infographic

Infographic / Data Poster Workflow

This skill handles complex visual compositions that need factual accuracy — infographics, data posters, educational visuals, timelines, comparison charts, event summaries, etc.

STEP 1 — Deep research

The infographic must contain accurate facts. Research thoroughly:

1. web_search for the topic (e.g. "Artemis 2 mission details", "climate change statistics 2025")
2. web_fetch the top 2-3 authoritative sources (NASA, WHO, official pages — not blogs)
3. Extract: key facts, numbers, dates, names, statistics, timeline events
4. Organize into a mental outline: what sections/blocks should the infographic have?

You must have at least 5-10 concrete data points before proceeding.

STEP 2 — Gather visual references

For real subjects, the image model needs visual references to render them accurately:

1. web_search for photos/images of key visual elements (spacecraft, people, logos, maps, diagrams)
2. web_fetch with saveTo to download 2-5 reference images locally
3. These will be passed as referenceImages in generate_image

Examples:

• Artemis 2 → Orion capsule photo, crew portrait, mission patch, lunar trajectory diagram
• Climate change → temperature chart, glacier comparison, carbon emissions graph
• Product launch → product photo, logo, hero image

STEP 3 — Research prompt techniques for the model

1. web_search for prompt tips specific to the generation model and output type
   • "best prompts for [model] infographic poster"
   • "how to generate data visualization with AI"
2. web_fetch at least 1 result to get proven keywords, styles, and structures
3. Extract: 3+ style keywords that work well for this type of image

STEP 4 — Style selection (if ambiguous)

If the user didn't specify a clear visual style, present 2-3 options via prompt_user:

• "Modern flat design with bold colors and icons"
• "Photorealistic collage with data overlays"
• "Minimalist editorial with large typography"

Include example descriptions so the user can judge. Skip if the user was very specific.

STEP 5 — Compose and generate

Combine everything into ONE comprehensive prompt:

• Layout: vertical poster, sections top-to-bottom, clear hierarchy
• Content: embed the researched facts/numbers directly in the prompt description
• Style: use the proven keywords from step 3 + user preference from step 4
• References: pass downloaded images as referenceImages
• Aspect ratio: typically 9:16 or 2:3 for vertical posters, 16:9 for horizontal
• Quality: always high

Call generate_image.

STEP 6 — Self-review and refine

1. Call read_file on the generated image to SEE it
2. Check: Are the facts visible? Is the layout clear? Is text readable? Any artifacts?
3. If not satisfactory, adjust the prompt and regenerate (up to 3 attempts)
4. On 3rd failure: show all attempts to the user, explain what the model struggles with

STEP 7 — Present

Call show_result with resourceType: "image" to display in the UI. Then return with file paths.