nanobanana Skill (GargantuaX): Nano Banana Images

Makes Claude generate and edit images with Gemini's Nano Banana models through one Python script.

Installs: 10.6K; on skills.sh
Stars: 7; on GitHub
Files: 6; besides SKILL.md
Updated: Apr 2026; last repo commit

nanobanana runs scripts/nanobanana.py with two subcommands: generate for one request and batch for repeated variants. It covers text-to-image, image edits with repeated local reference images, and three model aliases (nanobanana, nanobanana-2, nanobanana-pro). It validates options before sending and has a --dry-run that shows the request without sending it.

It needs Python 3.10 or newer and a GEMINI_API_KEY. The key is sent to the Gemini endpoint, or to a custom Gemini-compatible gateway if you set --base-url. The skill ships agents/openai.yaml for Codex.

Good for

Designers and developers who need images made from a text prompt
People who edit images using one or more local reference images
Teams that call Gemini through their own gateway

Install nanobanana

Claude Code

Copy this prompt.

Install the nanobanana skill in Claude Code. Run this command:

pnpm dlx skills add https://github.com/GargantuaX/openskills --skill nanobanana

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js, pnpm.

The repo is a two-skill collection (gpt-image-2 and nanobanana). The commands here install nanobanana only. The README also documents installing the whole collection with pnpm dlx skills add GargantuaX/openskills. The skill runs a Python script, so you need Python 3.10 or newer and a GEMINI_API_KEY (set it as an environment variable or in a .env file; see the README). Run a --dry-run command first, as the README advises. There is no Claude app upload route, because the skill runs a local script.

From the project README. Source.

What's in the skill

When it loads

Gemini-native Nano Banana image generation and editing across Nano Banana, Nano Banana 2, and Nano Banana Pro. Use when you need text-to-image, image-to-image edits, repeated local references, batch generation, dry-run request inspection, or a custom Gemini-compatible base URL such as a self-hosted gateway.

Sections in SKILL.md

Workflow
Commands
Rules
Resources

Key facts

Author: Gargantuax (copyright holder in LICENSE)
License: MIT (LICENSE file and README)
Installs on skills.sh: 10,568 (Nov 2, 2026)
GitHub stars: 7 (Nov 2, 2026)
Last repo commit: Apr 22, 2026
What it includes: 7 files including SKILL.md: README, agents/openai.yaml, 2 reference files, 1 Python script and 1 test file.
Needs: Python 3.10+, and GEMINI_API_KEY for a Gemini-compatible generateContent endpoint
Works in: Codex (agents/openai.yaml and .codex/INSTALL.md), and skill-aware agents that read SKILL.md; the README names Codex, Claude and Cursor as skill-aware agents

What it includes

Runs one text-to-image or image-edit request
It accepts repeated --input-image paths and writes the image to --output.
Runs code · generate
Runs repeated variants with a prefix, count, delay and optional parallel workers
The default batch count is 10, per references/config.md.
Runs code · batch
Prints the final request without sending it
SKILL.md says to use it first when the risk is the payload, endpoint or model options.
Runs code · --dry-run
Saves the raw JSON response body for debugging
Write · --save-response
Points at a custom Gemini-compatible gateway and chooses header style
Auto mode sends both Authorization: Bearer and x-goog-api-key to custom endpoints.
Network · --base-url / --auth-mode
nanobanana, nanobanana-2 and nanobanana-pro, or an exact Gemini model ID
SKILL.md maps them to gemini-2.5-flash-image, gemini-3.1-flash-image-preview and gemini-3-pro-image-preview.
Read · Model aliases
Rejects --size on nanobanana and allows 512 only on nanobanana-2
Read · Option validation

From the official docs

Permissions

Medium risk

nanobanana runs a Python script that sends your prompt, reference images and Gemini API key to an endpoint. Our advice: run --dry-run first and use a gateway you trust.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Your API key goes to whatever endpoint you set
SKILL.md says that with --auth-mode auto, custom endpoints receive both Authorization: Bearer and x-goog-api-key headers. Only use a gateway you trust. github.com
It sends your reference images
The script base64-encodes each --input-image file and puts it in the request body. Do not pass private images unless you are happy to send them to the endpoint. github.com
It searches upward for .env files
config.md says it uses --env-file, or the first .env found from the working directory up through parent directories. A parent .env can supply the key. Process environment variables override .env. github.com
Batch runs make many calls and write many files
The default batch count is 10 requests, with optional parallel workers, and images go to ./nanobanana-images unless you set an output folder. Set a small --count. github.com
Search tool can be enabled
GEMINI_USE_SEARCH adds the google_search tool to requests. It is off by default. github.com
It ships a Codex agent file
agents/openai.yaml holds the display name Nano Banana, a short description and a default prompt for Codex. github.com

Technical details

Turning on read-only	No read-only switch.
Talks to	generativelanguage.googleapis.com (default), or the base URL you set
Sign-in and keys	GEMINI_API_KEY from the environment or a .env file; --api-key is available but the script help says to prefer the environment or .env.
Scopes	Claude runs python scripts/nanobanana.py Reads reference images you name and .env files Writes generated images and optional response JSON to disk Sends prompts and images to the Gemini endpoint or your custom gateway

Questions

What does the nanobanana skill do?

It makes Claude run a Python script that generates and edits images through Gemini's Nano Banana models. It supports batches, dry runs, multiple reference images and custom Gemini-compatible endpoints.

How do I install it?

Use the install box at the top. The README gives a pnpm dlx command for this one skill and another for the whole collection. You also need Python 3.10 or newer.

Do I need an API key?

Yes. The script needs GEMINI_API_KEY for a Gemini-compatible endpoint, set as an environment variable or in a .env file.

What is the difference between nanobanana, nanobanana-2 and nanobanana-pro?

SKILL.md says to prefer nanobanana-2 unless you need the fastest low-cost default (nanobanana) or the highest-fidelity reasoning model (nanobanana-pro).

Can I use my own gateway?

Yes. Pass --base-url or set GEMINI_BASE_URL to a Gemini API root such as one ending in /v1beta. Set --auth-mode bearer if your gateway expects bearer auth.

Does it work in the Claude app?

There is no upload route, because the skill runs a local Python script. The install box covers Claude Code, Codex and the skills CLI. The repo includes agents/openai.yaml for Codex.

How is it different from gpt-image-2 in the same repo?

nanobanana covers Gemini-native image models. The README describes gpt-image-2 as the OpenAI-compatible GPT Image 2 skill, with its own README and config.

Is it safe?

It runs a Python script that sends your prompt, any reference images and your API key to the endpoint you configure. Our advice: run --dry-run first and only point it at a gateway you trust.

Information

updated Nov 2, 2026

Repository: gargantuax/openskills
Folder: skills/nanobanana

gargantuax/openskills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 7

nanobanana SKILL.md · github.com
openskills README · github.com
nanobanana README · github.com
.codex/INSTALL.md · github.com
LICENSE · github.com
skills.sh listing · skills.sh
config reference · github.com