What the painters were given

Pointed tips: round sable and rigger

notes/tip.md · 249 lines · Notes

Given to 3 painters:

Pointed tips: round sable and rigger

Branch tip. This addresses amnesia2 friction item 8: small marks became beads, blobs, pegs, "hearts" and ladders at 3200px (mountains M12, coast C14, winter W9).

What was wrong

I measured three causes in crates/paint/src/bristle.rs.

  1. Dashes and ladders. A hair touched the canvas only where the surface reached th = 1 - reach * 1.6. A lightly pressed hair has a small reach, so it touched only the weave's peaks. Every hairline, and the tapering end of every flick, became dry-brush dots at the weave pitch. At 1000px the weave is finer than a pixel, so the dots only appeared at 3200px.
  2. Pegs. The smallest footprint was still 0.45 of the tool's width, and the center hairs had ragged thresholds. A flick was a bar at full width that broke into dots where the pressure fell.
  3. Resolution. Each hair's track had a 0.55px minimum radius. Its paint spread as a uniform thin film over whole pixels, and Kubelka–Munk (KM) makes a thin film of opaque paint darker than the same paint covering part of a pixel. Small marks therefore came out about twice as heavy at 1000px as at 3200px. This was true for every tool (see the numbers below). The plough also dropped paint on whole pixels, which gave fine tracks a dotted ridge along one side.

The model

Tool::point (0..1) says how finely the hairs come to a point: 0 is a blunt tuft and 1 is a fine point. Since Round 7 every preset has point: 0.0, round_sable and rigger included: the pointed tip is opt-in (see "Round 7: opt-in" below). Everything below applies only when point > 0. With point = 0 the old code runs unchanged: the old golden matched bit for bit with both presets set to 0.

API

// a pointed tool: the point is opt-in (Round 7); Lua brush{kind="round", width=1.6, point=1}
let mut b = Held::new(Tool { point: 1.0, ..Tool::round_sable(1.6) }, seed);
b.load(Paint::body(dark), 0.8);

// a flick: pressed on the belly, lifted off to a hairline point
c.drag(&mut b, &Gesture::new(vec![root, mid, tip]).pressure(0.75, 0.0).ramps(0.1, 0.75), None);

// a hairline: the point only, light pressure
c.drag(&mut rg, &Gesture::line(a, e).pressure(0.3, 0.3).ramps(0.05, 0.1), None);

// think in widths: how wide at this pressure, what pressure for this width
let w = tool.mark_width(0.4);          // units
let p = tool.pressure_for(0.5);        // pressure for a 0.5-unit mark
// a custom tool with a blunter point
let t = Tool { point: 0.5, ..Tool::round_sable(3.0) };

mark_width is the geometric contact width, and the measured ink width comes within about ±30% of it (the table below). Touch with a pointed tool also gathers toward the point: a light touch leaves a small dot and a hard press leaves the belly.

Evidence

The study is cargo paint study_tip. It paints birds (V-flicks with a body touch), a brig's rigging, a signature-like line, grass flicked up, bare twigs and spruce tips on a Friedrich ground. For 3200px crops use --full --crop 20,20,330,220 (birds), 360,50,640,300 (rigging), 30,480,330,650 (grass), 680,360,980,650 (spruce) and 680,120,980,280 (signature). Before-and-after images are in notes/tip/:

Measured by cargo test --release -p paint probe_ink_width -- --ignored --nocapture. Ink width is the darkening summed across a straight mark, in units:

tool, pressure before: 1000px / 3200px after: 1000px / 3200px
rigger 0.5, p 0.4 1.22 / 0.71 0.30 / 0.30
sable 1.6, p 0.4 1.85 / 0.99 0.57 / 0.57
sable 1.6, p 0.7 2.35 / 1.51 0.96 / 0.94
sable 3, p 0.9 4.45 / 2.77 2.39 / 2.32

Tests are in bristle::tip_tests: resolution independence (within 20%, 500px against 1600px), width follows pressure and a flick tapers, a light hairline on linen at full size has no gaps, and feeding conserves paint. With the point switched off, the hairline test finds 13 gaps and the resolution test fails (1.50 against 1.01).

Changes to existing output

Round 3 review fixes (branch fix3-physics)

Round 7: opt-in (branch r7-d)

The winter A/B (notes/round7/winter_ab.md) showed the default thinned every tapered mark of a painter who had asked for a width: the winter oak's dark share fell from 10.2% to 6.1% at 1000 and the firs' from 16.4% to 10.7%. A painter who writes round_sable(1.6) or brush("round", 1.6) means a mark about 1.6 wide, and at ordinary pressures the cone laid a half to a third of that. Alice chose the variant without it ("we want d"). So round_sable, rigger, Style::detail and Style::line_tool are blunt again (point: 0.0): the pre-tip code path, which lays the width asked for. The pointed model is unchanged and one field away: Tool { point: 1.0, ..Tool::round_sable(w) }, Lua brush{kind="round", width=w, point=1}. Logs that already set point= behave as before.

Open issues

Source