Nano Banana Skill para Claude Code
dakl/claude-nanobanana-skill
Gere e edite imagens com Gemini 2.5 Flash via gcloud ADC em Claude Code
Ver no GitHubSkill Python puro para gerar e editar imagens com Gemini 2.5 Flash Image diretamente de sessões Claude Code. Autentica automaticamente via gcloud ADC (Vertex AI) sem necessidade de chaves de API, com fallback para API key do Gemini. Implementação stdlib-only, sem dependências externas.
Instalação
npx convez-ai dakl/claude-nanobanana-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar ícones e assets visuais durante desenvolvimento em Claude Code
- Editar/condicionar imagens existentes com prompts de IA
- Criar variações de designs sem sair da sessão de código
Veredito

Ferramenta bem arquitetada que integra geração de imagens ao workflow de Claude Code com autenticação inteligente. Ideal para desenvolvedores que já usam GCP; requer setup de gcloud e Vertex AI habilitada, mas elimina fricção de credenciais.
README
A tradução para o português deste README ainda está na fila.
claude-nanobanana-skill
A Claude Code skill for generating and editing images with Google's Gemini 2.5 Flash Image ("Nano Banana") — straight from your Claude Code session.
Unlike most Nano Banana tools, this one prefers local gcloud ADC (Vertex
AI), so no API key is required if you already use the Google Cloud CLI. It
falls back to a Gemini API key if you have one. Pure Python, stdlib only — no
pip install.
Install
Copy the skill into your Claude Code skills directory:
git clone https://github.com/dakl/claude-nanobanana-skill
cp -R claude-nanobanana-skill/skills/nanobanana ~/.claude/skills/
Then just ask Claude to generate an image ("make me an icon of …") and it'll use the skill.
Auth (auto-detected, in order)
- ADC / Vertex AI (default). Uses
gcloud auth print-access-tokenagainst Vertex AI for your gcloud project (GOOGLE_CLOUD_PROJECTorgcloud config get-value project) and location (GOOGLE_CLOUD_LOCATION, defaultglobal). Requires a logged-ingcloudwith a project set and the Vertex AI API enabled:gcloud auth login gcloud config set project <your-project> gcloud services enable aiplatform.googleapis.com - API key (fallback). Set
GEMINI_API_KEY(orNANOBANANA_GEMINI_API_KEY/GOOGLE_API_KEY) — get one at https://aistudio.google.com/apikey.
No secrets are stored in this repo or by the skill; it reads credentials from your
environment / gcloud at run time.
Usage
python3 ~/.claude/skills/nanobanana/generate.py "<prompt>" --out icon.png
| Flag | Effect |
|---|---|
--out PATH | output file (default: ./nanobanana-output/<timestamp>.png) |
--n N | generate N variations (separate calls) |
--image PATH | edit/condition on an existing image (repeatable) |
--model ID | override the model (or set NANOBANANA_MODEL); default gemini-2.5-flash-image |
Aspect ratio is steered in the prompt text (e.g. "square 1:1", "16:9 wide").
# four variations
python3 ~/.claude/skills/nanobanana/generate.py "$(cat prompt.txt)" --out icon.png --n 4
# edit an existing image
python3 ~/.claude/skills/nanobanana/generate.py "make it warmer and simpler" --image icon.png --out icon-v2.png
Notes
- Output is a raster PNG (~$0.04/image on
gemini-2.5-flash-image). - Vertex requires
role: "user"on the request content; the script handles it.
Credits
Inspired by kkoppenhaver/cc-nano-banana
and intellectronica/agent-skills;
this variant adds ADC/Vertex auth and a stdlib-only Python implementation.
License
MIT — see LICENSE.