Nano Banana Ui Prompts

nano-banana-ui-prompts

A Claude Code skill that turns the [YouMind-OpenLab Nano Banana Pro prompt library](https://github.com/YouMind-OpenLab/awesome-nano-banana-pro-prompts) into a structured, searchable index of **80 curated UI-relevant prompts** — then teaches Claude how to adapt them for three outputs:

1. Direct image generation (Gemini / Nano Banana Pro)
2. Google Stitch MCP (generate_screen_from_text, create_design_system, edit_screens, generate_variants)
3. Coded UI (React/Tailwind/SwiftUI via Picasso's DESIGN.md flow)

The skill bridges the gap between a massive image-prompt library (designed for flat Gemini images) and real UI production work. Instead of manually skimming 300KB of markdown to find a Bento-grid prompt, Claude filters by keyword/category/id, extracts composition + palette + typography tokens, and emits a target-appropriate artifact — with author attribution preserved end-to-end.

---

TL;DR

# Install
git clone https://github.com/cryptoxinu/nano-banana-ui-prompts ~/.claude/skills/nano-banana-ui-prompts

# Try it (no Claude needed — the search CLI works standalone)
python3 ~/.claude/skills/nano-banana-ui-prompts/scripts/search.py --featured
python3 ~/.claude/skills/nano-banana-ui-prompts/scripts/search.py --query "bento glass" --format md

# Use it through Claude Code
# Ask: "Design me a Bento grid landing page for Stitch"
# The skill auto-triggers, pulls featured-2, translates to a Stitch-ready prompt.

---

Why this exists

The upstream library is fantastic but has three frictions for agentic use:

Friction	Why it's a problem for an agent
Distribution	16 language READMEs, each ~300KB. Claude can't browse efficiently.
Format	Prompts written for Gemini image generation; describe photographs, not UIs.
No structured data	Author/source/category buried in markdown — not queryable.

This skill solves all three:

• Parses