Skill: Infographics & Data Visualization
What
Techniques for generating clear, accurate, visually compelling infographic panels and data visualizations using AI image generation. Covers chart types, visual hierarchy, Nano Banana prompting for data-heavy images, and mixed-media approaches.
Why
Infographic panels are the hardest to generate well. AI models struggle with:
- Accurate data representation (numbers, proportions, scales)
- Clean text rendering (still unreliable in most models)
- Precise geometric layouts (charts, grids, timelines)
- Balancing data density with visual clarity
These techniques maximize what AI CAN do (visual metaphor, illustrated data, comparative layouts) and minimize what it CAN'T (exact text, precise numerical accuracy).
1. What AI Can and Cannot Do for Infographics
CAN Do Well
- Visual metaphors for data (pie as actual pie, bar chart as buildings)
- Illustrated comparisons (side-by-side objects at different scales)
- Process diagrams with visual flow (arrows, sequences, stages)
- Map-style visualizations (geographic, conceptual maps)
- Icon grids (repeated symbols at different quantities)
- Conceptual timelines (visual progression through time)
- Proportional illustrations (elephant vs mouse for scale comparison)
CANNOT Do Reliably
- Exact text labels — text rendering is inconsistent
- Precise numerical values — numbers will be approximate at best
- Pixel-perfect charts — bar heights won't match exact data
- Small text at scale — illegible or garbled
Text Rendering Accuracy (Nano Banana 2, March 2026)
| Text Type | Accuracy | Notes |
|---|
| Short phrases (1-4 words) | ~98% | Brand names, titles, labels |
| Single-line text | <10% error | Headlines, short descriptions |
| Medium text (5-15 words) | ~75-80% | Captions, longer labels |
| Long text (paragraphs) | Unreliable | Frequent misspellings, dropped chars |
| Numbers in charts | Good but verify | Sometimes hallucinates specific values |
Strategy
Generate the VISUAL component with AI, add text/labels/numbers in post-production (viewer overlay, image editing, or composite). Or: lean into visual metaphor instead of traditional charts. For short labels (1-4 words), Nano Banana 2 can render them directly with ~98% accuracy — use the two-step method from the Nano Banana skill.
2. Infographic Panel Types
Replace abstract data with concrete visual comparisons the model understands.
| Data concept | Visual metaphor | Prompt approach |
|---|
| GDP comparison | Buildings of different heights representing countries | "Five buildings in a row, varying heights. Tallest (USA) 10 stories, shortest (country) 2 stories. Each building has the country's architectural style." |
| Population growth | Progressively larger crowds / expanding city | "Bird's eye view of a city expanding outward in concentric rings — 1950 core is dense, each decade adds another ring of growth" |
| Wealth inequality | Differently sized plates of food at the same table | "A long dining table. One end has an overflowing feast. The middle has a modest meal. The far end has a single cracker on a plate. Same table, same lighting." |
| Time passage | Object aging/evolving through positions | "A tree shown at 5 stages left to right: sapling, young tree, mature tree, old gnarled tree, dead stump. Ground consistent, sky shifting from dawn to dusk." |
Prompt pattern: [Concrete visual] representing [abstract data]. [Specific proportions]. [Style consistent with comic]. [No text in image — labels added in viewer].
2.2 Comparative Scale (Very Effective)
Show two or more things side by side with accurate relative sizes.
Example — healthcare spending:
"Split panel. LEFT: a single aspirin pill, tiny, on a vast white table. RIGHT: a golden palace made of pharmaceutical bottles, towering. Same perspective, same lighting, same surface — only the scale of spending differs."
Rules:
- Use the SAME environment for both comparisons (same table, same background, same lighting)
- Make the size difference dramatic and obvious
- The comparison should work WITHOUT labels
2.3 Process Flow / Sequence
Show a multi-step process as a visual journey.
Prompt pattern:
"A visual path from left to right showing [process]. Stage 1 at the left: [visual]. An arrow/path leads to Stage 2: [visual]. Then to Stage 3: [visual]. Each stage is clearly separated but connected by [visual connector]. Clean background, consistent style."
Key: Keep it to 3-5 stages maximum per panel. More than 5 gets cluttered.
2.4 Icon Grid / Unit Chart
Represent quantities through repeated visual units.
Example — mortality rates:
"100 human figures arranged in a 10x10 grid on a neutral background. 15 figures colored red (representing affected), 85 figures in grey (representing unaffected). Clean, flat style, no overlap."
Limitations: AI may not produce exactly 15 red figures. For precise numbers, describe the VISUAL PATTERN rather than expecting exact counts.
2.5 Map-Based
Geographic or conceptual maps with data overlaid as visual weight.
Prompt: "A simplified world map. [Countries/regions] glowing bright [color] with intensity proportional to [data]. [Low-data regions] are dark/muted. Clean cartographic style, no text labels, the visual intensity tells the story."
2.6 Timeline (Visual)
Prompt: "Horizontal timeline stretching left to right. At the left: [earliest event, with small illustration]. At the middle: [middle event, with illustration]. At the right: [latest event, with illustration]. A continuous line connects them. Background shifts color from [past tone] to [present tone]."
3. Nano Banana Prompting for Infographics
AVANZATO Tier (Full Control)
Infographic panels are ALWAYS AVANZATO (highest control tier from Nano Banana skill). They need maximum prompt specificity.
Structure for Infographic Prompts
[1. MEDIUM]: "Clean digital infographic illustration, flat design style"
[2. COMPOSITION]: "Split into [N] clear sections, [layout description]"
[3. DATA VISUAL]: "[What the data looks like as visual elements]"
[4. PROPORTIONS]: "[Relative sizes/quantities — be explicit]"
[5. COLOR CODING]: "[What each color represents]"
[6. STYLE]: "[Consistent with comic art style]"
[7. DEFENSIVE]: "No text labels. No numbers rendered in image. Clean background. No decorative flourishes."
Defensive Prompting (Critical for Infographics)
AI models LOVE to add decorative nonsense to infographics. Block it:
- "Do NOT add decorative borders, scrollwork, or ornamental elements"
- "Clean white/neutral background — no gradient, no texture"
- "No text — all labels will be added separately"
- "Simple and clear — prioritize readability over visual flair"
The Composite Strategy
For infographics that need accuracy:
- AI generates the visual/artistic component — the illustration, metaphor, background, character, environment
- Post-production adds data layers — text, numbers, labels, exact chart lines, annotations
Implementation in Comic Studio
The viewer can overlay text on panels. For infographic panels:
- Generate the visual base with AI
- Store labels/annotations in investigation.json metadata
- Viewer renders text overlay on top of the image
Prompt for the Visual Base
"An illustrated [subject] in [art style], with clear visual regions where [data categories] are represented by [visual differences]. Leave space for text labels — particularly [describe where labels will go]. Clean, uncluttered composition."
5. Educational Illustration Techniques
The Exploded View
Show internal structure by separating layers.
"Exploded view of [object] — outer shell separated and floating above, inner components visible and slightly separated, each layer distinct. Clean technical illustration style."
The Cutaway
Show interior while maintaining exterior context.
"Cross-section of [object/building/organism]. Left half shows the exterior intact, right half reveals the internal structure. Clean cut line, interior components labeled by color."
The Magnification Callout
Show a detail at larger scale connected to its source.
"[Full object] at center. A circular magnification callout connected by a thin line shows [detail area] enlarged 10x, revealing [what's normally invisible]."
Before/After Split
Direct visual comparison.
"Split image — LEFT side shows [before state] in [muted/grey tones]. RIGHT side shows [after state] in [vibrant/full color]. Clean vertical dividing line. Same angle, same scale."
Annotated Diagram (Anatomy, Engineering)
For technical/educational content requiring labeled parts.
"Detailed cross-section diagram of [subject]. Cutaway view showing [internal structures]. Leader lines connecting labels to each structure. Clean white background, textbook illustration style, color-coded regions with a legend box in the corner."
Flowchart / Decision Tree
Nano Banana Pro handles flowcharts well when explicitly structured.
"Design a 16:9 flowchart diagram. Topic: [process]. Layout: top-to-bottom with branching. Rounded rectangles for steps, diamonds for decisions. Green for success paths, red for error paths. Clean instructional style, white background."
6. Color Coding for Data
The Traffic Light System (Universal)
- Red: danger, negative, problem, high cost
- Yellow/amber: caution, moderate, transition
- Green: safe, positive, solution, growth
Sequential Color (For Gradients)
- Light to dark of a single hue = low to high
- "Gradient from pale blue (low values) to deep navy (high values)"
Categorical Color (For Distinct Groups)
- Use distinctly different hues for each category
- "Three groups: warm red for [A], cool blue for [B], bright green for [C]"
Prompt Integration
"Color-coded visualization: [category A] in [color A], [category B] in [color B]. The color difference is the primary way to distinguish categories — make it obvious and consistent."
7. Layout Principles for Data Panels
Visual Hierarchy
- Biggest/brightest = most important — the main data point should dominate
- Context smaller and muted — supporting data is secondary
- Empty space is data — gaps, margins, and whitespace help the eye parse
The 5-Second Rule
If the viewer can't understand the main message in 5 seconds, the infographic is too complex. For AI-generated panels, this means:
- ONE main visual comparison per panel
- Maximum 3-5 data points
- Clear visual dominance of the key finding
Grid-Based Layouts
For multi-element infographics, describe the grid:
"A 2x2 grid layout on clean background. Top-left: [element]. Top-right: [element]. Bottom-left: [element]. Bottom-right: [element]. Equal spacing, consistent style."
8. Model Allocation for Infographics
Infographic panels are ALWAYS PRO model (model_override in prompts.json). They are complex compositions that require:
- Higher spatial reasoning for layouts
- Better adherence to quantity/proportion instructions
- Cleaner geometric rendering
- More faithful color coding
Flash model produces muddy, inaccurate infographics. PRO is non-negotiable for data visualization panels.
Common Mistakes
- Expecting exact text rendering — AI text is unreliable. Design for text-free visuals, add labels in post.
- Too much data in one panel — more than 5 data points gets unreadable. Split across multiple panels.
- Using traditional chart formats — AI can't draw clean bar charts or line graphs. Use visual metaphors instead.
- No defensive prompting — AI adds decorative junk. Explicitly ban it.
- Inconsistent color coding — using red for "good" in one panel and "bad" in another. Establish and maintain a system.
- Using FLASH model — infographics need PRO's spatial reasoning. Always override.
- Describing data abstractly — "show the wealth gap" vs "one gold mountain 10 stories tall next to a single copper coin" — the model needs concrete visuals.
- Forgetting the 5-second rule — if the visual is too complex to parse quickly, simplify.