Nano Banana 2 Infographic

Create sleek, rich, non-noisy infographic prompts and review sets for Gemini image generation.

This skill uses Nano Banana 2 only. For API calls, use the live callable model ID rather than assuming the public marketing name is the exact endpoint name.

Decision Tree

What do you need to do?

• The brief is incomplete or fuzzy

Ask only for the missing essentials: topic, audience/context, must-include facts, and brand or style constraints.

• The user wants an infographic now

Prepare four review variants by default at 16:9 unless the user specified another ratio. Read references/patterns.md.

• The user wants live Gemini renders or proof that the prompt works

Read references/configuration.md, then run scripts/probegeminiimage_api.py.

• The user wants exact API syntax, model IDs, or request fields

Read references/api.md.

• The result looks noisy, text-heavy, or poster-like

Read references/gotchas.md, simplify the composition, and regenerate.

Default Operating Mode

• Offer four distinct variants by default unless the user explicitly asks for fewer.
• Default aspect ratio to 16:9.
• Use Nano Banana 2 only. Do not fall back to older image models unless the user explicitly asks.
• Render the default pack concurrently when you need live outputs fast.
• Use separate render passes for the variants instead of trusting one request to return the exact number of images requested.
• Keep visible text short: title up to 5 words, labels 1-3 words, no paragraphs in the image.
• Prefer editorial restraint over maximal detail. If a choice would add noise, cut it.

Intake Questions

Ask these only when they are not already answered:

If the user already gave the essentials, do not re-interview them. Build the variant pack immediately.

Quick Reference

Default Variant Quartet

Read references/patterns.md for the exact prompt shape and regeneration ladder.

Rendering Rules

• Say 16:9 explicitly unless the user asked for another ratio.
• Ask for a white or near-white base, restrained accents, and flat editorial graphics.
• Keep to 2-3 accent colours plus gray or white.
• Use one visual idea per image. Do not combine process, comparison, glossary, and hero illustration in the same render.
• Put the long explanation outside the image. Generate the copy first, then render only the short text that must appear.

Gotchas

1. Asking for a "detailed infographic" usually increases clutter rather than clarity. Ask for hierarchy, whitespace, and restraint instead.
2. Google documents that the model might not create the exact number of images requested. Treat the four default variants as four deliberate passes.
3. Google also documents that text generation works best when the text is decided first and then rendered into the image. Do not improvise long copy inside the image prompt.
4. If the image looks like a poster, reduce the number of panels, colors, and icon families before changing everything else.
5. When the user needs dense quantitative fidelity, hand-built charts or vector layouts may be a better fit than Gemini image generation.

Reading Guide