What it does This repo collects real-world examples and prompts for Nano Banana and Nano Banana Pro, Google’s Gemini-2.5-flash-image based models. Think of it as a recipe book with screenshots: 52 Pro examples and at least 80 regular ones, scraped from Twitter/X and Xiaohongshu, showing what the model can generate or edit when given the right prompt.

The interesting bit The maintainers also released Nano-consistent-150k, a 150,000-sample dataset built to keep the same person’s identity across 35+ different editing tasks and instructions. That’s the unglamorous but genuinely useful part—interleaved training data where one face stays consistent while everything else changes.

Key highlights

• Covers both Nano Banana and Nano Banana Pro with separate example sections
• Use cases range from practical (auto photo retouching, PPT generation from articles, flowcharts from documents) to whimsical (Ukiyo-e flash cards, fluffy plush toys, fake TikTok screenshots)
• Includes multi-image tasks: character cloning, cross-view generation, AR info overlays, recursive images
• Dataset available on Hugging Face; repo has multilingual READMEs (CN/EN/JP/KR/ES/TR)
• Licensed CC BY 4.0

Caveats

• The README is almost entirely in Chinese; English readers need to click through to README_en.md
• Examples are crowdsourced from social media, so quality and reproducibility will vary
• No code or tooling here—this is purely a gallery and dataset pointer

Verdict Worth a scroll if you’re prompting Gemini-2.5-flash-image and want to see what’s actually possible beyond the obvious. Skip it if you need model weights, API docs, or training scripts.