What Is Article Illustrator

This skill turns written articles into article-plus-illustration sets. It is not random decoration — every illustration has a position justified by the content, a type chosen for the information shape, and a style that reinforces comprehension.

Use it when the output should be:

• a technical article with diagrams at key arguments
• a tutorial with step illustrations
• an opinion piece with poster-style visuals
• a knowledge post with infographic summaries per section

If the task is a standalone infographic (no article), use infographic instead. If the task is a comic explainer, use knowledge-comic instead.

Three-Dimension System

Every illustration decision splits into three independent axes:

Dimension	Controls	Examples
Type	Information structure	infographic, scene, flowchart, comparison, framework, timeline
Style	Rendering approach	notion, warm, blueprint, vector-illustration, screen-print
Palette	Color scheme (optional)	macaron, warm, neon — overrides style's default colors

Combine freely. Or use presets that bundle all three — see style-presets.

Do not blur these together. "Blueprint" is a style, not a type. "Flowchart" is a type, not a style.

Types

Type	Best For
infographic	Data, metrics, technical concepts
scene	Narratives, emotional content
flowchart	Processes, workflows, tutorials
comparison	Side-by-side, options, trade-offs
framework	Models, architecture, principles
timeline	History, evolution, milestones

Styles and Palettes

Read:

• styles for the full gallery and Type × Style compatibility matrix
• style-presets for preset shortcuts (type + style + palette in one flag)
• Individual style specs in references/styles/<style>.md
• Individual palette specs in references/palettes/<palette>.md

Workflow

1. Analyze the article

Before anything visual, understand the content:

Analysis	Output
Content type	Technical / Tutorial / Methodology / Narrative
Purpose	information / visualization / imagination
Core arguments	2-5 main points to visualize
Positions	Where illustrations add value (not decoration)

Critical: If the article uses metaphors (e.g., "电锯切西瓜"), visualize the underlying concept, NOT the literal image.

What to illustrate:

• core arguments (required)
• abstract concepts that benefit from visualization
• data comparisons, processes, workflows

What NOT to illustrate:

• metaphors literally
• decorative scenes that add no information
• generic stock-image-style illustrations

2. Confirm settings with the user

One round of questions, max 4. Recommend based on content analysis.

Question	Options
Preset or Type	Recommended preset, alternative preset, or manual type selection
Density	minimal (1-2), balanced (3-5), per-section, rich (6+)
Style	Skip if preset chosen. Otherwise recommend from styles
Palette	Optional. Default uses style's built-in colors. Override with macaron/warm/neon

Use the content type → preset mapping in style-presets to make the recommendation.

3. Research models

pica prompt find "infographic"
pica skill find "banana prompting"
pica model search "image generation"
pica model info <model-id>

Choose the model based on what the illustrations need. Text-heavy infographics need text-strong models. Scene illustrations need atmosphere-strong models. Do not assume one model fits all types.

4. Build the outline

For each illustration, define:

• Position: which section/paragraph it follows
• Purpose: why this illustration helps the reader
• Visual Content: what to show
• Type Application: how the chosen type applies here
• Filename: NN-{type}-{slug}.png

Present the outline to the user for confirmation before generating.

5. Construct prompts and generate

For each illustration:

1. Build the prompt using type-specific templates from prompt-construction
2. Include actual data from the article — real numbers, terms, metrics, not placeholders
3. Apply style rendering rules from references/styles/<style>.md
4. Apply palette override from references/palettes/<palette>.md if specified

Generate:

pica generate \
  --model <model-id> \
  --kind image_generation \
  --input '{
    "prompt": "<constructed prompt>"
  }'

After each generation, show the result and ask for feedback before continuing to the next illustration or iterating on the current one.

6. Finalize

Insert markdown image references into the article at the planned positions:

![description](NN-{type}-{slug}.png)

Prompt Construction Principles

Good illustration prompts must include, in this order:

1. Layout structure — composition, zones, flow direction
2. Specific data/labels — actual content from the article
3. Visual relationships — how elements connect
4. Semantic colors — meaning-based color choices
5. Style characteristics — line treatment, texture, mood
6. Aspect ratio — typically 16:9

Full templates per type: prompt-construction

Iteration Rules

When a draft fails, diagnose which layer failed:

• hierarchy wrong → rewrite layout structure
• layout family wrong → choose different type, not just reword
• labels wrong → shorten and specify text
• too crowded → reduce content or switch to denser type
• too sparse → add content or switch to sparser type
• style wrong but structure works → only change style language

Do not rewrite the whole prompt when only one layer failed.

Common Failure Modes

• describing an illustration instead of a structured layout
• omitting exact text labels and data from the article
• illustrating metaphors literally
• no clear position justification for an illustration
• chasing style before information architecture works
• using scene type for content that should be infographic
• generating all images before getting feedback on the first one

Related Skills

• pica skill find "infographic" — standalone structured infographics
• pica skill find "knowledge comic" — comic-style explainers
• pica skill find "banana prompting" — prompt architecture techniques