Editorial Flatfolk

/editorial-flatfolk — locked editorial flat-folk illustration

This skill produces one square editorial illustration in a locked hand-drawn naive book-illustration style. The line work, color medium, surface texture, and overall feel are constants. The caller only dials five creative knobs (palette, scene metaphor, perspective, cast, light / weather).

The aesthetic sits in the contemporary editorial / picture-book / pop-folk space — flat saturated color blocks with loose hand-drawn ink line work sitting ON TOP of the color, subtle paper grain across the canvas, and a charming slightly imperfect hand. Adjacent to Pol Mara, Klaus Haapaniemi, and Carson Ellis editorial work. Not anime, not vector, not photoreal, not soft watercolor wash, not flat-vector poster.

Required model

This style is locked to gpt-image-1.5 (OpenAI image generation through whichever provider the executing tool routes to). The combination of flat saturated color, loose hand-drawn ink overlay, paper grain, and naive architectural depth requires this specific model; it does not survive on harder-line vector models or softer painterly models.

Required generation parameters:

• Model: gpt-image-1.5
• Size: 1024x1024 (square)
• Quality: high — the paper grain and ink hatching need resolution to read
• Mode: image-to-image with ref-harbor.png (the locked style anchor that lives next to this SKILL.md) passed as the style reference
• Style: caller must opt out of stacked style injection (the executing tool should pass --skip-style or its equivalent — the SKILL itself is the style)

These are semantic parameters. The executing tool decides the invocation mechanics (CLI command, REST call, gateway, local runtime).

Reference anchors

Three reference images live alongside this SKILL.md:

• ref-harbor.png — the canonical style anchor (coastal-peach palette, harbor metaphor, golden-hour light, solo wanderer cast, one-point dock perspective). This is the locked anchor every generation must use for image-to-image style hold.
• ref-alpine.png — clean output from the validation batch: winter-cool palette, alpine-village metaphor, snow-midday light, solo child cast, one-point sloped lane perspective.
• ref-park.png — clean output from the validation batch: autumn-rust palette, park metaphor, midday light, seated character cast, low-angle hero perspective.

ref-harbor.png is the required style reference. The other two are exemplars showing how the dials move within the locked style.

Locked style axes (NEVER vary)

Color medium

• Flat, saturated color blocks. Each shape is one solid color.
• No airbrush gradients, no soft pastel haze, no internal form modeling, no 3D rendering inside shapes.
• A faint vertical sky gradient is acceptable (warm-to-cool low horizon); everywhere else stays flat.

Line work

• Loose hand-drawn black ink lines that sit ON TOP of the color blocks — not as outlines that perfectly trace the shape's perimeter.
• Lines wobble slightly, occasionally cross or fall short of a color edge, never look ruler-straight.
• Line weight is fairly uniform; this is not a brush-pen calligraphic style.

Surface texture

• Subtle paper grain across the entire canvas. Visible but quiet.
• The grain is the ONLY texture allowed; nothing inside any shape should be brushy, impressionistic, or gouache-thick.

Sky treatment

• Simple ink sun-ray hatching radiating from the sun, OR ink rain ticks for wet scenes, OR ink-tick snowflakes for cold scenes.
• Cloud forms are drawn as simple soft humps with a single optional outline.

Composition

• A clear depth axis. Default is one-point perspective with two flanking walls of color receding to a vanishing point.
• Tall narrow architecture lines the receding walls when the scene metaphor allows (row houses, harbor-front shops, alpine chalets, market stalls).
• The scene metaphor IS the composition; do not retrofit a generic urban template with relabeled props (see Pitfalls).

Mood

• Editorial, picture-book, naive, observational. Calm and warm by default; pivots cleanly into rainy, snowy, or twilight tones via the Light dial without changing the underlying flatness.

Dialable knobs (vary per illustration)

1. Palette

Pick one. Always name 4–6 specific anchor colors in the prompt.

• warm-primary (default) — mustard yellow / tomato red / cobalt blue / leaf green / terracotta / cream sky.
• coastal-peach — soft peach / coral / cobalt / mustard / terracotta / rose-gold sky.
• jewel-night — deep indigo / magenta / jade green / lantern yellow / plum / charcoal.
• winter-cool — cream snow / pale icy blue / evergreen / plum / terracotta-red roofs / mustard window frames.
• autumn-rust — burnt orange / ochre / plum / sage green / deep brick red / cream sky.
• monsoon-twilight — slate blue / teal / dusty rose / mustard / cream / charcoal.

2. Scene metaphor

The scene IS the metaphor — pick a place that itself embodies the idea. Examples that respect the style:

• street — tall narrow row houses lining a cobblestone lane.
• harbor — wooden dock between two rows of fishing boats with colored hulls, harbor-front shops behind (the canonical anchor).
• market — covered stalls lining a market street, awnings, hanging lanterns.
• transit — a small train platform, a tram stop, a ferry terminal.
• park — a clearing with a bench, fountain, or stone bridge; tall trees as the flanking depth elements.
• square — a town plaza with a clock tower or fountain centered, shopfronts framing the edges.
• interior — a small café, library reading room, or shop interior with tall narrow windows.
• alpine-village — chalet houses with steep snow-capped roofs along a sloped lane, evergreens flanking.

If the scene metaphor is a pure-nature scene with no buildings, FORCE one architectural anchor element (a bench, a stone bridge, a wooden bandstand, a fountain). Otherwise the style softens — see Pitfalls.

3. Perspective

• one-point street (default) — straight depth axis with two flanking walls of architecture receding to a vanishing point.
• frontal facade — flat parallel composition, one wall of buildings filling the frame.
• low-angle hero — looking slightly up at a central figure or structure framed by trees / buildings.
• isometric — for top-down map-like scenes; the style still holds but loses the receding-rhythm feel.
• top-down — bird's eye map view; use sparingly, only for explicit "map" scenes.

Default to one-point street unless the brief explicitly requests another. Composition without a clear depth axis reads flat — see Pitfalls.

4. Cast

• solo wanderer (default) — a single figure walking, cycling, or standing in the scene, typically rear- or three-quarter-view.
• paired — two figures, often sharing an umbrella, walking together, or facing each other.
• small crowd — 4–8 background figures plus one foreground anchor figure.
• none — no figures, only architecture and atmospheric elements.

Figures are drawn small-to-medium relative to the frame; they anchor the scale of the architecture, they are not portraits.

5. Light / weather

• midday sun (default) — warm cream sky with simple ink sun-ray hatching.
• golden hour — warm peach / rose sky, low sun, longer ink sun rays.
• rain — wet cobblestone reflections, ink rain ticks, NO atmospheric haze — see Pitfalls.
• snow — pale icy sky, ink-tick snowflakes, snow caps on architecture.
• blue hour — deep blue sky with one or two ink-line clouds, lit windows glowing.

Pitfalls (lock these in the prompt)

These are validated failure modes from the calibration batch:

1. Dark + wet drift. At the dark-and-rainy extreme, the model drifts toward impressionist atmospheric texture (Van Gogh nightscape territory). The prompt must explicitly include flat color blocks, no atmospheric texture, no impressionist brushwork, no soft shading whenever Light is rain or blue hour.
2. Nature scenes soften. Pure forest / pure park / pure field scenes lose the style. If the scene metaphor has no architecture, FORCE one architectural anchor element into the prompt (bench, fountain, stone bridge, bandstand).
3. Flat composition. Without a clear depth axis the result reads vector-flat. Default to one-point perspective; if the brief asks for low-angle or frontal, explicitly call out a strong vertical-line rhythm to keep the depth feel.

Prompt template

Given a brief, assemble this prompt and pass it to the model together with ref-harbor.png as the image-to-image style reference:

Editorial flat-vector illustration in EXACTLY the same hand-drawn naive book-illustration style as the reference image: flat saturated color blocks, loose imperfect hand-drawn ink line overlay, subtle paper grain texture, organic edges, no rendering, no soft shading, no atmospheric texture, no impressionist brushwork.

SCENE: {SCENE_DESCRIPTION_WITH_ARCHITECTURAL_ANCHOR}.

PALETTE: {4–6 anchor colors from the chosen palette dial}.

PERSPECTIVE: {one-point street / frontal facade / low-angle hero / etc.} — emphasize a clear depth axis.

CAST: {solo wanderer / paired / small crowd / none}; figures small-to-medium, no portrait close-ups.

LIGHT: {midday sun / golden hour / rain / snow / blue hour} — render sky with simple ink hatching (sun rays / rain ticks / snowflakes).

No text, no logo, no watermark.

How to evaluate the output

A clean render satisfies all of:

• Color reads as flat solid blocks, not airbrushed or painted.
• Black ink lines sit on top of the color and occasionally cross or fall short of a color edge.
• Paper grain is visible across the whole canvas; nothing inside any shape is brushy.
• The scene has a clear depth axis.
• Architecture (or the forced architectural anchor) is present and reads as the scale of the place.
• Sky uses simple ink hatching (sun rays / rain / snow) — no soft cloud rendering.

If the result drifts toward soft watercolor, oil-paint impasto, anime, or clean vector, the style failed and the prompt should be re-run with the Pitfall-lock language reinforced.