| name | consulting-article-illustrator |
| description | Generate hand-drawn "whiteboard / excalidraw"-style explainer images for articles and posts — the doodle-marker diagrams that sit inside a pillar to explain a concept (a flow, a comparison, a loop, a stack, a hub-and-spoke, a numbered framework). Made with gpt-image-2 in the brand palette: light-grey background, black marker line work + hand-lettering, a single blue accent. Plans a figure set — a hero image at the top (above the first line; doubles as the social/OG preview) plus ~1 diagram per section — and embeds them inline. Use when asked to "make an article image / diagram / explainer / infographic / hero / thumbnail", a "hand-drawn / sketch / whiteboard" visual, or to illustrate a blog post / LinkedIn article / pillar. NOT for clean HTML social graphics (use consulting-graphics) or anything animated (use consulting-hyperframes-video). |
Consulting Article Illustrator
Hand-drawn whiteboard explainer diagrams for articles — the friendly doodle-marker visuals that make
a concept legible inside a pillar post. Made with gpt-image-2, so this is a prompt + style system.
- This skill = hand-drawn, doodle, marker — concept explainers inside articles.
consulting-graphics = clean HTML→PNG social graphics (statement / stat / framework-blocks).consulting-hyperframes-video = anything animated/motion.
The look
| Role | Value | Notes |
|---|
| Background | light grey #e9ebee | soft, flat — not pure white, no gradient |
| Line work + lettering | black marker | wobbly hand-drawn boxes/arrows; legible hand-lettering |
| Accent | one blue #2a5bd0 | highlighter swashes, underlines, numbered circles, checkmarks, ticks |
Doodle vocabulary: slightly-wobbly rounded-rectangle boxes, hand-drawn solid + dashed arrows, simple
line-doodle icons (brain, gears, laptop, clock, lightbulb, magnet, file, server…), an optional friendly
mascot, lots of whitespace, flat — no photoreal, no gradients. Match the bundled samples/. Defer brand
tokens to the project DESIGN.md.
Density & placement
- Open with a hero. Every article gets a hero — the title concept as one bold metaphor (a cover, not a diagram) — at the very top, above the first line. It's the bundle's first image,
images/image1.png, and it doubles as the social/OG preview.
- Name by reading order. Every image lives in the bundle's
images/ folder, named image1.png, image2.png, … in the order it appears (hero = image1). The meaning lives in each embed's alt text, not the filename.
- Then ~1 diagram per major section/concept. A 4-section pillar → a hero + ~4 section figures. Cap section figures at ~4–6.
- Place each section figure at its break, right after that section's point lands.
- Lead with the strongest archetype (a comparison or a flow usually reads fastest).
Workflow
- Plan the figure set. Read the draft; list its major sections; per Density & placement pick a
hero (the title concept, for the top) plus the concept + archetype for each section. One idea per figure.
- Write the prompt = the style block + the archetype block, filled with the concept's real
labels (templates: references/prompt-kit.md). Keep labels short (≤3–4 words)
— gpt-image-2 garbles long text.
- Generate with gpt-image-2. Attended / local: the
GenerateImage tool in-session, or the
Higgsfield CLI (gpt-image-2 --image $REF, a samples/ image as the style reference). Unattended /
cloud pipeline: the Higgsfield MCP connector with model gpt-image-2 (the CLI's browser auth
can't run there) — don't substitute another model. Landscape ~3:2 in-article; 1:1 or 4:5 if it also
runs as a post.
- Run the bar (below); regenerate on any miss.
- Save + embed. Save every image in the bundle's
images/ folder, numbered in reading order: the
hero is images/image1.png (embed it at the very top, above the first line), then each section
figure is images/image2.png, image3.png, … embedded at its break with descriptive alt text that
carries the meaning:  (see content/AGENTS.md).
Scratch/tests → content/_work/article-images/.
Layout archetypes
| Archetype | Use for | Shape |
|---|
| Hero / title card | the lead image (top of the article) + social preview | a COVER, not a diagram: ONE central doodle (the core metaphor) + the hand-lettered TITLE with a blue underline, nothing else — no steps, lists, panels, or flow arrows. Wide 16:9, poster-like, mostly empty space |
| Flow / pipeline | a process, before→after, a path | 2–4 boxes left→right joined by arrows; dash + blue-highlight the pivotal box |
| Comparison | X vs Y, old vs new, tool vs system | vertical split; two headings (blue underline); blue-highlight pill verdict under each |
| Cycle / loop | a compounding/repeating system | 4–6 boxes in a ring, curved arrows, blue numbered circles |
| Stack / layers | a layered architecture | stacked rows + a left brace label; blue-highlight the key layer |
| Hub-and-spoke | one core + its parts | center node + satellite boxes on connectors |
| Framework | a named N-part model, inputs→output | numbered rows (blue circles) between a "messy inputs" cluster and a clean "output" |
The bar — check every render before showing
Fail any box → don't ship it; fix the prompt and regenerate.
Samples
samples/ holds approved exemplars to match: flow.png, compare.png, cycle.png.
Nightly pipeline hook
consulting-nightly-content can auto-illustrate a staged pillar: generate every image via the
Higgsfield MCP connector with model gpt-image-2 (never the in-session GenerateImage tool or a
fallback model), apply Density & placement, save into the draft's images/ as image1.png (hero) …
imageN.png, embed inline, and leave it staged for review (never auto-publish). If the MCP or gpt-image-2 is unavailable,
stage without figures and flag — don't swap models.
Taste authority: consulting-tasteful-design + DESIGN.md govern brand look across every medium;
this is their hand-drawn-explainer implementation.
