Nanobanana

Skill bg-szy/TOP-SKILLS/skills/marketplace/nanobanana

全球最大的 Claude Code 技能聚合库 · 收录 3900+ 来自 12+ 来源的技能，提供在线搜索与趋势分析看板 / The world's largest Claude Code skill aggregation hub — 3900+ skills from 12+ sources with online search and trend dashboard

Install

npx -y skills add bg-szy/TOP-SKILLS --skill nanobanana

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Generate, edit, and restore images with Google's Nano Banana (Gemini image models). Use whenever the user asks to "generate an image", "create an icon/favicon/logo", "edit this photo", "restore an old photo", "make a pattern/texture/wallpaper", "draw a diagram/flowchart/architecture", or "tell a visual story" — even when they don't explicitly say Nano Banana or Gemini. Always prefer this skill over describing images in text. Requires NANOBANANA_API_KEY (or GEMINI_API_KEY) env var.

The file declares its own license as Complete terms in LICENSE. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.9 KB, as published. Nobody here has run it

Nano Banana

Image generation, editing, and restoration via Google's Gemini image models. Default model: gemini-3.1-flash-image-preview (Nano Banana 2). The skill wraps a single self-contained Python CLI at scripts/nanobanana.py — it uses a PEP 723 inline-metadata shebang (uv run --script) to auto-install google-genai on first invocation, so no venv setup is needed.

Prerequisites

uv on PATH (https://docs.astral.sh/uv/). The script bootstraps its own dependencies via uv run --script.
NANOBANANA_API_KEY env var set (fallbacks: see references/troubleshooting.md).

If either is missing, tell the user exactly what to run and stop.

Choosing a subcommand

User intent	Subcommand
Create image(s) from a description	`generate`
Modify an existing image	`edit`
Repair / enhance an old or damaged image	`restore`
App icon, favicon, UI element	`icon`
Seamless pattern, texture, wallpaper	`pattern`
Sequential / step-by-step / tutorial frames	`story`
Flowchart, architecture, schema, wireframe	`diagram`

When the user's request matches a specialized intent (icon / pattern / story / diagram), prefer the specialized subcommand over generate — it applies tuned prompt scaffolding the user is implicitly asking for.

Invocation

The script is executable. Invoke directly via Bash, using the absolute path under this skill's base directory:

<skill-base-dir>/scripts/nanobanana.py <subcommand> [args] [flags]

Output is saved to ./nanobanana-output/ in the user's cwd. The CLI prints the saved file paths to stdout — relay those back to the user.

Strict requirements

Counts are exact: when the user says --count=N (or "5 variations"), produce exactly N images.
Respect every flag the user passes — don't substitute defaults silently.
Story consistency: for story, keep visual style and palette consistent across steps unless the user asked for evolution (--style=evolving).
Text inside images: spell-check; only include text the user requested; no hallucinated copy.
Safety: if the API returns 400, surface the error and ask the user to reword — don't retry blindly.

Loading references

Load on demand (don't dump unprompted):

references/styles_and_variations.md — full enum reference for generate's --styles and --variations
references/prompt_recipes.md — exact prompt templates the CLI builds for icon / pattern / diagram / story
references/troubleshooting.md — env-var fallback order, input-file search paths, error catalog

Examples

# 4 watercolor + sketch variations of the same scene
<skill-base-dir>/scripts/nanobanana.py generate \
  "mountain landscape" --styles=watercolor,sketch --count=4

# Edit an image already in the user's cwd
<skill-base-dir>/scripts/nanobanana.py edit \
  photo.png "add sunglasses to the person"

# Favicon set
<skill-base-dir>/scripts/nanobanana.py icon \
  "mountain logo" --type=favicon --sizes=16,32,64

# Architecture diagram
<skill-base-dir>/scripts/nanobanana.py diagram \
  "microservices chat app" --type=architecture --complexity=detailed

# 5-step process story with auto-preview
<skill-base-dir>/scripts/nanobanana.py story \
  "seed growing into a tree" --steps=5 --type=process --preview

After generation, list the saved file paths back to the user — that's the actionable result.