered (A) |
| green earth | #3a4843 | 0.2 | 0.35 | 0.3 | translucent, weak, little body [AP1 p.146; FIELD p.129]; Munsell 7.5G/2.9/1.5 converted |
| Rinmann's green | #5f8f76 | 0.35 | 0.5 | 0.4 | semi-transparent, weak (A for the numbers) |
| copper green | #3f7f6a | 0.25 | 0.4 | 1.0 | "poor hiding power in oil" [AP2 p.132]; numbers A |
API:
Palette::friedrich_early_greens() // early + Prussian blue + green earth
Palette::friedrich_1820_greens() // 1820 + Prussian blue + green earth + Rinmann's green
Palette::friedrich_1820().with(vec![Palette::copper_green()]) // a rare tube for one passage
Palette::green_tubes() // the two green-making tubes, to add anywhere
let st = Style { palette: Palette::friedrich_1820_greens(), ..Style::friedrich() };
let sky_pal = st.palette.only(&["lead white", "cobalt blue", "pale smalt", "yellow ochre", "red earth"]);
st.broad().mixed(&sky_pal, 0.45) // paint the sky from its own family
The base palettes are unchanged, on purpose. I first added the greens
to friedrich_early/friedrich_1820. With them, the aimed search picked
the very strong Prussian blue for the smalt/cobalt sky ramp. The miss
halved (0.049 → 0.023 OKLab), but recipe switches made seams at thin coats,
so palette::tests::smooth_targets_make_no_seams failed (0.028 against a
0.018 limit). It would also have changed every existing sky, and his skies
are smalt or cobalt [ALF; NPJ25]. So the greens live in separate palettes,
and existing paintings and the golden are unchanged. The golden passed
without being re-recorded.
Finding: for muted summer greens, the base palette already mixes well
(cobalt blue + chrome yellow + bone black, ΔE ≈ 0.003–0.006). The greens
palette adds green earth to those piles (for example "bone black 0.08 +
chrome yellow 0.17 + green earth 0.75"). Prussian blue seldom wins for
these targets (test green_tests::greens_are_mixed_closer). The aim code
is untouched (it belongs to fixes-paint).
3. Geometry (crates/paint/src/growth.rs)
Habit::leaf: Leafing: how a species carries leaves, as data. Presets:oak,birch,beech,alder,willow,spruce,none. Fields:years: wood this young carries leavesclump: clump radius / tree heightspacingsquash: flat beech sprays < 1 < hanging birch/willowdroopfill: how much sky shows through a clumpragged: lobed outlinebare: twigs with no leaves, the sky holestip: leaves crowd toward shoot endsinner/inner_w: short leafy shoots along older thin wood, so the crown's shell fills inspray: extra clumps around each live tip for the real twigs the model does not grow
- New habits:
Habit::beech()(distichous, level sprays, shade tolerant),alder()(a clear stem through a narrow oval crown) andwillow()(steep rising limbs, arching shoots, a leaning bole). Limb::age: the age in years at each point, so leaves go on young wood.Skeleton::foliage(sun, seed) -> Foliage(orfoliage_with(&Leafing, ..)):clumps: Vec<Clump>, where eachClumphasat,z,r,squash,tilt,fill,lit,shade,limbandmass(its main limb: the masses the eye groups)- live wood only: a dead oak's dead limbs stay bare
litis the clump's facing on the crown's envelope times the shadow cast by the clumps between it and the sun- methods:
mask(frame): leaves = 1, with sky holes; a leaf-sized noise breaks every partly covered place into leaves and skylit(frame): front clump over back, each slightly rounded; a clump covers what is behind it by itsfill(the same share as inmask), so an airy clump in front lets the light of the leaves behind show through instead of overwriting it (tested:transparent_foliage_keeps_the_light_behind)envelope(frame, reach)andgaps(frame, reach): the holes inside the crownback_to_front(),bounds(),grain()
Sward { horizon, near, height, spacing, thin, smallest, blades, fan, curl, flowers, kinds, patch, patch_size, wind: Wind }:grow(®ion_mask, seed) -> Vec<Tuft>scatters tufts with a minimum distance, so there are no rows. They crowd in lush patches, lean withWind { lean, gust, period, seed }and shrink toward the horizon.- Spacing grows as
scale^-thin, so there are fewer marks per area far away. Tufts shorter thansmallestare dropped: that's where the painter paints tone instead. - Each
Tufthasat,scale,height,lean,lush,blades(quadratic curves foot → control → tip) andflower: Option<Flower { at, r, kind }>. Tufts come far first.
- Tests:
foliage_on_live_young_woodcovers every habit. It checks that clumps are only on live wood, that the sunward side is lit more, that sky shows through the crown (2–60%), that lit ≤ mask and that winter and dead wood are bare.sward_recedeschecks far-first order, that far tufts are shorter and sparser, the mean lean against the wind, flowers present and no marks belowsmallest.
let oak = Habit::oak().grow((690.0, 570.0), 360.0, seed);
let leaves = oak.foliage((-0.55, -0.75, 0.35), seed); // toward the sun
let mask = leaves.mask(c.frame());
let light = leaves.lit(c.frame());
for cl in leaves.back_to_front() { /* touches at cl.at, sized by cl.r, colored by cl.lit */ }
let hedge = Habit { years: 15, ..Habit::alder() }.grow(..)
.foliage_with(&Leafing { clump: 0.05, spray: 0.5, ..Leafing::alder() }, sun, seed);
let tufts = Sward { horizon, near: h, height: 26.0, wind: Wind { lean: 0.12, gust: 0.18, period: 140.0, seed }, ..Sward::default() }
.grow(&meadow_mask, seed);
4. Study (paintings/src/bin/study_green.rs)
A summer meadow in daylight, 1.4:1: a leafy oak, a hedgerow of young alders and beeches, far fields, a clear sky. The study's own habits:
- The sky is painted from a sky family of tubes and fused.
- Far woods are stippled.
- The meadow is laid in lean, with the oak's shadow and soft cloud shadows glazed over it.
- Leaf masses are laid in with short hatched strokes cut to
Foliage::maskand colored bylit, then broken with dark touches, mid touches on the sunward side of lit clumps and a few crisp yellow-green touches wherelit > 0.45. - Grass goes last, as fine upturning rigger strokes, far first [NG p.56], with a sheen field that follows the gusts; a flower is one touch.
--geom shows the geometry flat (diagnostic only).
Evidence (in [local path redacted]):
p1.jpg: the first painted pass. The hedge was lollipops on stems, the clouds cotton spots, the far woods a straight strip.geom.jpg(whole) andgeom4.jpg/geom1.jpg(the oak): geometry. Clumps only at limb ends first ("lollipops"), then withinnerandspray, a filled crown shell with sky holes.p3.png/p3.jpg: the 1000px study as committed (alsoout/study_green.png).oak3200.png/.jpgandoak3200_detail.jpg:--full --crop 540,170,960,600at 3200px, 3m22s on a busy machine. The 1000px render takes about 115 s from scratch; the hedge (47 shrubs, 21k clumps) is the slowest stage at about 41 s.
Honest judgment
At 1000px it reads as a green summer landscape: meadow, hedgerow, a tree in full leaf, cloud shadows. It is a clear step away from last night's twilight palette. It does not yet read as a Friedrich summer passage.
- The oak reads as a savanna tree. This skeleton (oak habit, 24 years) is a flat umbrella on long bare limbs. The model grows an open young oak, not a mature, rounded, dense-crowned one.
- At 3200px the leaf masses are cut-outs. The hatched lay-in is clipped to the mask, so each mass is flat dark green with a crisp digital edge. The light touches are uniform round dabs, like polka dots. This is closer to a video-game tree than to Friedrich's foliage, whose drawings use "small strokes and hooks" and "jagged lines" [KH].
- The meadow works best. It has varied texture and upturning blades at the front, but the far half is a smooth lay-in, and the cloud shadows are a bit blotchy.
Next
- Foliage painting:
- Leaf-shaped marks at the silhouette (hooked, jagged strokes following
tilt/droop) instead of clipping. - Layered greens (MÄD's "up to four layers"): a warm mid layer under
the dark, with sky scumbled back into
gaps. - Light touches varied in size and shape, oriented along the twig.
- Leaf-shaped marks at the silhouette (hooked, jagged strokes following
- Growth: a mature oak habit (more years with less self-thinning, and reiteration from dormant buds), so crowns are rounder and denser. Pollard willows are worth doing for Greifswald meadows.
- Performance: the hedge's 21k clumps cost ~40 s at 1000px. Paint far
hedges with a coarser
Leafing(bigger clumps,spray0) or a stipple overmask. - Easel bindings:
Skeleton::foliage,Foliage::{mask, lit, gaps},Sward::growand the greens palettes.