Log inGet started
World

Style Cookbook: Mix Your Own Art Style from Stackable Look Blocks

README.md

pencilBlocks

Give your game a hand-drawn art style while it runs and the camera moves, each in one call: a pencil sketch look, a charcoal drawing style (a charcoal study), colored pencil over watercolor (coloured pencil hatched over a watercolor wash), a conte drawing on tan paper, a ballpoint notebook sketch, a margin doodle, a bold ink sketch, a technical sketch, a blueprint or a soft storybook pencil. Each style has its own how-to guide in the kit: "How to make my game look like a pencil sketch", "Charcoal drawing look", "Coloured pencil over watercolor" and "Ballpoint notebook doodle look", plus an example game built from an empty world. Under the hood, the scene is read as a hand drawing: six drawing media (graphite, coloured pencil, charcoal, conte crayon, ballpoint and notebook doodle) in twenty-five named variants, each as a shade block whose strokes build up layer by layer as the picture darkens and a line block for its edges, plus a technical pen line and ruled technical hatching, three Fill blocks to start a drawing from (a pale sketch blank that keeps as much of the original colour as you choose, a light coloured pencil underlayer and a soft storybook underlayer), twelve whole looks you turn on with one call, a hero material that lets a moving character carry its own marks, guides and a workflow. Every block is a styleKit block, so it stacks, swaps and masks with every other block: pencil over a watercolor wash, an ink outline over pencil shading, any drawing on any paperBlocks paper.

Version 1.4.8. New in 1.3.0: coloured pencil by object, so the hero, the enemies and the pickups read apart at a squint under one drawing: an object's style channel value draws it in its own colour, its line weight sets its line width and how densely it is shaded. The version history at the end of this README lists every earlier change.

Install and use

Install pencilBlocks from ZeroMind (search "pencilBlocks" or "pencil sketch look"). An install brings what the package's own files use, and nothing needs installing by hand: styleKit (the look stack every block runs in, and its shared skStrokes stroke module), paperBlocks (the papers every look is drawn on), watercolorBlocks (the washes under the coloured pencil looks), inkBlocks (the edge finding the pencil lines share, and the ink lines that swap for them) and lookbookKit (the package's measurements, which brings starterBlocks with it). In a brand-new world the whole install took about 5 seconds. Installs with several dependencies can be slow (#21420): give the install time to finish and check /zero/source for the pencilBlocks.package folder.

Then, from the console or any script:

local StyleKit = require("~styleKit.StyleKit")

StyleKit.applyLook("pencilSketch")              -- HB pencil hatching on warm cold-press paper: one change
StyleKit.swap("shade", "charcoalVineShade")     -- the same drawing in vine charcoal: one change
StyleKit.swap("shade", "graphiteHBShade")       -- back: the picture from before, tuning included
StyleKit.swap("line", "inkPenLine")             -- an ink outline over the pencil shading
StyleKit.swap("paper", "paperKraft")            -- the drawing on brown kraft paper
StyleKit.clear()                                -- the frame exactly as rendered again

The look lives on the scene's synced StyleStack, so every player in the world sees the same drawing, and it is saved with the scene. Swapping back gives the frame the first block would have drawn at that moment: with the stack's clock held (StyleKit.setClock(function() return 1000 end)) the two are identical pixel for pixel; with the boil running, the strokes are redrawn 8 times a second either way.

How a drawing is made

A pencil look is three entries, each in its own slot:

EntrySlotWhat it decides
shadeToneThe medium and how it shades: one of the <variant>Shade blocks.
lineLineThe edges: one of the <variant>Line blocks, or any ink or styleKit line.
paperGroundThe sheet: any paperBlocks paper.
fill (optional)FillWhat the drawing starts from: a pale sketch blank, a coloured pencil underlayer, or a watercolorBlocks wash. The shade's over lets it show through.

The shade draws each place by the light falling on it (and localTone of its material's own darkness), from bare paper at lightest to the darkest tone at darkest. Darker places get more strokes laid on top of the lighter places' strokes, layer by layer up to layers, each stroke coming in at its own threshold, so a slow change of light adds or removes strokes one at a time instead of switching patterns. Strokes come in patches of parallel strokes laid on the surfaces in the world, so they stay put while the camera pans, orbits or zooms, and they keep their size on screen at any distance (two octaves of the pattern blended by distance: walking closer, strokes grow into the next octave instead of sliding). They catch on the paper's tooth and skip its valleys, harder leads more so. Textures such as brick and planks cancel out of the light, so they do not print as strokes.

Charcoal and the soft leads lay much of their tone as masses instead (the shade's mass setting, new in 1.1.0): soft, smoky tone rubbed into the sheet, uneven like a finger's smear and sitting on the paper's tooth peaks, its darkest parts turning to the medium's deep colour. Strokes show fully only at the edges of the masses (where tone or form changes, or at a silhouette) and only faintly inside them, so a charcoal drawing reads as smudged masses with a few strokes at the edges, as a real one does, and soft graphite keeps its directional strokes over a rubbed tone. Every layer of strokes at one place shares one patch grid and one slant, and darker layers cross the first at clear angles, so strokes never beat against each other into broad zebra bands. The strokes are sized on screen across their own width, so their spacing is the same number of pixels whichever way they run and however steeply a floor or wall is seen.

Every shade block runs one shader (pencilShade) and every pencil line another (pencilLine); each variant is the same shader with its own defaults. So every shade takes the same settings, and StyleKit.swap("shade", id, { keepSettings = true }) always fits.

The media

Each variant is two blocks: <variant>Shade (Tone) and <variant>Line (Line).

MediumVariant idShadeLineWhat it is
Graphitegraphite2Hgraphite2HShadegraphite2HLineA hard lead: pale, thin, crisp strokes at one even slant.
GraphitegraphiteHBgraphiteHBShadegraphiteHBLineAn everyday lead: even grey strokes following the form, the plain pencil sketch.
Graphitegraphite4Bgraphite4BShadegraphite4BLineA soft lead: dark, broad strokes crossing in the shadows over a rubbed tone (some of it laid as masses), a soft shine where heavy.
Graphitegraphite6Bgraphite6BShadegraphite6BLineA very soft lead: a dark tonal study, much of it laid as velvety rubbed masses, with directional strokes over them.
GraphitegraphiteMechanicalgraphiteMechanicalShadegraphiteMechanicalLineA fine mechanical pencil: even, thin, close parallel strokes.
Coloured pencilcoloredPencilWaxycoloredPencilWaxyShadecoloredPencilWaxyLineSoft waxy pencils in each thing's colour, filling the tooth, indigo shadows.
Coloured pencilcoloredPencilHardcoloredPencilHardShadecoloredPencilHardLineHard pencils in many thin crossing layers, the paper sparkling through.
Coloured pencilcoloredPencilBurnishedcoloredPencilBurnishedShadecoloredPencilBurnishedLineLayers rubbed into a smooth, saturated finish.
Coloured pencilcoloredPencilSketchcoloredPencilSketchShadecoloredPencilSketchLineQuick open hatching in each thing's colour, lots of paper left.
CharcoalcharcoalVinecharcoalVineShadecharcoalVineLineLight, powdery vine charcoal laid as soft smudged masses, strokes at their edges.
CharcoalcharcoalCompressedcharcoalCompressedShadecharcoalCompressedLineDeep velvety black masses, bold broad strokes at their edges.
CharcoalcharcoalWhitecharcoalWhiteShadecharcoalWhiteLineWhite charcoal on a dark sheet: the scene drawn in light, in soft masses with strokes at their edges.
CharcoalcharcoalSmudgedcharcoalSmudgedShadecharcoalSmudgedLineCharcoal rubbed almost wholly into smoky masses, a few strokes at the edges.
Conte crayonconteSanguineconteSanguineShadeconteSanguineLineRed-brown sanguine darks, white heightening on a toned sheet.
Conte crayonconteBlackconteBlackShadeconteBlackLineDense black conte, white heightening on a toned sheet.
Conte crayonconteThreeCrayonsconteThreeCrayonsShadeconteThreeCrayonsLineSanguine, black and white: the three-crayon figure drawing.
Conte crayonconteSepiaconteSepiaShadeconteSepiaLineWarm sepia-brown with soft white heightening.
BallpointballpointBlueballpointBlueShadeballpointBlueLineBlue pen in patient crossing layers, blobs where strokes start, a faint shine.
BallpointballpointBlackballpointBlackShadeballpointBlackLineFine crossing black hatching building to deep black.
BallpointballpointRedballpointRedShadeballpointRedLineA red correction pen: quick red hatching, confident red outlines.
BallpointballpointStudyballpointStudyShadeballpointStudyLineClose blue-black cross-hatching in eight layers.
Notebook doodledoodleMargindoodleMarginShadedoodleMarginLineLoopy blue pen scribbles and wobbly, exaggerated lines.
Notebook doodledoodleFeltTipdoodleFeltTipShadedoodleFeltTipLineBold, rounded black felt-tip lines and chunky fills.
Notebook doodledoodleStickyNotedoodleStickyNoteShadedoodleStickyNoteLineA fast pencil doodle; pair it with a smooth paper tinted pale yellow.
Notebook doodledoodleCrayondoodleCrayonShadedoodleCrayonLineA child's wax crayon scribble in the colours of things.
Technical pentechnicaltechnicalHatchShadetechnicalPenLineRuled 45 degree hatching, crossed in the darks, and fine even pen lines with construction lines past the corners. Dark ink on white; on a dark or blue sheet both are drawn in white.

The technical pair is named for what it draws rather than <variant>Shade and <variant>Line; it swaps like every other pair. Each block's README lists its own defaults, what it gives up in motion and how it treats glass, water, particles, sky, fog, characters, foliage and world text.

Fill blocks: what a drawing starts from

Three Fill blocks lay the sheet a drawing starts from, under the shade. All three run one shader (pencilFill), so they swap with keepSettings = true. A shade block starts from a clean sheet at over = 0; set its over toward 1 to draw over the fill.

BlockWhat it is
sketchBlankFillA pale sketch blank: the picture lifted toward bare paper by paleness, keeping keepColour of its colour (0 a grey value drawing, 1 every colour). Pixel by pixel, so it adds nothing that can slide or shimmer.
storybookFillA soft storybook underlayer: warm, light pastel colour laid with the side of a soft coloured pencil and rubbed smooth, keeping more of each thing's colour than the sketch blank. The base of the storybookPencil look.
coloredPencilUnderlayFillA light first layer of coloured pencil: each thing's own colour in soft, even, light-pressure strokes over a pale blank, laid with the same surface-anchored strokes as the shade blocks. Surfaces seen edge-on and things past fadeEnd take it as rubbed colour, so a field never fills with long even bands.
local StyleKit = require("~styleKit.StyleKit")
StyleKit.applyLook("pencilSketch")
StyleKit.add("sketchBlankFill", { id = "fill" })                   -- a pale blank under the graphite
StyleKit.set("fill", { keepColour = 0.6, paleness = 0.6 })          -- keep more of the colour
StyleKit.set("shade", { over = 1 })                                 -- hatch over it
StyleKit.swap("fill", "coloredPencilUnderlayFill")                  -- a light coloured pencil first layer instead

Fill settings: keepColour, paleness and skyBare shape the blank; strokes (0 none, 1 full) and coverage set how much underlayer is laid; hardness, pressure, tooth, smudge, direction, angle, strokeLength, spacing, width, layers, patchSize, jitter, taper, fadeEnd, boil, wobble and seed mean what they mean on a shade block. Only what light reached is lifted, so with every light off the frame stays black.

Whole looks

StyleKit.applyLook(id) sets the blocks, their settings, the hold rate and the soft radius at once; { keep = { "line" } } keeps your own line. Every look boils at 8 drawings a second (holdRate = 8).

LookWhat it isShadeLinePaper
pencilSketchHB hatching that follows the forms, soft wandering outlines, lit planes left as paper.graphiteHBShadegraphiteHBLinepaperColdPress
charcoalStudyVine charcoal smudged into soft masses, compressed charcoal outlines, grey-white rough paper.charcoalVineShadecharcoalCompressedLinepaperRough
whiteCharcoalNightWhite charcoal on black paper: the scene drawn only in light.charcoalWhiteShadecharcoalWhiteLinepaperBlack
conteFigureStudyThree crayons on tan: sanguine, black and white heightening.conteThreeCrayonsShadeconteSanguineLinepaperTonedTan
ballpointNotebookBlue ballpoint cross-hatching on a ruled page.ballpointBlueShadeballpointBlueLinepaperNotebook
doodleMarginsLoopy blue pen scribbles and wobbly black felt-tip outlines on a ruled page.doodleMarginShadedoodleFeltTipLinepaperNotebook
coloredPencilWashWatercolor washes (wcWash, wcPaperFade, wcEdgeDarken, springPastel set) with hard coloured pencil hatched over them.coloredPencilHardShadecoloredPencilHardLinepaperColdPress
fieldNotesOnKraftThin washes (wcWash, earthy set), coloured pencil with white heightening, a fine black ballpoint outline.coloredPencilHardShadeballpointBlackLinepaperKraft
boldInkSketchDeep, contrasty 6B hatching under a heavy ink brush line that swells and tapers: inked pencils, graphic and punchy.graphite6BShadeinkBrushLine (inkBlocks)paperHotPress
technicalSketchRuled technical hatching and fine technical pen lines with construction lines, on smooth white paper.technicalHatchShadetechnicalPenLinepaperHotPress
blueprintWhite technical lines and white ruled hatching on cyanotype blue, the darks left as the sheet.technicalHatchShadetechnicalPenLinepaperBlack, tinted blue
storybookPencilWarm pastel colour laid with a soft coloured pencil and rubbed smooth (storybookFill), light coloured pencil hatching over it, a soft pencil outline.coloredPencilSketchShadecoloredPencilSketchLinepaperColdPress, warm

The four looks added in 1.1.29 are in every table below: the Lookbook motion and cost tables and the dim-room table.

The controls

StyleKit.set("shade", { ... }) and StyleKit.set("line", { ... }); StyleKit.settingsOf(id) lists every setting with its meaning, default and range.

Shade settingWhat it does
hardness0 very soft (6B, charcoal: dark, broad, fuzzy, fills the tooth) to 1 hard (2H: pale, thin, crisp, only on the tooth's peaks).
pressure0 a light touch (paler, thinner, skipping the tooth's valleys) to 1 pressed hard.
toothHow much the paper's tooth shows through the marks.
smudgeHow much tone is rubbed in between the strokes.
mass0 to 1: how much of the tone is laid as soft, smoky masses rubbed into the sheet rather than as strokes. Strokes show fully only at the edges of masses (where tone or form changes, or at a silhouette) and faintly inside them. Charcoal starts at 0.6 to 0.92, 4B at 0.7, 6B at 0.65, every other variant at 0 (strokes only).
direction0 follow the form, 1 one fixed angle, 2 crossed (later layers turn across the first), 3 scribbled loops.
angleDegrees: tilt from the form direction (0, 3) or from vertical (1, 2).
strokeLength, spacing, widthIn pixels at 1080p, kept on screen at any distance.
screenCalmHow much the strokes are rubbed into tone of the same value when most of the screen is dark enough to pile strokes up (a dark room, a night, a shadowed wall filling the view), so the screen never fills edge to edge with stripes. Each layer of strokes keeps its average coverage and only its stripes soften, so a calmed place is exactly as dark as it would be without the calm and darker places stay darker. An ordinary daylight view keeps its strokes: 1 (the default) fully, 0 never.
layersMarks per tone: how many layers of strokes the darkest tone builds up, 1 to 8.
lightest, darkest, contrast, toneMiddleWhich values stay bare paper, which get the darkest tone, how tone climbs between, and (on a toned sheet) which value the bare sheet stands for.
localTone, formShade, detailHow much a material's own darkness counts, how much faces turned from the sun are darkened, how much fine texture the tone follows.
hueHow much marks take the colour of what they draw (coloured pencil).
over0 strokes on a clean sheet, 1 strokes over what the blocks before made (a watercolor wash).
heighten, heightenFromWhite heightening on a toned sheet: the lights put on in their own strokes.
wax, burnish, gloss, blobsWaxy fill, burnished layers, shine where the medium is heavy, ballpoint blobs.
lead, deepLead, deepFromThe mark's colour, the colour the darkest layers turn to, and from which layer.
jitter, taper, patchSize, scribble, boil, seedHow loose the hand is, how strokes fade at their ends, patch size, loops, the redraw on each held step, another hand.
Line settingWhat it does
width, darkness, strokesPass width, how dark it lays, 1 to 3 passes per edge.
wander, wanderPx, overshoot, scribble, variationHow far passes wander off the edge, how long a wander is, how far lines run past corners, a doodler's wobble, pressure swelling and lifting.
hardness, pressure, tooth, smudge, blobs, hueThe same medium controls as the shade.
silhouette, crease, creaseAngle, contact, thresholdWhich edges draw.
thinNear, thinFar, farWidth, fadeEndHow lines thin and fade with distance.

Characters: PencilHero

Full-frame strokes are laid on world positions, so they slide across a character that moves. PencilHero draws a character or hero prop in the same medium inside its own materials, with every stroke on the character's own position:

local PencilHero = require("~pencilBlocks.PencilHero")
local count, why = PencilHero.paint("hero")          -- an entity name, id or entity: every model under it
PencilHero.list()                                    -- what is painted, with the shade block each was drawn with
PencilHero.unpaint("hero", { removeFiles = true })   -- every original back, the derived materials deleted

paint writes derived materials on the pencilSurface shader under /zero/source/pencilHero (each original's colour and texture kept, glass and water skipped), in the medium and tuning of the stack's first pencilBlocks Tone entry; opts.shade names another entry or shade block, and any shade setting in opts overrides it. It records the originals on a PencilDrawn component (synced, so every player sees the same hero), puts the target in mask group 3 and keeps every pencilBlocks Tone entry and every Fill entry (a watercolor wash, a sketch blank, a coloured pencil underlayer) off that group, since a Fill repaints the frame from the picture under it and would wash the hero's own strokes out (opts.mask = false leaves masks alone). Line and paper still draw over the hero. The materials are written when you paint: after swapping or tuning the shade, paint again to match. On each machine the component keeps the hero's stroke size matched to that screen and its boil on the stack's held clock.

A body is many small surfaces turning every way, and strokes that follow the form there change direction patch by patch and read as a woven texture, so the hero material hatches a figure the way a figure study is drawn: long parallel strokes at the shade's angle, laid in the figure's own space so they ride on it (figureHatch = 1, the default; 0 follows the form as the full-frame shade does). It follows the body's smooth normal, carried into the figure's own space per pixel, not its triangles. Shadow sides are drawn as a middle tone, as the full-frame shade draws them from the sky's light in the picture (shadowFill, default 0.4: how light a side the sun does not reach is drawn), and with every light off a painted figure goes black with the rest of the frame. Set either on the derived material, or pass it to paint (PencilHero.paint("hero", { figureHatch = 0 })).

With paperBlocks, inkBlocks and watercolorBlocks

The kits share one stack and one order (Fill, then Tone, then Line, then Ground), so mixing them is plain slot stacking:

  • Paper. Every paperBlocks paper swaps for the look's paper in one change. White sheets take darks; on toned sheets (paperTonedTan, paperTonedGrey, paperKraft) the bare sheet stands for the middle value and a conte block's heightening puts the lights on in white; on dark sheets (paperBlack, paperBlackboard, paperGreenChalkboard) a dark lead is drawn light. A sticky note is any smooth paper tinted pale yellow: StyleKit.set("paper", { tint = "#f8e58c" }) on paperHotPress.
  • Ink. Every inkBlocks line (inkPenLine, inkBrushLine, inkNibLine, inkSketchPen, inkDryLine) and styleKit's depthLine and sketchLine swap for the pencil line: StyleKit.swap("line", "inkPenLine"). The pencil lines find edges with inkBlocks' own edge rules, so they draw the same edges.
  • Watercolor. watercolorBlocks' Fill blocks (wcWash, wcGlaze, wcPaperFade, wcEdgeDarken and the rest) run before the shade; set the shade's over near 1 to hatch over the wash, as coloredPencilWash does. The guide "Coloured pencil over watercolor" goes step by step.

Quality

StyleKit.setQuality("low") is the cheaper setting for slower machines: the chain runs at half resolution, shade blocks draw at most 4 layers of strokes and give strokes up for soft tone sooner (at 0.6 of fadeEnd), and lines draw at most two passes. Strokes thinner than about a drawn pixel fade rather than breaking up, and far away strokes give way to a soft rubbed tone, so the horizon never fills with fine marks. Black stays black: every block returns through styleKit's shared exit, which holds every pixel at or below the value the scene gave it.

What is in the package

AssetWhat it is
blocks/Fifty-two blocks (25 shade, 25 line, 2 fill), each a .styleBlock naming one of three shaders, pencilShade.shader, pencilLine.shader and pencilFill.shader.
shaders/pencilCore.shaderModuleHow a medium lays its marks: lead and deep colour, hardness, pressure, tooth, smudge, wax, burnish, hue, shine, blobs, heightening. Shared by the shade blocks and the hero material.
shaders/pencilSurface.shaderThe hero material: the same drawing inside an object's own material.
PencilHero.module, PencilDrawn.componentPaint, unpaint and list hero characters.
looks/Twelve whole looks.
guides/Five guides: how to make a pencil sketch look, charcoal, coloured pencil over watercolor, ballpoint notebook doodles, and an example game given a pencil sketch look from an empty world.
tests/The test suite (pencilBlocks.testSuite) and the closed room its every-light-off check is built from (pencilTestRoom.material on pencilTestRoomSurface.shader, lit only by the scene's lights).
workflows/pencilWorld.workflowTurn an existing world into a drawing, report what changed, leave it removable.

Lookbook numbers

Measured on 2026-11-07 with 1.1.8 and lookbookKit 1.0.21 on a freshly started engine, the 3D view at 1920 x 1080 (data in the Cookbook's design/lookbook/pencil_runs_118.json): one discarded warm-up run, then the unstyled frame and every whole look played over the Lookbook's full path (pan, orbit, dolly, a 10x zoom, a slow light, a turning sun, every light off). The table is one pass on this release's shaders. An earlier two-pass batch on another freshly started engine, made just before the last shader change (the screen calm under a lit sun, see the 1.1.8 history), agreed with itself to 0.01 on every number and with this pass to 0.02, so a reader on the same engine should expect these numbers within a few hundredths. They were measured on the desktop engine; another engine build can draw a look a little differently (the dark-room table below differs by up to 12 levels in its lamp-lit room between builds), so on another build read each look against the unstyled row measured on that same engine.

The marks stay on things for every camera move. "Marks on things" correlates the look's own marks (the frame less the frame before the look) with the frame before, once moved along with the scene and once left on the same screen spot: negative means the marks went with the things they are drawn on, positive that they stayed on the screen. Every look reads -0.48 to -0.78 in the pan, the orbit, the dolly and the zoom. The Cookbook's design/shots/pencilstage/marks_stay_put.png (its numbers in marks_stay_put.json) shows it by eye, one row per move, Pencil sketch on the campaign stage: before the move, after it cropped where the spot now is, and after it cropped where the spot used to be on the screen.

  • Pan (3 degrees) and orbit (10 degrees): the strokes on the back wall sit in the same places on the wall before and after; at the old screen spot the third crop shows other things.
  • Zoom (50 to 30 degrees): the strokes keep their drawn width, so as the wall comes nearer the shade adds a finer layer of strokes, the way a draughtsman draws a nearer wall with the same pencil. The strokes are laid on the wall, not on the screen: the Lookbook scores the zoom -0.63 to -0.78 for every look.
  • Moving figure (figure and camera move 0.4 m together): painted with PencilHero, the figure's hatching is the same stroke for stroke after the move. The last row is the same move unpainted under the full-frame shade, for contrast: there the strokes are laid on the world around the figure, so they change as it walks through them, which is why PencilHero exists for characters (the suite measures it: the painted figure changes by about 6, full-frame strokes by 19 to 31).

Why the 1.1.0 table read positive. Up to lookbookKit 1.0.20 the Lookbook took the frame before the look one pixel to the left and up: it turned a pixel position into a whole pixel by rounding down a value that came out a hair under the whole number. The "marks" it measured were then the picture's own one-pixel edges, which sit on the screen, so every look scored positive, and so did plain paper, whose grain is laid on world positions (+0.95). The strokes had not moved onto the screen; the measurement had. lookbookKit 1.0.21 samples at the pixel's centre: an empty stack has no marks, plain paper reads -0.52, and every pencil look reads as above.

How to read the other columns, against the unstyled row rather than as pass marks:

  • Warped difference (0..255, lower is steadier): how much each frame differs from the one before once motion is taken out. The held clock's redraw is part of a look's number; every look is steadier than the unstyled render in the pan.
  • Screen-stuck in the pan (the whole picture's detail against the same screen spot; more negative is less stuck). The two notebook looks read near 0 because the ruled page is a sheet lying under the drawing, as a real page does: its rules stay put on the screen while the strokes move with the things they draw (their own marks read -0.51 to -0.78 in every move).
  • Mark growth over the 10x zoom: about 1 means marks keep their size in the picture, about 10 that they grow with the things they sit on. Pencil strokes keep their drawn size by laying finer or coarser octaves of strokes as things come nearer, as a draughtsman does, while staying on the surfaces.
  • Every light off: the frame's mean brightness at the Lookbook's every-light-off moment; every look stays at or below the unstyled frame's black.
StackWarped difference, whole pathWarped difference, panMarks on things: panorbitdollyzoomScreen-stuck in the panMark growth over the 10x zoomEvery light off (0..255)
unstyled2.306.57-----0.537.881.22
pencilSketch1.882.41-0.53-0.68-0.70-0.78-0.730.951.00
coloredPencilWash2.664.62-0.48-0.59-0.62-0.63-0.341.041.00
charcoalStudy3.053.96-0.55-0.68-0.70-0.77-0.740.980.99
conteFigureStudy2.293.54-0.56-0.66-0.69-0.69-0.710.980.92
ballpointNotebook1.962.13-0.51-0.68-0.70-0.78-0.050.941.02
doodleMargins2.162.13-0.51-0.68-0.70-0.78-0.050.941.03
fieldNotesOnKraft2.524.76-0.56-0.63-0.65-0.61-0.590.980.94
whiteCharcoalNight1.712.59-0.54-0.68-0.70-0.77-0.721.001.07
boldInkSketch1.572.14-0.53-0.68-0.69-0.79-0.660.940.45
technicalSketch1.442.14-0.53-0.68-0.70-0.79-0.660.941.01
blueprint3.104.31-0.56-0.65-0.70-0.72-0.701.000.75
storybookPencil2.565.00-0.41-0.45-0.48-0.47-0.341.190.85

The last four rows were measured on 2026-11-08 with 1.1.30 and lookbookKit 1.0.34, by the same method (data in the Cookbook's design/lookbook/pencil_runs_1129.json). The unstyled row and pencilSketch were measured again in the same session and came out exactly as above, so all rows read on one scale; each of the four looks was played twice and the two runs agreed within 0.01. storybookPencil reads less negative on "marks on things" and grows a little more over the zoom because most of what it adds is the storybook fill's soft colour, which sits on things pixel by pixel rather than as strokes, as with coloredPencilWash.

GPU cost, the style passes alone at 1920 x 1080 over the Lookbook's pan, measured in a test client with the editor open beside it and no other test running (expect a little less on a machine that runs the game alone):

LookHigh (ms)Low (ms)The shade's pass on high (ms)
pencilSketch10.52.66.7
coloredPencilWash10.92.77.2
charcoalStudy10.82.76.7
conteFigureStudy11.52.87.5
ballpointNotebook11.42.76.7
doodleMargins9.92.56.6
fieldNotesOnKraft12.83.27.9
whiteCharcoalNight12.63.08.6
boldInkSketch9.12.16.8
technicalSketch9.02.36.7
blueprint9.72.47.2
storybookPencil12.63.07.1

Measured on the desktop engine (data in the Cookbook's design/lookbook/pencil_cost_118.json), the 3D view at 1920 x 1080. The shade is most of the cost on high; the low quality tier costs about a quarter of it.

Block by block: marks on things, and their size in a 10x zoom

Measured on 2026-11-07 with 1.1.8 at 1920 x 1080 on a freshly started engine (data in the Cookbook's design/lookbook/pencil_blocks_motion_118.json, script beside it). Every block holds still on things at least as well as sketchLine, and every shade draws its marks at a pencil's width after the zoom.

Every Tone and Fill block alone (no paper, the held clock fixed, boil = 0), and styleKit's sketchLine as the yardstick, on the campaign stage (lookbookShowcaseKit) under its own sun with the props' weathering taken off, the camera at 7, 4.5, 15 from the stage looking at its centre. For each block the marks alone are taken (the styled frame less the unstyled frame at the same camera pose, two by two pixels averaged, over the middle of the view):

  • On things: the camera turns one degree (the scene moves 20.2 px on screen). The marks after the turn are correlated with the marks before, moved by the camera's own shift. 1 means every mark is still on the same spot of the same thing; marks that swim or warp score lower.
  • On the screen: the same, unmoved. A screen texture (a shower-door effect) would score high here and low on things. For every block it is lower than on things, and the best match is at the camera's shift.
  • Mark size (mean |marks| / mean |gradient|, in pixels) before and after a 10x zoom (field of view 50 to 5.3 degrees) onto the shadowed wall behind the figure. A texture fixed to the surfaces would grow about 10 times. Pencil strokes are laid in octaves, so as the wall comes nearer the shade lays finer strokes and the marks keep a pencil's width, as a draughtsman draws a nearer wall with the same pencil.

"Holds still at least as well as sketchLine" is yes when the block's on-things correlation is at least sketchLine's, its best match is at the camera's shift (within 2 px) and it scores higher on things than on the screen. "Drawn at a pencil's width in the zoom" is yes when the marks after the zoom measure 2 to 12 px and did not grow 4 times or more. Up to 1.1.7 this column asked for a size change between 0.25 and 4 times; that read the wide view's number as a stroke width, but from 8 m it mostly measures the broad tone masses across a whole wall, so the light shades whose zoomed view is the most finely hatched read lowest (2H, white charcoal, hard and sketch coloured pencil, the sticky-note and felt-tip doodles and red ballpoint at x0.18 to x0.24). The zoomed views themselves are in the Cookbook's design/shots/pencilstage/zoom_by_eye_118b.png. Before 1.1.8 the zoomed shadowed wall carried almost no strokes in any shade (the screen calm, see the 1.1.8 history); now every shade hatches it, which is also why every mark size after the zoom is smaller than the 1.1.0 table's. The two fills change colour rather than laying strokes, so their size follows the picture's own shapes. sketchLine scores lower on things (0.61) because its lines are a few pixels thin, so a one-pixel resampling difference costs more correlation; its lines still sit on the edges they draw.

BlockOn things (at the camera's shift)On the screen (unshifted)Best match atMark size before → after the 10x zoom (px)Size changeHolds still at least as well as sketchLineDrawn at a pencil's width in the zoom
sketchLine (styleKit, the yardstick)0.610.24-20 px2.1 → 2.0×0.92
ballpointBlackShade0.860.44-20 px14.0 → 4.5×0.32yesyes
ballpointBlueShade0.880.50-20 px16.4 → 4.6×0.28yesyes
ballpointRedShade0.930.64-20 px22.8 → 5.4×0.24yesyes
ballpointStudyShade0.820.28-20 px10.3 → 4.2×0.41yesyes
charcoalCompressedShade0.920.54-20 px14.1 → 5.8×0.41yesyes
charcoalSmudgedShade0.920.55-20 px16.2 → 6.5×0.40yesyes
charcoalVineShade0.930.56-20 px18.2 → 6.4×0.35yesyes
charcoalWhiteShade0.960.75-20 px38.4 → 6.9×0.18yesyes
coloredPencilBurnishedShade0.860.45-20 px10.0 → 4.8×0.48yesyes
coloredPencilHardShade0.940.66-20 px26.1 → 5.3×0.20yesyes
coloredPencilSketchShade0.950.66-20 px28.4 → 5.8×0.20yesyes
coloredPencilWaxyShade0.890.40-20 px14.3 → 4.2×0.30yesyes
conteBlackShade0.910.36-20 px12.4 → 5.0×0.40yesyes
conteSanguineShade0.920.47-20 px16.4 → 5.4×0.33yesyes
conteSepiaShade0.900.41-20 px14.4 → 5.0×0.35yesyes
conteThreeCrayonsShade0.920.43-20 px14.8 → 5.1×0.34yesyes
doodleCrayonShade0.920.57-20 px18.0 → 5.0×0.28yesyes
doodleFeltTipShade0.920.60-20 px22.6 → 5.3×0.23yesyes
doodleMarginShade0.900.56-20 px19.0 → 4.8×0.25yesyes
doodleStickyNoteShade0.950.70-20 px29.9 → 6.0×0.20yesyes
graphite2HShade0.960.74-20 px33.9 → 6.4×0.19yesyes
graphite4BShade0.910.57-20 px23.2 → 6.0×0.26yesyes
graphite6BShade0.910.50-20 px12.2 → 5.5×0.45yesyes
graphiteHBShade0.850.41-20 px13.3 → 4.3×0.32yesyes
graphiteMechanicalShade0.930.64-20 px22.3 → 5.6×0.25yesyes
coloredPencilUnderlayFill0.950.64-20 px28.8 → 18.3×0.63yes(a fill)
sketchBlankFill0.960.72-20 px41.9 → 37.1×0.89yes(a fill)

Dim scenes and dark rooms

A pencil look draws a lit scene on white paper, so a sunny scene gets brighter where it is left as bare sheet. A dim scene must not: no look brightens a dim or unlit room by more than a few levels, and with every light off the drawing is black. Two things keep it so.

  • The sheet is no brighter than the scene's lights. Every shade and fill block reads how bright the picture's brightest parts are as the renderer drew them, and draws the sheet no brighter than that. In daylight or a well lit room the sheet is white; in a dim room the whole drawing starts dim.
  • styleKit's dim guard keeps a dark scene dark (styleKit 1.4.0). Every shade block sits in the Tone slot, so it keeps dim with no setting to turn on. After the whole stack is drawn, styleKit compares the drawing with the frame as rendered, smoothed over about 36 pixels, and scales the drawing down wherever it came out brighter than the room was. It scales rather than clips, so every stroke and mass keeps its shape. It acts only on a dim picture: a room with bright spots but a low overall level (a night street, a room of dark walls and floor under one lamp) is guarded, a room with no light switched on counts as unlit however bright the engine's fallback draws it, while daylight, an overcast day and a well lit room (an overall mean of 0.42 or above, brightest parts at 0.55 or above, some light other than an ambient one on) are handed on exactly as drawn.

Before 1.1.0 the shade held each pixel of the wash or blank it lay over to the rendered brightness under it. That kept dim rooms dim, but in daylight it printed the render's brick and tile texture back over the watercolor wash, so the Coloured pencil wash look read like a plain render. That cap is gone: the wash is laid as painted, and the dim guard keeps dim rooms dim for the whole stack instead.

When most of the screen is dark enough to pile strokes up (a dark room, a night, a shadowed wall filling the view), each layer of strokes is rubbed into its own average tone: the value stays exactly what the strokes gave, and only the stripes soften (to a fifth of their contrast with no sun, as in a dim room or a night; under a lit sun only to 60%, so a shadowed wall filling a daylight view, as when the camera zooms onto it, stays hatched), so the screen never fills edge to edge with fine high-contrast stripes and the darkest band stays the darkest. Up to 1.1.23 the calm turned the strokes down and rubbed in only one layer's worth of tone instead; a medium that builds its darks from many light layers (blue ballpoint, coloured pencil) then drew its darkest band lighter than its mid band whenever the calm was on. A dim picture (the rendered mean under about 0.45, or under about 0.55 with no sun: the engine's fallback light when nothing is switched on draws a room mid grey) is always calmed. In a brighter picture only dark tones count (a tone of 0.55 and up of the block's range, where strokes pile layer on layer), worked out the way each pixel works out its tone (the light on a surface over its material's colour, its form, the block's range), so a sunlit dark brick counts as lit; calming starts when 35% of the screen is that dark and is full at 60%. Before 1.1.7 light tones counted too, so an ordinary daylight view (the cottage from 8 m) calmed and every look lost its strokes there: Pencil sketch read as a pale render with outlines. screenCalm = 0 turns it off. Both readings are taken on a fixed 5 x 5 grid of the screen, the same for every pixel, so they add no pattern of their own; they follow the camera, never the clock.

The test suite reads every look, and both Fill blocks over the graphite shade, in six lights on every run, with a figure painted with PencilHero in the room:

  • a lamp in a grey room (lit);
  • the same lamp in a room of dark walls and floor (a dim room with bright spots, mean about 82 of 255);
  • dim ambient light;
  • the lamp turned low (0.35);
  • no light enabled (the engine's fallback ambient);
  • every light off.

It fails if any stack makes the dark-walled, dim, low-lamp or fallback room more than 3 of 255 brighter than the unstyled frame or fills more than 5% of it with fine high-contrast stripes, fills more than 45% of the lamp-lit room with them (a lit room keeps its hatching since 1.1.7: Ballpoint notebook, the most, reads 30% or 16% depending on which state the room lands in, see below), leaves the lamp-lit room as rendered (a stack that drew nothing), lays a light rim around the pale props of the dark-walled room on more than 1% of its dark places, or lifts the every-light-off room more than 1.5 above black. In the dark-walled lamp room the Ballpoint notebook look lifted the frame by 8.5 before the dim guard (styleKit 1.4.0) and draws it about 7 darker than rendered now.

Measured by the suite on 2026-11-08 with 1.1.24 on the desktop engine (built-in library f39f044), the 3D view at 1920 x 1080: the mean brightness of the middle half of the frame, 0..255, at the suite's own camera in the suite's own closed room. Two runs in one session gave the same number in every cell. The suite's notes say which view size a run measured at. The first row is the frame as rendered; every other number is the drawing's. The dim rooms read a few levels lower than up to 1.1.23 (Pencil sketch: dim ambient 41.61 then, 38.22 now), because the screen calm now keeps the full value of the strokes it rubs in.

StackLamp, grey roomLamp, dark wallsDim ambientLow lampNo light enabledEvery light off
unstyled158.1382.0466.1041.16126.730.00
pencilSketch215.0675.1438.2226.98126.580.00
coloredPencilWash180.6275.3744.0531.24126.610.00
charcoalStudy182.9667.6828.5719.87125.200.00
conteFigureStudy168.6266.3429.8319.83126.590.00
ballpointNotebook222.8975.4345.4131.85126.550.00
doodleMargins236.1476.3651.3333.68126.530.00
fieldNotesOnKraft140.0368.5551.7531.81126.050.00
whiteCharcoalNight57.3129.8725.9222.4236.520.00
sketchBlankFill (over graphite HB)197.2673.4538.6219.38126.600.00
coloredPencilUnderlayFill (over graphite HB)206.0773.5737.7318.91126.580.00
boldInkSketch (1.1.29)223.4673.8533.2421.04126.090.00
technicalSketch (1.1.29)229.1476.0447.8632.95126.610.00
blueprint (1.1.30)88.0654.1561.2835.3382.100.00
storybookPencil (1.1.29)187.9474.5448.3826.55126.580.00
storybookFill (over graphite HB, 1.1.29)179.7170.1336.4018.12126.560.00

Only the lamp-lit grey room comes out brighter, as a pencil drawing of a lit room on white paper does (except Field notes on kraft and White charcoal night, whose sheet is darker than the room). In every dim room every stack is drawn darker than it was rendered, by 0.1 to 1.6 in the room with no light enabled (where the engine's fallback light draws the room mid grey) and by 5.6 or more in the others; White charcoal night, drawn in light on black paper, far below. With every light off the room stays at 0.00 under every stack. In this run fine high-contrast stripes cover at most 30.0% of the lamp-lit room (Ballpoint notebook; Pencil sketch 21.3%) and at most 1.2% of any dim room, and at most 0.28% of the dark-walled room's dark places are drawn more than 25 levels brighter than rendered.

The lamp-lit grey room has two states, and a run lands in one of them. The table above is the brighter one. In the other, four looks draw that room darker and less finely hatched (measured on 1.1.15: Ballpoint notebook 213.41 with 16.3% of the room in fine stripes, Doodle margins 224.12, Pencil sketch 214.96 with 16.8%, Coloured pencil wash 178.22), the dim rooms a level or two darker, and the unstyled row identical. Both states have been seen on the same engine twenty minutes apart and on fresh engines, so it is not the engine build (as this README said up to 1.1.14), nor the view size (at 1280 x 720 Ballpoint notebook reads 212.1 with 19.2% stripes); what selects the state is not known yet. So expect the lamp-lit grey room to read up to about 12 levels below this table, and 16% to 30% stripes. The promise is in the dim rooms and the every-light-off room, and it holds on every build measured: every stack draws every dim room darker than unstyled, and the every-light-off room is 0.00. A long session can add a little light to the closed room as well (the 1.1.7 table, measured late in a long editor session, read 1.4 to 4.3 levels higher; in the Lookbook scene the every-light-off room read 4.67 unstyled). What the suite checks holds in every case, because it compares each stack with its own unstyled frame in the same run: no dim room more than 3 levels brighter than rendered, every light off within 1.5 of black, and stripes in the lamp-lit room under 45%, which leaves 15 points of room above the most finely hatched look on any build measured.

What repeats from run to run, and what does not. Measured on 2026-11-08 with 1.1.22 and styleKit 1.4.14: the suite run twice in one engine process in a brand-new world with only pencilBlocks installed, right after the first example above. Both runs passed (the wide-band case skipped both times, as it must without the campaign stage), and all 133 readings in the suite's notes were identical. With 1.1.24 in the Style Cookbook (all thirteen cases, the campaign stage included), two runs in one session read every dark-room and ramp number identically; the campaign stage readings moved (below). With 1.1.21 the same check passed twice but 19 readings differed: styleKit kept a stale camera anchor after a clear (fixed in 1.4.14, see the 1.1.22 history). Two runs on separately started engines (1.1.20) agreed on 125 of 133 readings. What still varies between engine processes:

  • A few dark-room readings can land 0.01 apart, even between two runs in one session. A reviewer's two runs in one session read Pencil sketch in the dark-room case at 215.06 and then 215.07, and Conte figure study's share of fine stripes at 0.19% and then 0.20%. Our own pairs in one session matched to the last digit (all 133 readings on 2026-11-08, and the dark-room case twice more to four decimals), and pairs on separately started engines differed by 0.01 on the same few lines. Within one build of the room every frame is identical to four decimals, with the engine's per-frame clock running or held, so the frame the suite reads is not the cause; what moves those last digits from one room build to the next is not known. So read the room numbers as repeating within 0.01 of 255, and stripe shares within 0.01 of a percentage point, not to the last digit.
  • The lamp-lit grey room can land in a second state. In about one fresh engine start in six today, every stack drew that room darker and less finely hatched (Ballpoint notebook 213.41 with 16% fine stripes instead of 222.89 with 30%), and the dim rooms a level or two darker; the room table above is the brighter state. The room's figure, the render path, the scene's direct light and styleKit's start-up writes were each ruled out by a direct test; what selects the state is not known yet.
  • Campaign stage readings move between stage builds where lookbookShowcaseKit's campaign stage is installed. With 1.1.24 in the Style Cookbook, two runs in one session read the unstyled stage at 48.2 and 46.4 and each medium within 4.3 of its other reading, and three stage builds over an hour read it at 46.4, 48.2 and 66.7; within one build the stage is steady to 0.003 over 45 seconds, and the scene's sky is held to one fixed day (since 1.1.17). The stage is assembled from other kits whose state outlives a build (its weathering overlays among them), and what moves it was not pinned down. Every stage case compares readings taken within one build, so each passes either way; read the stage numbers as a few levels wide. A fresh install draws on the suite's own stand-in set instead, which repeated to the last digit (above). The density check's stroke counts can land in one of two close states, up to 3% apart, on the campaign stage.

Every limit the suite checks against has a wide margin over all of these, so whether a case passes never depends on which value a run lands on.

Players, enemies and pickups in one drawing

Every block honours styleKit's style channels. An object's line weight sets the width of every pencil line drawn on it and how densely its shading strokes are packed (a boss shaded dark and close, the hero lighter, props sparse); how much shading (tone), colour (fill) and paper reach it follows its marks, and its value draws it in coloured pencil (0 the look's lead, 1 strokes and a soft layer in its own colour, pushed toward its hue so a gold coin lit pale by a white sky is still gold; a metal falls back to the colour it shines). A coin with value 1 is drawn in gold among the graphite. The styleRoles package's pencil sketch set does this for sixteen game roles in one call (StyleRoles.use("pencilSketch")). Objects with no channel are drawn exactly as before.

Known limits

  • Offscreen pictures do not draw the look (#17427). Post effects have no frame history or per-camera inputs, so offscreen captures, mirrors and render-to-texture cameras show the scene without the drawing. Look at the presented frame (capture with source = "presented").
  • Strokes slide on moving characters unless the character is painted with PencilHero. Lines follow moving things exactly, since edges are found every frame.
  • A screen that darkens into strokes rubs them in. With screenCalm on (the default), when a scene darkens so far that most of the screen is dark tone, the strokes are rubbed into tone of the same value instead of adding up edge to edge: a deliberate trade, so a dark room never fills with stripes. Each layer keeps its average coverage, so a calmed band is as dark as it would be without the calm (1.1.24). Under a lit sun the strokes keep at least 60% of their contrast (1.1.8), so a daylight view zoomed onto a shadowed wall reads as drawn, not as a soft render. Within any one frame darker places still carry more strokes laid over the lighter places'; screenCalm = 0 keeps every stroke at full strength.
  • Strokes follow the light. A moving light or a day cycle adds and removes strokes one by one; that is the drawing reacting, not flicker. A scene with very strong ambient light gives a pale drawing: raise formShade or lower lightest.
  • Glass, water and particles have no depth, so strokes and lines of what is behind them are drawn through them.
  • The hero material draws its own paper grain. No shared paper reaches a material, so a painted character's strokes catch on a soft grain of its own rather than on the stack's paper. Materials on a custom shader keep their colour and texture but lose their own texture layout. The material shows flat grey until its first compile lands.
  • Post-process shaders have no preview image, so pencilShade.shader, pencilLine.shader and pencilFill.shader show none.
  • Right after a block is first added, its first frames can draw without it, or dark (#22321). An effect reads as published while its shader still compiles; the look appears a moment later. Before judging a first frame, check two things. In the shell, zero styleKit stack (in the console, tools.use("styleKit", "stack")) lists under running.effects every effect of the stack by name (sk.shade.graphiteHBShade and so on); each must read true, which means it drew last frame. Then, in the console, #renderer.compiledShaders() must stop growing: read it, wait a second, read it again, and judge the frame once the two numbers match. That is what the suite waits for. A block shader's own compileStatus() (on the shader asset, asset.resolve("pencilBlocks.blocks.pencilShade", "shader"):compileStatus()) is not the check: it can read pending while the look draws.
  • A setting written before the shader compiles reads as its default (#22741). Since styleKit 1.6.5 a new block's settings are written every frame for its first three seconds, so a look switched on draws its settled picture from its first frames; after that styleKit renews its writes twice a second.
  • The Lookbook and any per-frame measurement need every-frame drawing (#22740). A world script that steps once per drawn frame stalls while the editor declines frames; lookbookKit holds every-frame drawing for its runs, and the pencilBlocks suite does the same.
  • Run the suite by its asset, in a scene nobody else is changing (#22322, #22937). A bare test run also runs every installed package's suite, and a run cannot pick single cases: asset.resolve("pencilBlocks.tests.pencilBlocks", "testSuite"):run() runs pencilBlocks' thirteen tests and nothing else. The suite draws its stacks on the open scene's synced StyleStack (and puts the stack back after), so a second editor or agent changing that scene's look, or loading another scene into it, while the suite runs changes what the suite measures. Run it where you are the only one working, and let it finish: a run stopped part way leaves the scene's lights and stack as that case had them (zm restore <scene> discards those edits).
  • Every light off means every light off. With no ambient light enabled the engine falls back to a default ambient, and an outdoor scene keeps the sky's captured light with every light disabled, so "lights off" is only black where the scene's ambient and sun are kept at 0 and no sky light reaches. The suite's check builds its own closed room for that, so it holds in any world, whatever the world's own scene holds.
  • Installs with several dependencies can be slow (#21420). Give the install time to finish, and check /zero/source for the package folder.
  • Search can serve an older description for a while after a release (#19258).
  • Multiplayer. The look lives on the scene's synced StyleStack and painted characters are synced, and the held clock runs on wall-clock UTC, so every machine shows the same strokes and the same boil step. A real second machine was not tried from here (#14448: two engines on one account join as one player).
  • Frame rate on the web and on a second peer. On a slower machine call StyleKit.setQuality("low"). Every number in this README was measured on the desktop engine; a browser engine cannot be started from an agent's tools (#20946), so the web build is checked by hand.

The limits every styleKit block shares are listed in styleKit's README under Known limits.

  • Other engine issues that apply here: every render pass is queued again every frame, so script cost grows with the passes a look queues (#24548); a per-object value cannot reach a geometry pass, so per-object looks go through styleKit's object mask (at most eight distinct settings) or need passes of their own (#24549); a two-client editor sometimes refuses to load two different scenes at once (#24666, #24667); a world boot can wait about 30 s on a network handshake before content starts (#24669).

Removing it

StyleKit.clear() (or deleting the scene's StyleStack entity) takes the look off. The only material change pencilBlocks makes is PencilHero.paint; PencilHero.unpaint(target, { removeFiles = true }) puts every original material back and deletes the derived files.

Version history

1.4.7 (2026-11-12): pins inkBlocks 1.2.10, lookbookKit 1.0.52, paperBlocks 1.2.2, styleKit 1.7.8, watercolorBlocks 1.2.12 (styleKit 1.7.8 logs no second refusal for the old roles field on a scene saved before styleKit 1.5). Nothing else changed.

1.4.6 (2026-11-12): pins inkBlocks 1.2.9, lookbookKit 1.0.51, paperBlocks 1.2.1, styleKit 1.7.7, watercolorBlocks 1.2.11 (styleKit 1.7.7: a role set or look another player changes reaches everyone in the session). Nothing else changed.

1.4.5 (2026-11-12): pins inkBlocks 1.2.8, lookbookKit 1.0.50, paperBlocks 1.2.0, styleKit 1.7.6, watercolorBlocks 1.2.10 (styleKit 1.7.6: an ink set by hand on one object reads as set, and a set for a kind of game carries its groups before a look is picked). Nothing else changed.

1.4.4 (2026-11-12): pins inkBlocks 1.2.7, lookbookKit 1.0.49, paperBlocks 1.1.32, styleKit 1.7.5, watercolorBlocks 1.2.9 (styleKit 1.7.5: its checks pass where styleKit is installed alone). Nothing else changed.

1.4.3 (2026-11-12): pins inkBlocks 1.2.6, lookbookKit 1.0.48, paperBlocks 1.1.31, styleKit 1.7.4, watercolorBlocks 1.2.8 (styleKit 1.7.4: installs on its own into an empty world again). Nothing else changed.

1.4.2 (2026-11-12): pins inkBlocks 1.2.5, lookbookKit 1.0.47, paperBlocks 1.1.30, styleKit 1.7.3, watercolorBlocks 1.2.7 (styleKit 1.7.3: a role reaches what a kit draws in an object's place (pickupsKit's coins), and looks installed after the first scan are found). Nothing else changed.

1.4.1 (2026-11-12): the suite waits for styleKit to draw before it reads a frame. styleKit 1.7.0 shows the frame as rendered until every shader a look uses has compiled; in a fresh engine the first compiles take seconds, and the suite's wait (the engine's compile queue going quiet for a second) let it read frames while styleKit was still waiting, so half its frames were the plain scene. "The redraw runs on the held clock without flashing" then failed on the first run in a fresh world (same held moment 0.53 apart, brightness 211 against 182: a plain frame against a drawn one) and the add/remove round trips saw a 0.62 noise floor; warm, the same checks read 0.000 and 0.028 with the brightness steady. The suite now also waits while StyleKit._status().waiting is set. No block, shader or look changed: the pencil redraw itself always held still on its clock (on its own, one held step changes the stand-in stage by 0.028 and its brightness by 0.03 of 255).

1.4.0 (2026-11-12): an object's line weight now presses the pencil harder or lighter: a heavier weight darkens its tone (more layers, and some tone even on lit faces), a lighter one takes most of it away, where before weight only packed the strokes closer or spread them and the overall tone barely moved. On styleKit's channel probe: weight 3 reads 37 against 204 with no channel, weight 0.3 reads 245 (styleKit 1.7.0's dim guard lets a lighter weight lighten; before, a heavy weight read only a few points darker and a light one stayed at the rendered value). Objects without a channel draw exactly as before. Known limits drop #19849, #22425, #22812 and #20750, which are closed. Pins styleKit 1.7.0, paperBlocks 1.1.29, watercolorBlocks 1.2.6, inkBlocks 1.2.4 and lookbookKit 1.0.46.

1.3.4 (2026-11-11): pins lookbookKit 1.0.45 (lookbookKit 1.0.45 brings its Lookbook scene with it when it is installed as another package's dependency, so the toonBlocks, inkBlocks and paperBlocks unlit-frame tests and zero lookbook run in an installed world). Nothing else changed.

1.3.3 (2026-11-11): The every-light-off test judges the fallback room (no light switched on, the engine's fallback light alone) as a dim room only where it draws dim, under 90 of 255. In the Style Cookbook it does; in a freshly made world the fallback light draws it mid-lit (126.7 of 255), and there the pencil look rightly draws it as a lit room, which the test had called lifting a dim room. Pins styleKit 1.6.5, lookbookKit 1.0.44, inkBlocks 1.2.3, paperBlocks 1.1.28 and watercolorBlocks 1.2.5 (styleKit 1.6.5 draws a newly switched-on look's settled picture from its first frames instead of a lighter, half-set one for two to four seconds; lookbookKit 1.0.44 finds its Lookbook scene where it is installed).

1.3.2 (2026-11-11): pins inkBlocks 1.2.2, lookbookKit 1.0.43, paperBlocks 1.1.27, styleKit 1.6.4, watercolorBlocks 1.2.4 (styleKit 1.6.4 puts a role given to a player template on the body every joining player is copied from and rebuilds the role mask at most four times a second, and toonBlocks 1.3.6 starts its toon ink by itself and draws every role ink through one shared flood). Every README in the package (each block, look, shader and guide) now ends with a Known limits section naming the engine issues that apply there.

1.3.1 (2026-11-11): pins inkBlocks 1.2.1, lookbookKit 1.0.42, paperBlocks 1.1.26, styleKit 1.6.2, watercolorBlocks 1.2.3 (styleKit 1.6.2 finds the scene's stack again when the one it held was removed (after a play restart a look or role set change failed with "the component this proxy names was removed")). Nothing else changed.

1.3.0 (2026-11-11): pencil shading honours an object's style channel value as coloured pencil: 0 the look's lead, 1 strokes and a soft layer in the object's own colour, pushed toward its hue so a gold coin lit pale by a white sky is still drawn in gold (a metal's dark surface colour falls back to the colour it shines), with its darkest layers kept in that colour. Together with styleRoles 1.1.0's pencil set (blue, red and ochre leads by variant, enemies packed dark, players light), players, enemies and pickups now read apart at a squint under every pencil look. Objects with no channel draw exactly as before. A style channel test on styleKit's ChannelProbe. Pins inkBlocks 1.2.0, lookbookKit 1.0.41, paperBlocks 1.1.25, styleKit 1.6.1, watercolorBlocks 1.2.2.

1.2.1 (2026-11-11): pins inkBlocks 1.1.1, lookbookKit 1.0.40, paperBlocks 1.1.24, styleKit 1.5.1, watercolorBlocks 1.2.1 (styleKit 1.5.1 keeps the scene's role set working in an editor that had an older styleKit open). Nothing else changed.

1.2.0 (2026-11-11): style channels. pencilLine takes the object's line weight as its width and pencilShade as its stroke density (heavier packs the strokes closer), and every block is weighed by the object's marks and its entry's variant through styleKit 1.5.0. Objects with no channel are drawn exactly as before. Pins styleKit 1.5.0, paperBlocks 1.1.23, watercolorBlocks 1.2.0, inkBlocks 1.1.0, lookbookKit 1.0.39.

1.1.31 (2026-11-08): moves to the engine's current built-in library (@builtin e89b9db), whose update rewrote small reference files inside the package (a world on an older built-in library is hidden from players). Pins styleKit 1.4.21, paperBlocks 1.1.22, watercolorBlocks 1.1.18, inkBlocks 1.0.21, lookbookKit 1.0.35. The package description now says what it gives you and how to install and use it. Nothing else changed.

1.1.30 (2026-11-08): the Lookbook numbers now cover all twelve looks: boldInkSketch, technicalSketch, blueprint and storybookPencil were measured over the full camera path and for GPU cost, by the same method as the first eight, with the unstyled row and pencilSketch measured again in the same session as anchors (both matched the existing rows exactly). The measurement found blueprint flickering in a pan (frame-to-frame difference 9.91 against 6.57 for the plain render): its thin white hatching lay right at the black paper's threshold for putting lights on, so the soft edges of the strokes switched on and off as they moved. Its hatching is now a little broader and softer, in two layers (width 2.0, spacing 10, smudge 0.35, layers 2): 4.31 in the pan and 3.10 over the whole path (was 4.52), steadier than the plain render, and it still reads as white crosshatching on cyanotype blue. Nothing else changed.

1.1.29 (2026-11-08): a polish pass. Each medium was looked at on the stage in sun, at a quarter of the light and up close, and under a sky that drives the sun. Coloured pencil: on lit, colourful ground the colour laid with the side of the pencil, and every layer laid as its average tone (ground seen at a low angle, a calmed shadow), lay as flat printed colour; both are now grained by the paper's tooth, the peaks taking more and the valleys less, at about the same value, so a sunlit lawn reads as pencil on paper. The rest read as their medium at both ends (charcoal smoky and rich, ballpoint crisp, conte warm on tan) and were left as they were. Defaults: pencilSketch on cold press with HB stays the starting point; under a stock sky that drives the sun it reads as well as under the guide's own sun. New: technicalPenLine (Line), technicalHatchShade (Tone) and storybookFill (Fill), and the looks boldInkSketch, technicalSketch, blueprint and storybookPencil, each one StyleKit.applyLook and each swapping like the rest. Example game guide: a "Before you start" section says which sky setting keeps your own sun and how area tones and keeping players clear work together, in plain words (an entry has one mask: whole-picture entries are kept off players, an area entry keeps its area and is drawn over a player inside it), and every step ends with "What you should see now". The README's opening keeps only this release; earlier notes are below. Also moves to the engine's current built-in library (@builtin edb50c2), whose update rewrote small reference files inside the package (a world on an older built-in library is hidden from players). Pins styleKit 1.4.20, paperBlocks 1.1.21, watercolorBlocks 1.1.17, inkBlocks 1.0.20 and lookbookKit 1.0.34.

1.1.28 (2026-11-08): completes 1.1.27 (a contribution, pull request #6). PencilHero.paint put its group mask on every pencilBlocks Tone entry and every Fill entry, so a medium given to one area (a sphere, a depth band) spread over the whole frame as soon as a character was painted; it now masks only entries that cover the whole frame, as the example guide's player step does. The example guide's Sketch.keepPlayersClear wrote groups = { players }, which removed PencilHero's group: in a scene with a painted shopkeeper (as Step 5 suggests) entries were left masked on an emptied group, so neither the player nor the shopkeeper was kept clear. Players now go into PencilHero's group beside painted characters, every other group is kept, and a player who leaves is taken out without touching the rest. Tried live: the sky keeps the guide's sun (1.1, warm, from the south-east instead of 7.76 from the sky), a sphere-masked charcoal entry keeps its sphere, and one group holds the painted shopkeeper and the player; when the player leaves it is taken out and the shopkeeper stays.

1.1.27 (2026-11-08): two fixes to the example game guide ("From an empty world to a small pencil-sketch game"), found by following it in a fresh world. The sky now sets syncSunToLight = false: a ProceduralSky drives the scene's directional light from its time of day by default, so the guide's low warm sun (intensity 1.1) was replaced by the sky's own (about 7.8, another direction) and its lines did nothing. And Sketch.keepPlayersClear now masks only the Tone and Fill entries that cover the whole frame: it used to give every Tone and Fill entry the player group mask, which replaced the sphere of an entry added for one area of the world (a charcoal ruin, a coloured pencil garden), so that medium spread over the whole frame; an entry with a mask of its own now keeps it. Guide text only; no block, look or shader changed.

1.1.25 (2026-11-08): moves to the engine's current built-in library (@builtin f4425a5), whose update rewrote small reference files inside the package (a world on an older built-in library is hidden from players). Pins styleKit 1.4.19, paperBlocks 1.1.20, watercolorBlocks 1.1.16, inkBlocks 1.0.19 and lookbookKit 1.0.33. Nothing else changed.

1.1.24 (2026-11-08): the darkest band stays the darkest. A reviewer found that with the default screenCalm blue ballpoint and coloured pencil drew their darkest band lighter than their mid band, and that turning the calm off made them darken steadily again. The calm turned every stroke down toward a fifth (60% under a lit sun) and rubbed in a tone worth one layer of strokes; a medium that builds its darks from many light layers lost most of its value there. The calm now rubs each layer into its own average coverage instead, the way ground seen edge-on was already drawn: the stripes soften just as much, and the value stays what the strokes gave. On one material turned lit, glancing and shadow side at a fifth of the light, the old calm left the shadow band 2.9 to 3.5 lighter than with the calm off (graphite, coloured pencil, ballpoint); now the two agree within 0.4 at every light level, and the bands darken lit to glancing to shadow for every medium. The suite has a thirteenth case for this, "the darkest band stays the darkest, with the screen calm on". The dim rooms read a few levels darker than before (the room table is measured again on 1.1.24); every dim room is still drawn darker than rendered and every light off reads 0.00. The README now says honestly how repeatable the readings are: a few dark-room readings can come out 0.01 apart even between two runs in one session (a reviewer saw it; our own pairs matched), and campaign stage readings move by a few levels between stage builds in the Cookbook. Also moves to the engine's current built-in library (@builtin f39f044), whose update rewrote small reference files inside the package. Pins styleKit 1.4.18, paperBlocks 1.1.19, watercolorBlocks 1.1.15, inkBlocks 1.0.18 and lookbookKit 1.0.32. The Lookbook and block tables were measured on 1.1.8 and are unchanged; the calm only engages when most of the screen is dark, which the Lookbook reaches only at its every-light-off moment.

1.1.23 (2026-11-08): a reading pass as a newcomer. The README names each guide by its real title, says what swapping back gives with the boil running, and its known limits name every tracked issue a user can meet (#19849 long waits, #20946 the web build is checked by hand, #22425 retry after a connection reset, #22812 no zm restore on an installed package's files). The how-to guide says PencilHero keeps Fill entries (a wash) off a painted character as well as Tone entries, as it has since 1.1.15; the guides no longer carry version history in their explanations; the workflow says a painted character is kept clear of washes too. Pins styleKit 1.4.15, paperBlocks 1.1.18, watercolorBlocks 1.1.14 and lookbookKit 1.0.31 (descriptions only). No block, look or shader changed.

1.1.22 (2026-11-08): needs styleKit 1.4.14. Two suite runs in a brand-new world with only pencilBlocks 1.1.21 installed both passed, but 19 of 133 readings differed between them, the unstyled stand-in stage included. Tracking it down: the stand-in stage drew identically in every build, and so did every stack after the first, but the first stack drawn after the scene had run a look on the normal clock laid its strokes and paper grain elsewhere. styleKit writes the camera anchor its position data is stored against only when the anchor moves; a clear tore that buffer down while the runtime still remembered the last anchor, so the suite's stack, drawn again and again at one camera pose 400 m from the origin, ran on an unwritten anchor. A game hit the same thing whenever a look was cleared and put back without the camera moving. Fixed in styleKit 1.4.14. Pins styleKit 1.4.14 and lookbookKit 1.0.30. No block, look or shader changed.

1.1.21 (2026-11-08): PencilHero keeps its mask group clean. A character painted with PencilHero and then despawned without unpaint (a scene unloaded, a player who left, a test room torn down) left its id in the stack's mask group 3 for good, so the group grew by one dead id per paint; paint and unpaint now drop the ids of entities that no longer exist whenever they write the group (names are kept, since a named object can come back with a reload). Checked on the Cookbook: with a painted object despawned without unpaint, painting another left only the live one in the group, and unpainting it left the group empty. The README's "What repeats from run to run" now says plainly what still varies between engine processes, measured on two fresh-install runs of 1.1.20 (every case passed both times, 125 of 133 readings identical). Pins styleKit 1.4.13 (a released stack now removes every effect it registered) and lookbookKit 1.0.29. No block, look or shader changed.

1.1.20 (2026-11-08): moves to the engine's current built-in library (@builtin f110cb3), whose update rewrote small reference files inside the package. Pins styleKit 1.4.12, paperBlocks 1.1.17, watercolorBlocks 1.1.13, inkBlocks 1.0.17, lookbookKit 1.0.28. Nothing else changed.

1.1.19 (2026-11-08): moves to the engine's current built-in library (@builtin a29b227), whose update rewrote small reference files inside the package. Pins styleKit 1.4.11, paperBlocks 1.1.16, watercolorBlocks 1.1.12, inkBlocks 1.0.16, lookbookKit 1.0.27. Nothing else changed.

1.1.18 (2026-11-08): the suite repeats to the digit in a fresh install. Run twice in a brand-new world with only pencilBlocks installed, 1.1.17 passed 11 of 12 both times (the wide-band case skipped, as it must without the campaign stage) and 78 of its 79 readings were identical; the held-clock case's brightness read 209.35 in one run and 209.71 in the next. Without the campaign stage the suite draws on a stand-in set whose figure is the engine's humanoid, and that humanoid arrives as a player's body: a character controller that poses and idles it, and a name label above it. The campaign stage's own figure was already held still; the stand-in figure now is too (its controller and label are taken off before anything is measured), in the stage cases and in the dark-room case. The suite README says so. Also moves to the engine's current built-in library (@builtin c61f473), whose update rewrote small reference files inside the package. Pins styleKit 1.4.10, paperBlocks 1.1.15, watercolorBlocks 1.1.11, inkBlocks 1.0.15, lookbookKit 1.0.26. No block, look, shader or guide changed.

1.1.17 (2026-11-07): stage readings repeat. The suite switched off every light in the open scene for its campaign-stage cases but left the scene's sky, which lights the open-air stage too, so every stage reading followed whatever sky the open scene had: moving one scene's sky from 15:30 to 10:00 took the unstyled stage from 134 to 199, and a reviewer's two runs in one world read the unstyled stage at L 46.4 and 44.8. The stage cases now hold every sky in the open scene to one fixed day (the ProceduralSky defaults) and put it back afterwards. Two runs, the first with the scene's sky at 15:30 and the second at 10:00, then read the unstyled stage at 48.1 and 48.0 and every medium within 0.2; the stroke counts land in one of two close states (up to 3% apart) and the README now says so under "What repeats from run to run". Search: the README's opening and the package description now name the styles the kit gives in the words people search with (pencil sketch look, charcoal drawing style, colored pencil over watercolor, in both spellings) and the guide for each, and the tags carry them. Also moves to the engine's current built-in library (@builtin 7ca1f19). Pins styleKit 1.4.8, paperBlocks 1.1.14, watercolorBlocks 1.1.10, inkBlocks 1.0.14, lookbookKit 1.0.25. Passed 12 of 12 in both runs. No block, look or shader changed.

1.1.16 (2026-11-07): the README now says under its version line what is new, so the package's description on ZeroMind names it. No block, look, shader or guide changed.

1.1.15 (2026-11-07): a painted hero keeps its own strokes under a wash. PencilHero kept only pencilBlocks' Tone entries off a painted character, so a look with Fill entries repainted the hero from the picture under it: in Coloured pencil wash a painted figure came out as a smooth pale shape with no marks (it carried crisp hatching the moment the three watercolor entries were switched off). paint now keeps every Fill entry (a watercolor wash, a sketch blank, a coloured pencil underlayer) off the hero as well; checked by eye on the campaign stage with the whole look on. Every code block in the how-to guide and the three medium guides now runs when pasted on its own: 17 blocks used StyleKit without the require line, which only the first block of each guide had, so pasted alone they raised attempt to index nil. The coloured pencil guide's "Taking the wash out" block starts from the whole look, since the block before it removes two of the entries it switches off. The README's install line names starterBlocks (it arrives through lookbookKit). The note under the room table is corrected: the lamp-lit grey room has two states and a run lands in one of them, on one engine; it is not the engine build as 1.1.11 to 1.1.14 said. Checked in a brand-new world with only pencilBlocks installed: all 26 README and guide blocks run alone with no error and no refused call, and the example game guide built and played word for word (2H to coloured pencil over five pencils) and pushed. The suite passed 12 of 12 in each of the three runs on the editor's engine this round, the last two with the PencilHero change. Also moves to the engine's current built-in library (@builtin fdc9a99). Pins styleKit 1.4.7, paperBlocks 1.1.13, watercolorBlocks 1.1.9, inkBlocks 1.0.13, lookbookKit 1.0.24.

1.1.14 (2026-11-07): releases 1.1.13 (merged from a contribution) from main, and moves to the engine's current built-in library (@builtin 7e13a48), whose update rewrote small reference files inside the package. Pins styleKit 1.4.6, paperBlocks 1.1.12, watercolorBlocks 1.1.8, inkBlocks 1.0.12, lookbookKit 1.0.23. Measured on main in the Style Cookbook, the suite run twice back to back: 12 passed both times (the wide-band case runs here, on the campaign stage). The second run started from the room materials the first run's end had let the engine release, and drew the room at 158.13 (colour spread 6.022) and the dark wall at 0.64 to 0.73, as a first run does, so the in-place rebuild 1.1.13 added holds on a released material. No block, look or shader changed.

1.1.13 (2026-11-07): the suite can be run twice in one session. At the end of a run the scene swap back let the engine collect the test room's materials, and on the next run they still resolved and reported resident but drew as the magenta error pattern, so a second run failed the every-light-off case ("the test room did not draw as a lit grey room") and the dark-wall case measured the error pattern; the suite now rebuilds the room materials (through the material type's own change handler for mat.yaml, so nothing is written and no preview is re-rendered) before it builds a room. Measured in a fresh world: a run starting from the broken state passed (11 passed, 1 skipped) with the room's numbers identical to a healthy first run. The how-to guide now says "every light off" outdoors includes the sky (lighting.setSky("none")), and the suite README says the wide-band case is skipped without lookbookShowcaseKit's stage and how to run it. The example game guide now says each hand-made component, module and scene needs a README.md and a .metadata, without which zm push refuses the world. No block, look or shader changed.

1.1.11 (2026-11-07): the README's dark-room numbers say which engine they hold on. A reviewer's three fresh engines read the lamp-lit grey room brighter and more finely hatched for Ballpoint notebook (222.9, 30% stripes), Doodle margins (236.1) and Pencil sketch (21% stripes) than the table (213.4 and 16%, 224.1, 17%), the same every time. Fresh desktop test clients and the editor give the table's numbers to the hundredth, the first test client of the session (on that morning's engine program) gave the reviewer's, and the view size is not the cause (1280 x 720 reads 212.1 and 19%), so the difference is the engine build; the README now gives both and the range to expect. The lamp-lit room's stripe limit in the suite goes from 30% to 45%, so the most finely hatched look on any build measured (30%) has room to spare; the suite also fails a stack that leaves the lamp-lit room as rendered (it has drawn nothing, and the dim-room checks would pass on the plain render), and its notes record the view size each run measured at. The troubleshooting line on first frames now says exactly what to check: zero styleKit stack in the shell, whose running.effects must all read true, and #renderer.compiledShaders() in the console, which must stop growing. Also moves to the engine's current built-in library (@builtin 2809239), whose update rewrote small reference files inside the package. Pins inkBlocks 1.0.11, lookbookKit 1.0.22, paperBlocks 1.1.11, styleKit 1.4.5, watercolorBlocks 1.1.7. No block, look or shader changed.

1.1.10 (2026-11-07): checked as a newcomer meets it. In a brand-new private world the package installed in 2 s, the first example above ran word for word (swapping the shade back gave the same frame, clear() gave the frame back exactly), and the suite passed twice (11 cases each, the wide-band case skipped there because that world has no campaign stage). The GPU cost note now says what ran beside the measurement. No block, look or shader changed.

1.1.9 (2026-11-07): the README's Lookbook table, block table and pictures are measured on 1.1.8's final shaders (1.1.8 was published with the Lookbook table measured just before its screen-calm change; the numbers moved by 0.02 at most). The block table's zoom column now asks whether the zoomed marks are drawn at a pencil's width (2 to 12 px, not grown with the surface) instead of a size change of 0.25 to 4 times, which read the wide view's broad tone masses as a stroke width. No block, look or shader changed.

1.1.8 (2026-11-07): the suite gives the same answer on every run, and the README's numbers come from freshly started engines. The check that failed about one run in three was "no wide bands from a normal distance": it read a plain dark wall lit by whatever sun the open scene had, and most of its score was the rough paper's tooth raked by that sun (a lit wall with no strokes on it at all scored 15.5 against a limit of 18), so it passed or failed with the session's light, and under any fixed light it could not tell the 1.1.6 bands from the fix. It now reads the look's own marks (the drawing less the frame as rendered) over the campaign stage's back wall and cottage from the recipe camera, under the suite's own light, where the 1.1.6 vine, compressed, 4B and 6B shades read 58 to 167 and this release 21 to 42 on every run (the Charcoal study look and smudged charcoal, whose 1.1.6 readings do not stand clear of the current ones, are checked against new bands only); where that stage is not installed it is skipped and says why. The suite now reads a frame only once every style shader has compiled and the frame was drawn after the change being judged, and needs three agreeing reads of different frames (a frame of the stack before, or of a setting not yet applied, could pass before). Every measuring case holds the 3D view at 1920 x 1080 and brings its own sun and ambient light, and the campaign stage's idle animation, smoke and weathering overlays are taken off before anything is measured, so the stage is the same picture every run. The Lookbook table is measured again with lookbookKit 1.0.21, whose marks score had read the frame before the look one pixel off (every look now reads on things for the pan, orbit, dolly and zoom), and the room table on three freshly started engines, which agree to the hundredth. Every guide carries a short description that search shows. One shader change: zoomed onto a shadowed wall in daylight, every shade drew a soft grey copy of the render, because the screen calm (strokes giving way to rubbed tone when most of the screen is dark) took the strokes down to a fifth; under a lit sun it now keeps at least 60% of them, so the zoomed wall stays hatched (the Cookbook's design/shots/pencilstage/zoom_by_eye_118b.png), while a dim room or a night, with no sun lit, is calmed as before. Pins lookbookKit 1.0.21.

1.1.7 (2026-11-07): every look reads as its medium from a normal distance, charcoal masses carry no wide bands, and the README's room table is measured again. The screen calm (strokes giving way to rubbed tone when most of the screen would be dense with them) counted light tones too, so it engaged in ordinary daylight views such as the cottage from 8 m and cut every stroke to a fifth: Pencil sketch read as a pale render with outlines, Ballpoint notebook and Doodle margins as the same blue-grey render. A dim picture is still calmed as before; in a brighter one it counts dark tones only and engages from 35% of the screen, so a daylight view keeps its hatching. Inside a charcoal mass (the four charcoals and the 4B and 6B leads) each layer of strokes is laid as its average tone, and only the strongest edges keep strokes, one layer of them: several faint, near-parallel layers had beaten against each other into wavy dark bands 12 to 20 pixels apart across walls (the Charcoal study recipe's back wall and cottage). Doodle margins scribbles larger and looser (spacing 10, stroke length 70, three layers), and leaves lit faces bare. Coloured pencil over watercolor simplifies its wash fully, so brick and crate textures no longer show through, and lays five layers of pencil reaching further into the middle tones. The suite gains a case for wide bands seen from 8 m and a check that pale props do not glow in the dark-walled room, loads every look with its own options so a reading repeats from run to run, and has twelve cases. The example guide's first step prints true 52. Also moves to the engine's current built-in library (@builtin 278f02f), whose update rewrote small reference files inside the package. Pins inkBlocks 1.0.10, lookbookKit 1.0.20, paperBlocks 1.1.10, styleKit 1.4.4, watercolorBlocks 1.1.6.

1.1.6 (2026-11-07): charcoal and the soft leads no longer streak close to a dark wall. The streaks came from styleKit's shared position data (fixed in styleKit 1.4.3, see its history); no pencil block changed. The suite gains a case that reads a dark wall 90 m from the world's centre, close up through a long lens, at three camera positions along a slide, for every charcoal, the 4B and 6B leads and the Charcoal study look (limit 1.5; 3.4 to 5.4 before the fix, 0.65 to 0.78 after), so the suite has eleven cases. Also moves to the engine's current built-in library (@builtin 6656162), whose update rewrote small reference files inside the package. Pins inkBlocks 1.0.9, lookbookKit 1.0.19, paperBlocks 1.1.9, styleKit 1.4.3, watercolorBlocks 1.1.5.

1.1.5 (2026-11-07): moves to the engine's current built-in library (@builtin 808b4a1), whose update rewrote small reference files inside the package. Pins inkBlocks 1.0.8, lookbookKit 1.0.18, paperBlocks 1.1.8, styleKit 1.4.2, watercolorBlocks 1.1.4. No block, look or shader changed.

1.1.4 (2026-11-07): the moving-figure check measures the figure's own pixels (those that change when the figure is taken out of the unstyled frame). 1.1.3's mask kept every point where the unstyled scene looked the same at both camera poses, which on the stand-in set's plain ground was nearly all of them, while the strokes on that ground still moved with the camera; the check then read 10.6 against 17.6 in a fresh world and failed. On the figure alone the painted figure changes by 5.1 against 13.6 for world strokes there. No block, look or shader changed.

1.1.3 (2026-11-07): the moving-figure check holds in any world. In a fresh world, where the suite draws on its stand-in set, the ground and walls behind the figure shift with the camera and set both of the check's numbers, so the painted figure read 11.2 against 18.0 and failed the 0.6 ratio although its marks stayed on it. The check now measures only where the unstyled scene looks the same at both camera poses (the figure itself and plain background). No block, look or shader changed.

1.1.2 (2026-11-07): no zebra bands on open ground. A fresh world's plain ground plane, seen at a low angle, showed wavy zebra bands in the Charcoal study look and the soft leads: each patch of strokes squeezed into a short row and the rows of stroke ends lined up. The shade now lays each layer as its average tone on surfaces seen nearly edge-on (styleKit 1.4.1's sks_edge_on and sks_layer_mean), so the ground keeps its value as even rubbed tone and the strokes stay on the forms. On open ground the band share fell from 0.215 to 0.052 for 6B and from 0.066 to 0.006 for the Charcoal study look, with the ground's brightness within about 6 of 255. The suite's zebra case now also looks over open ground (limits 0.055 charcoal, 0.09 the leads, 0.03 the look), and the moving-figure case compares the painted figure with 4B strokes only (mass = 0), since since 1.1.0 4B lays most of its tone as masses, which gave the comparison too little to slide on the suite's stand-in set. The per-block and whole-look tables were measured on 1.1.0; 1.1.2 changes only surfaces seen at a low angle.

1.1.1 (2026-11-07): installs into any world with only the five packages it needs again (styleKit, paperBlocks, watercolorBlocks, inkBlocks, lookbookKit). The test suite builds its stage from lookbookShowcaseKit where that is installed and from a stand-in set where it is not; 1.1.0's install took that optional stage as a dependency, pulled in lookbookShowcaseKit and the ten kits its stage is made of, and was refused in a fresh world because three of those kits could not be fetched at the versions it named. The suite now names the stage's module at run time, so an install records nothing for it. No block, look or shader changed.

1.1.0 (2026-11-07): charcoal now draws like charcoal. Charcoal and the soft leads lay their tone as soft, smudged masses rubbed into the sheet, with strokes showing at the edges of the masses and only faintly inside them (the new shade setting mass: vine 0.8, compressed 0.82, white 0.6, smudged 0.92, 4B 0.7, 6B 0.65, every other variant 0 and unchanged). The Charcoal study look had shown zebra-like bands of stripes; it no longer does, because every layer of strokes at one place now shares one grid and one slant and darker layers cross at clear angles (styleKit 1.4.0), and the shade sizes its strokes on screen across their own width (sks_layer_d, from the view rays and the smooth normal), as does the PencilHero material. Dim rooms stay dim through styleKit 1.4.0's dim guard: the Ballpoint notebook look had lifted a dim room by about 16 of 255, against the promise that no look brightens a dim or unlit room by more than a few levels. With that guard in place the shade no longer caps the wash or blank it lies over at the rendered brightness, a cap that printed the render's brick and tile texture back over the watercolor wash in the Coloured pencil wash look in daylight, so that look reads as a wash again; and the sheet turns white once the scene's brightest parts reach about 0.55 (was 0.7), so a drawing of an overcast day sits on white paper. Coloured pencil lays a little more of each thing's own colour on lit coloured surfaces. The suite has ten tests: the dark-room case reads every look and both fills in six lights (a lamp in a grey room, the same lamp in a dark-walled room, dim ambient, a low lamp, no light switched on, every light off) and allows 3 of 255 in the dim and unlit rooms (was 4); a new case, "charcoal and the soft leads draw masses, never zebra bands", counts the share of the stage in a periodic band pattern and fails the Charcoal study look over 0.05, a charcoal shade over 0.11 and 4B or 6B over 0.15 (the shaders before the fix read 0.083, up to 0.149 and up to 0.191 on the same stage). Needs styleKit 1.4.0.

1.0.4 (2026-11-03): the suite's scene clean-up runs as the last step of each case, after the world's lights are back on: switching a light off and on marks it, and the scene saver then records every difference the light holds from the scene file, so 1.0.3 could still leave a clean scene with two edits after a run. The clean-up now writes the saver's pending records and then discards them, only when the scene held no unsaved edits before the run.

1.0.3 (2026-11-03): the test suite leaves the open scene as it found it in any world: a look stack it had to create is removed again, and when the scene held no unsaved edits before the run, the edits the run left are discarded (in a fresh world, 1.0.2's suite left the scene marked as edited, because the first look it applied created the scene's stack).

1.0.2 (2026-11-03): two Fill blocks, sketchBlankFill (a pale sketch blank keeping as much of the original colour as you choose) and coloredPencilUnderlayFill (a light coloured pencil first layer), on a third shader, pencilFill. The hero material hatches a figure in long parallel strokes on its smooth normal (it read as blotchy, woven cells), draws shadow sides as a middle tone, and goes black with every light off. The shade and fill blocks draw a dim scene dim (the sheet is no brighter than the scene's own lights, a wash under the strokes included) and give their strokes up for rubbed tone when most of the screen would carry them, so a dark room never fills with stripes. The suite's dark-room case builds its own closed room of six slab walls, stops if the room does not draw lit and grey (a missing material drew magenta and let the sky in before), and reads every look and both fills in a lamp-lit, a dim, an unlit-fallback and an every-light-off room. The suite's stage is temporary and brings no day clock, so a run leaves the open scene's unsaved changes as they were and the next run starts from the same state. A per-block table measures every Tone and Fill block against styleKit's sketchLine on the campaign stage.

1.0.1 (2026-11-03): the example game guide, followed word for word in a fresh world, now runs as written: Field.bool for the pencil's flag, the fourth pencil swaps in soft vine charcoal (it changed nothing before), and every player's body is kept out of the shade and the wash with a group mask (a body close to the camera read as stripes).

1.0.0 (2026-11-03): first release: six media in twenty-five variants (50 blocks on two shaders), eight whole looks, the PencilHero hero material, five guides and the pencilWorld workflow. Built on styleKit 1.3.0 (skStrokes: stroke direction and stacked strokes by tone).