A field guide to Nano Banana, written by 900 people at once

What it does An awesome-list-style gallery of prompts for Google’s Gemini image models — Nano Banana and Nano Banana Pro. Entries are gathered from X, Xiaohongshu, Discord, Reddit, and Facebook, then human-screened; each one shows the prompt text, the actual generated image, the author, and a link back to the original post. The main README is Chinese with an English mirror at README_EN.md, and everything is browsable at aiart.pics. Coverage spans 20+ areas: photography, illustration, 3D rendering, UI design, posters, infographics.

The interesting bit The value is the pairing — exact words next to the exact output, which turns prompt-engineering folklore into something you can study rather than believe. Some entries are parameterized templates: drop in {city_name} and {current_weather} and you get a tilt-shift ‘city micro slice’ of anywhere. It also stitches the Chinese and English prompt-sharing communities together, two worlds that barely overlap. (A few prompts still carry Midjourney flags like --ar — read before you paste.)

Key highlights

• 3,200+ curated cases from 900+ creators, refreshed daily via automated collection with human screening
• Every entry pairs the raw prompt with the real generated image, plus author and source attribution
• Bilingual — Chinese-first, with a full English README
• Ranges from one-liners (‘Make an infographic that explains how Tesla FSD works’) to multi-tool workflows chaining Nano Banana Pro with Kling 2.6 and Freepik Spaces
• Licensed CC BY 4.0

Caveats

• It’s curation, not software — no code or tooling lives in the repo, and the collection pipeline itself is invisible
• The counts wobble: 3,000+ in one paragraph, 3,200+ in another, and an update note claiming 5,233 new prompts
• Legal footing is soft — content is harvested from social media creators, the README says it’s for learning only and will delete on request, yet the repo wears a CC BY 4.0 badge

Verdict If you prompt Gemini’s image models, or you study how phrasing maps to output, this is a genuinely useful corpus. If you wanted a tool, a structured dataset, or a commercially safe asset library, keep scrolling.