Sumi - Générateur d'infographies artistiques

Générez de superbes infographies à partir de n'importe quel sujet textuel en utilisant le pipeline Sumi. Claude analyse, structure et crée des prompts, puis génère l'image via l'API Gemini.

Spar Skills Guide Bot
ContenuIntermédiaire
1026/07/2026
Claude Code
#infographic#generative-art#data-visualization#sumi#creative-coding

Recommandé pour


name: sumi description: Generate artistic infographics from any topic. Runs the Sumi pipeline (analyze → structure → craft prompt → generate image) entirely within Claude Code. Use when "generate infographic", "create infographic", "sumi", "make an infographic about", or "visualize topic".

Sumi — Artistic Infographic Generator

Generate a beautiful infographic from a text topic. Claude does all the thinking (analysis, structuring, prompt crafting) using curated reference files. Only the final image generation calls an external API (Gemini).

Usage

The user provides:

  • topic (required): The subject or pasted text to turn into an infographic
  • style (optional): A style ID like ukiyo-e, bauhaus, sumi-e, etc.
  • layout (optional): A layout ID like bento-grid, hub-spoke, iceberg, etc.
  • aspect_ratio (optional): e.g. 16:9, 9:16, 1:1. Default: 16:9

If style or layout are not specified, recommend 3 combinations and let the user pick.

Reference Files

All reference files live in this skill's references/ directory:

  • references/analysis-framework.md — Framework for analyzing content
  • references/structured-content-template.md — Template for structuring content
  • references/base-prompt.md — Base prompt template for image generation
  • references/styles/{style-id}.md — Style definitions (60+ styles)
  • references/layouts/{layout-id}.md — Layout definitions (20 layouts)

IMPORTANT: Read the relevant reference files before each step. These contain the secret sauce — detailed visual instructions, color palettes, compositional patterns, and visual metaphor mappings that make the infographics excellent.

Instructions

Step 1: If style/layout not specified, recommend combinations

  1. Read references/analysis-framework.md
  2. Analyze the user's topic using the framework (just do the analysis yourself — you ARE Claude)
  3. Based on the content type, complexity, and audience, recommend 3 layout×style combinations:
    • best_match — most appropriate for the content
    • creative — unexpected, visually striking pairing
    • accessible — broadest appeal, balances clarity and aesthetics
  4. For each recommendation, give a 1-2 sentence rationale
  5. Ask the user to pick (1/2/3 or name their own)

To make good recommendations, consult the analysis framework's Content Type Classification table which maps content types to layouts and styles.

Step 2: Analyze content

Read references/analysis-framework.md and apply it to the topic. Produce a thorough analysis covering:

  • Content type classification (timeline, process, comparison, hierarchy, etc.)
  • Learning objectives (what the viewer will understand)
  • Audience analysis
  • Complexity assessment
  • Visual opportunity mapping
  • Data verbatim extraction (preserve all statistics, quotes, dates EXACTLY)

This analysis is internal working material — don't dump it on the user. Just mention you're analyzing.

Step 3: Structure content

Read references/structured-content-template.md and transform the analysis into designer-ready structured content:

  • Title and overview
  • Sections with key concepts, content points, visual elements, and text labels
  • All data points preserved verbatim
  • Clear visual element descriptions per section

Step 4: Craft the image generation prompt

This is where it all comes together. Read:

  • references/layouts/{selected-layout-id}.md — the full layout definition
  • references/styles/{selected-style-id}.md — the full style definition
  • references/base-prompt.md — the base prompt template

Then craft a rich, detailed image generation prompt with these sections:

Fixed header (one line):

Create a single {aspect_ratio} infographic in {Style Name} style using a {Layout Name} layout. All text in English. Use rich visual scenes with clear hierarchy and ample whitespace.

Layout Guidelines: Adapt the generic layout structure to this specific content. Replace placeholders with content-specific descriptions.

Style Guidelines: Include the FULL style description plus:

  • Color Palette with specific usage instructions
  • Visual Elements list
  • A Compositional Patterns table mapping content structures to style-specific compositions
  • Visual Metaphor Mappings that translate EACH content section into a concrete, style-specific visual scene
  • Typography instructions

Content: Describe each content section as a VISUAL SCENE. Each section must have:

  • A descriptive title (e.g., "CENTER HUB: The Castle")
  • A vivid description of what to DRAW — objects, characters, spatial layout, colors, mood
  • The exact text to render

Text Labels: Organize ALL text that should appear in the infographic, grouped by area: Title, Hub/Center, Section Labels, Section Content, Key Phrases, Attribution.

CRITICAL PROMPT CRAFTING RULES

  • Describe VISUAL SCENES, not abstract concepts. Tell the image generator what to DRAW.
  • Visual metaphor mappings must be SPECIFIC to the content AND style (e.g., for ukiyo-e: "Trust Signals → guardian komainu statues or a samurai standing watch")
  • Keep all source quotes and data VERBATIM — never summarize or rephrase
  • The prompt should be 1500-3000 words — rich enough for Gemini to produce a detailed infographic

Save the crafted prompt to a file for reference.

Step 5: Generate the image

The ONLY step that needs an external API call. Use the Sumi backend's image generator:

cd /Users/paolo/playground/sumi/backend && source .venv/bin/activate && python -c "
import asyncio
from sumi.engine.image_generator import generate_image

PROMPT = open('/tmp/sumi-prompt.md').read()
ASPECT_RATIO = 'ASPECT_RATIO_HERE'
OUTPUT_PATH = '/tmp/sumi-infographic.png'

async def main():
    path = await generate_image(prompt=PROMPT, output_path=OUTPUT_PATH, aspect_ratio=ASPECT_RATIO)
    print(f'Image saved: {path}')

asyncio.run(main())
"

Before running this:

  1. Write the crafted prompt to /tmp/sumi-prompt.md
  2. After generation, open the image with open <path> on macOS
  3. Tell the user the file path

Step 6: Iterate (if requested)

If the user wants to tweak:

  • Different style/layout: Re-do steps 4-5 with new references (skip analysis/structuring)
  • Prompt adjustments: Edit the saved prompt and re-run step 5
  • Content changes: Re-do from step 2

Available Styles

aged-academia, airline-travel-poster, art-nouveau, art-nouveau-mucha, atomic-age, axonometric, bauhaus, bold-graphic, botanical-illustration, chalkboard, charley-harper, claymation, constructivism, corporate-memphis, craft-handmade, cubism, daniel-clowes, de-stijl, dia-de-muertos, dr-seuss, fantasy-map, futurism, golden-age-comics, googie, ikea-manual, isometric-technical, isotype, jack-kirby, kandinsky, kawaii, keith-haring, knolling, ligne-claire, matsumoto, memphis, moebius, origami, osamu-tezuka, patent-drawing, paul-rand, pixel-art, pop-art-lichtenstein, renaissance-diagram, richard-scarry, rinpa, saul-bass, shan-shui, storybook-watercolor, studio-ghibli, subway-map, sumi-e, superflat, synthwave, technical-schematic, tibetan-thangka, treasure-map, ukiyo-e

Available Layouts

bento-grid, binary-comparison, bridge, circular-flow, comic-strip, comparison-matrix, dashboard, funnel, hierarchical-layers, hub-spoke, iceberg, isometric-map, jigsaw, linear-progression, periodic-table, story-mountain, structural-breakdown, tree-branching, venn-diagram, winding-roadmap

Skills similaires