Illustrations

Illustrations are Visualist’s illustrated voice: the flat-plane style that carries the brand where a photograph would be too literal and a screenshot too cold. There are five families, and they are not interchangeable. Each has a job and a pass/fail bar. This chapter is the source of truth for all of them — what each is, when to reach for it, how it is generated, and how it is judged.

The five families

The organizing question is always who or what is depicted — it decides what the illustration is and where it starts.

Three of them depict a person and must not be confused: Figure is an archetype commanding a world; Portrait is a real individual rendered whole; Avatar is a small identity mark. One illustration family per surface — a Figure does not share a composition with a Scene or a Backdrop. It has all the space, or it is not there.

How it is built

Every family is constructed the same way. This grammar is what makes a drawing read as Visualist and not as generic vector art; learn it before briefing or judging any illustration.

• Flat-plane construction. Cubist-influenced: each form is built from confident overlapping flat geometric planes — deliberately angular and faceted, a few tonal planes per element (three to four per garment area on a figure; two to three per object in a scene), each a flat matte fill. The faceting is the look — but it stays flat and 2D. It is not a 3D render: no rendered lighting, no depth or perspective shading, no glossy or plastic surfaces, no origami-style folded volume.
• Flat matte colour, hard edges. Every colour field is a single flat matte fill meeting the next at a hard edge. No gradients, no shading within a plane, no rendered lighting, no drop shadows, no glow, no texture overlays.
• No text. Illustrations carry no lettering of any kind — no titles, prices, labels, captions, signage, or UI copy inside the image. If words are needed, they belong on the surface around the illustration, never rendered into it.
• Shape economy. Enough planes to read as considered, few enough to stay clean — roughly twelve to fifteen flat shapes in a Figure composition, two to three planes per object in a Scene. More reads mesh-like; fewer reads primitive. Avatars are the deliberate exception: displayed small, they run a denser, finer faceted mosaic (see the Avatar family).
• Palette only. Every colour is named in the Visualist system — the neutrals (Cotton, Parchment, Soft Brew, Brew, Strong Brew, Charcoal, Leather), the three persona cores (Picardy, Wolfe, Gretna, each with a soft and strong step), and the accents (Malibu, Amalfi, Byron, Goa, Cap Ferrat, Burgundy). The full palette with values is in color. No pure black, no pure white, no off-palette colour — if a colour can’t be named, it doesn’t belong.
• Palette setting. One of three: neutrals only (just the neutrals, no persona colour), core (the persona’s core colour on the neutrals), or core with accents (core plus the accent palette). The brief names which.
• Modernist Western lineage. The 1920s–40s European modernist-poster tradition — Cassandre, Paul Colin, Tom Eckersley, Patrick Nagel — extended by Bauhaus, Memphis, and mid-century studio still-life for object work, applied to a contemporary 2026 subject.
• Delivery. Figure, Avatar, and Portrait deliver on a transparent background, placed on whatever surface needs them. Scene and Backdrop may carry a grounded field, because the field is the point.

The generation process

The method is one pass through one script: the producer writes a small spec — subject, composition, family, colours — and build-prompt.mjs assembles the outgoing prompt, which goes to the image model with the starred references attached.

• Figure, Scene, and Backdrop are claim-first. They make an argument, so the spec’s subject leads with the claim — the one sentence on what the image says — before any composition or object choice. A Backdrop’s claim is atmospheric (what the space says) and stays behind the content it grounds, but it is stated first all the same. When a human wants to work the concept out conversationally, the optional Architect interview develops the brief first; producers with the surface in hand write the claim directly.
• Avatar and Portrait are direct. They carry no metaphorical claim: the subject simply names the person, the family’s fixed parameters, and the surface. An avatar is “this persona, head-and-shoulders, her core colour, transparent”; a portrait is “this photo, in our style, likeness kept.”

Calibration (all families). Always attach, as style references, the starred illustrations of the same family from visualist-assets (the star: true gold-standard set Blair maintains). This self-maintains: as the canonical set improves, so does every new generation. Never hand-pick references or work from memory. References calibrate style only — a generation that clones a reference’s woman or mirrors its composition fails.

The prompt is built by a script, never by hand. The producer writes a small JSON spec and skills/visualist-illustrations/scripts/build-prompt.mjs assembles the outgoing prompt — injecting the fixed style blocks, translating internal keys to plain language + exact hexes, and refusing to build if owned vocabulary or required fields are wrong. --help prints the spec reference, --keys the colour and identity menus; the optional human-led Architect interview is references/architect-prompt.md. This chapter owns the standard and the tests; the script owns the words. Edit the script if the wording is wrong — never paraphrase it.

Variation is part of the grammar. Consecutive same-family illustrations must differ in facing direction, in which side of the frame is occupied versus quiet, and — for a Figure — in the woman’s identity: skin tone, hair silhouette, dress. Check the recent library before briefing; a library converging on one recurring woman, one pose, or one composition side is a systemic fail even if each single image passes.

Figure

A brand persona is the fixed, commanding point of a composition; the objects of her profession expand to environmental scale and move around her; the image makes one metaphorical claim about what it feels like to be her. A Figure carries a marketing hero, a section lead, a manifesto page.

The grammar (four principles, non-negotiable).

• The fixed point. She is still, centred, commanding; everything moves around or toward her. Posture is authority settled in its own domain, never posed or aggressive.
• The activated world. Her tools don’t sit at natural scale — they expand, orbit, float, drape, or activate in response to her. A tape measure becomes a river she wields; a room assembles itself around her. Charged, not decorative.
• The metaphorical claim. Each Figure claims one thing about what she is, not what she does. Her domain doesn’t contain her: she contains it. The brief names this claim before any composition choice.
• The posture of command. Directing, presiding, stepping through, leaning in — never overwhelmed, reactive, or small. Facing direction (left, right, forward, away) is a deliberate compositional choice that varies across the library — never a default.

Character. Persona identity by colour, consistent across the library — Peyton (stylist) in Picardy orange, Indigo (interior designer) in Wolfe purple, Emery (planner) in Gretna green; a Figure in the wrong colour for her vertical breaks the system. Proportions are lightly stylised, fashion-illustration adjacent but grounded — a person, not an avatar. Across the library, figures vary in skin tone, hair, and dress — consecutive Figures must not repeat the same woman (the identity menus and variation check live in the producer skill); the face itself is never rendered — it is always the featureless Cassandre geometry (see Construction), so a figure reads as a specific person through posture and proportion, not through eyes, nose, or mouth.

Construction. She is contemporary, never period — present-day designer clothing (The Row, Khaite, Toteme, Jil Sander), never a hat. Her age, colouring, and dress follow the subject the illustration is for; a present-day professional is only the default when the brief doesn’t specify. The face is Cassandre profile geometry: one flat plane for the face mass, one sharp diagonal shadow plane for the far side of the nose and cheek, a triangular sliver at the neck — no rendered eyes, nose, or mouth. Hands are faceted geometric planes that articulate a gesture (holding, directing, presiding), never dangling passively; one may rest on a hip, a surface, or in a pocket when the working hand clearly dominates. She stands in confident contrapposto — weight on one leg, the opposite hip raised, shoulders counter-rotated, a slight three-quarter turn — an editorial magazine stance, not a runway or corporate pose.

Where it appears. Marketing heroes and long-form leads; the most expressive social posts; email headers (one, carrying the emotional register); product onboarding and empty states. Never sharing a surface with another illustration, never as background texture, never cropped, never where the product itself must be shown.

Passes only if — the tests. All must hold.

• Claim. You can state in one sentence what it claims about the professional. “She’s at her desk” is a portrait, not a Figure. Fail.
• Command. She is the fixed, commanding point, in relationship with her activated world — not a bystander adjacent to it.
• Vertical. Her colour is correct for her persona and the objects are from the right professional world.
• Medium. Flat-plane construction, flat matte colour, hard edges, no gradient/glow/shadow/texture, no text. It reads as illustration, not a photoreal render.
• Face. The face is Cassandre geometry — a flat mass plane, a diagonal shadow plane, a neck sliver — with no rendered eyes, nose, or mouth. Any drawn facial features fail.
• Palette. Every colour names in the Visualist system.
• Generic. It could not sit unremarked in a competitor’s marketing. If it could, the claim isn’t specific enough to our ICP. Fail.

Scene

The world without a protagonist. A figureless arrangement of the objects of a professional’s world, composed so the arrangement itself makes a claim — tools at rest at the end of a day, three practices held in one frame, a working space mid-thought. It carries a blog opener or an editorial lead where a Figure would over-claim, and grounds a section where copy sits on top.

The grammar.

• The compositional claim. One sentence on what the arrangement says about the world. Objects carry the idea; if they’re only decorative, it isn’t a Scene.
• No figure. The rejection is load-bearing: a person sneaking into frame turns it into a weak Figure. No people, silhouettes, or human forms — the objects carry the whole narrative.
• Front-to-back layering. Objects hold clear depth — foreground, middle, ground plane — never a flat mesh of shapes.
• Scaffolding. The recurring compositional armature: a Soft-Wolfe halo, a diagonal Cotton plane, a Strong-Brew triangular wedge. Constant in position across a sequence, varying only in opacity per frame.
• The held-empty zone. A Scene splits into a high-density zone (roughly 30–40% of the canvas) and a held-empty content zone (60–70%) where the copy or UI will sit. Brief where the empty zone is and keep activated objects out of it — only scaffolding appears there.

Character. Density is deliberate — 4–6 objects when the Scene leads a surface, 2–3 when it backs dense content; above or below that it reads cluttered or primitive. For brand-level, all-three-personas content, prefer cross-persona objects that read across verticals (a fabric drape, a ceramic vessel, a pillar candle, an open book). An optional time-of-day register (quiet warm, cool morning, closing evening) sets mood through a single warm accent or a Cotton-and-Soft-Wolfe ground.

Avoid the common misfires: flat collage (objects sitting beside each other with no depth); objects leaking into the held-empty content zone; and uniform trios (the same shape three times in three colours — give a trio distinct silhouettes).

Where it appears. Blog and editorial openers, page-section grounds, presentation backing, scrollytelling frames. Never where a human presence is the point (that is Figure), never as repeated pattern.

Passes only if — the tests. All must hold.

• Claim. The arrangement says something statable in one sentence, beyond “some objects.”
• No figure. No person, silhouette, or human form of any kind.
• Density. Within the family’s range — not cluttered, not sparse.
• Content zone. The held-empty region is genuinely clear for overlaid copy or UI.
• Medium & palette. Flat-plane construction, flat matte colour, Visualist palette, no effects, no text.
• Generic. Reads as Visualist, not as stock editorial still-life.

Backdrop

A restrained atmospheric or architectural field that sits behind content and never competes with it — a spatial ground, an arched threshold, a horizon, a plane of light. Its job is to give a surface warmth and depth while the foreground copy or UI stays first.

The grammar (claim-first). Name five things: the claim (one sentence on what the space says — an atmosphere, a threshold, a time of day; quieter than a Scene’s claim, but stated first), the space (the architectural or atmospheric cue — a doorway, a floor, a wash of light), the density (minimal — a Backdrop is quieter and emptier than a Scene), the content zone the foreground occupies (kept clear and legible), and the palette ground (a muted core or neutral field that recedes).

Scene vs Backdrop. A Scene’s claim is focal — the image can lead a surface; a Backdrop’s claim is atmospheric — it stays behind the content it grounds. If overlaid copy has to fight the illustration, it should have been a Backdrop.

Character. Low density, generous negative space, a muted grounded field. Architectural and spatial cues over object clutter. It may be grounded (a field, not transparent) because the field is the point.

Where it appears. Page and section grounds, hero grounds behind copy, scrollytelling frames, anywhere a surface needs depth without a subject. Never foregrounded — its claim stays atmospheric, behind the content.

Passes only if — the tests. All must hold.

• Claim. You can state in one sentence what the space says — while everything still recedes.
• Recedes. Foreground copy or UI stays clearly first; the Backdrop never competes.
• Low density. Restrained — emptier than a Scene, with real negative space.
• Legibility. Anything set over it remains readable.
• Medium & palette. Flat-plane construction, flat matte colour, Visualist palette, no effects, no text.

Avatar

A single person — a persona or a real individual — as a small head-and-shoulders mark in the Visualist style, built to stay legible at 40px. It makes no metaphorical claim; its job is recognisable identity at small size. Field Notes author bylines, testimonial and team faces, comment and profile chips.

The grammar (direct brief, no Architect). Head-and-shoulders only, centred, generous margin so nothing crops when it’s masked to a circle. Flat-plane construction and flat matte colour like everything else, but densely faceted: because an avatar is displayed small, the face, hair, and shoulders break into a fine mosaic of many small angular tonal planes — denser than the poster families — so the mark reads as crafted rather than plain at small size. Richness comes from the faceting while the silhouette stays simple: accessories and fine detail that don’t survive at 40px are removed. A persona avatar wears her core colour; a real-person avatar takes a restrained ground from the palette. Transparent background. No activated world, no scene, no claim.

Where it appears. Field Notes author avatars, testimonial and team headshots, profile and comment chips, anywhere a small identity mark is needed. Never scaled up to carry a hero — that is Figure or Portrait work.

Passes only if — the tests. All must hold.

• Legible at size. Reads cleanly at 40px; no detail that turns to mud when small.
• Faceting. A dense, fine mosaic of small flat planes across face, hair, and shoulders — it reads as crafted at small size. A few plain flat fields is a fail.
• Identity. The persona or person is recognisable — right core colour for a persona; a fair likeness for a real subject.
• Medium & palette. Flat-plane construction, flat matte colour, Visualist palette, no effects, no text.
• Restraint. No activated objects, no background scene, no claim. An avatar that reaches for a claim is over-built. Fail.

Portrait

A real, named person rendered whole in the Visualist style, likeness preserved — the in-style alternative to a photograph for a real individual. A profile subject, a guest contributor, a named practitioner we want to feel drawn into the brand rather than shown in a raw headshot or replaced by a persona.

The grammar (direct brief, no Architect). Start from a source photo and name: keep the likeness (the person stays recognisable), apply the house style (flat-plane construction, flat matte colour, the Visualist palette), set a restrained palette ground, and choose the crop (head-and-shoulders or fuller). No activated world, no metaphorical claim — a Portrait renders a person, it does not make an argument about her.

Portrait vs Figure vs Avatar. Figure is an archetype persona with an activated world and a claim. Portrait is a real individual, rendered whole, no activated world, no claim, from a photo. Avatar is the small identity mark. When the subject is a real, named person at editorial scale, it is a Portrait.

Where it appears. Field Notes profile heroes for real subjects, guest-contributor images, press and about surfaces. Never as a small chip (that is Avatar), never given an activated world (that is Figure).

Passes only if — the tests. All must hold.

• Likeness. Recognisable as the actual person from the source.
• Medium & palette. Flat-plane construction, flat matte colour, Visualist palette, no effects, no text.
• Restraint. No activated world, no metaphorical claim. It is a rendered person, not a Figure.

The library

The approved library lives in visualist-assets under illustrations/ — that repository, not this chapter, is the source of truth for what exists. Before commissioning anything new: browse it, and check the starred set first (the star: true best-in-class references, which are also the calibration images the generator attaches). If an approved asset already makes a similar claim for the same family and vertical, use it; new work is for genuine gaps, not surface variety.

The producer reviews her own work; Blair guards the brand separately. The producer runs her own review checklist (in visualist-illustrations: the family’s “passes only if” tests plus colour, construction, and variation checks) on the finished output, regenerates anything that fails or feels unsure, and lands the result directly in illustrations/, ready to wire. On her own cadence, Blair runs periodic brand audits of the landed library (visualist-asset-review) as brand guardianship: she rejects drift post-hoc with a binding critique (the asset moves to _rejected/ and is replaced) and maintains the starred calibration set. Wire illustrations/ paths only; _review/ stages the other asset types.