nano-banana-image-gen

name	nano-banana-image-gen
description	Generate images using the `imagen` CLI (Gemini and OpenAI models). Use when the user says "nano banana", "generate an image", "create an image", "make me a picture", "image gen", "draw", "illustrate", or wants to create images from text prompts.

Image Generation with imagen CLI

Generate images using the imagen CLI — a unified interface for Google Gemini and OpenAI image models. Supports text prompts, multiple aspect ratios, output formats, batch generation, and more.

Prerequisites

1. imagen CLI: Install via curl -fsSL https://raw.githubusercontent.com/ozten/imagen/main/scripts/install.sh | bash
2. API Keys: Configure via environment variables or config file (see below)

Quick Start

# Basic generation (uses default nano-banana / gemini-3-pro-image-preview)
imagen "A cat wearing a top hat" -o cat.jpg

# With aspect ratio
imagen "Mountain panorama" --aspect-ratio 16:9 -o mountains.jpg

# Using OpenAI model
imagen "Cyberpunk cityscape" -m gpt-1.5 -o city.png

# Multiple images
imagen "Abstract art" -n 3 -o art.jpg

# Higher resolution
imagen "Detailed landscape" --size 2K -o landscape.jpg

# Read prompt from file
imagen --prompt-file prompt.txt -o result.jpg

Models

Alias	Model ID	Provider
nano-banana (default)	gemini-3-pro-image-preview	Gemini
gpt-1.5	gpt-image-1.5	OpenAI
gpt-1	gpt-image-1	OpenAI
gpt-1-mini	gpt-image-1-mini	OpenAI

Options

Option	Values	Default	Description
-m, --model	see Models table	nano-banana	Image generation model
-a, --aspect-ratio	1:1, 16:9, 9:16, 4:3, 3:4, etc.	1:1	Output aspect ratio
-s, --size	1K, 2K, 4K	1K	Image resolution
-q, --quality	auto, low, medium, high	auto	Quality (OpenAI models only)
-f, --format	jpeg, png, webp	jpeg	Output format
-o, --output	path	auto-generated	Output file path
-n, --count	integer	1	Number of images to generate
-p, --prompt-file	path	none	Read prompt from file
-v, --verbose	flag	off	Verbose output

Workflow

If the user provides $ARGUMENTS, treat them as the image prompt. Otherwise ask what they'd like to generate.

Step 0: Check imagen is installed

Before generating, verify imagen is available:

which imagen

If imagen is NOT found:

1. Tell the user that the imagen CLI is required but not installed
2. Ask them: "Would you like me to install the imagen CLI? It's a lightweight Rust binary from https://github.com/ozten/imagen"
3. If they agree, run:

   curl -fsSL https://raw.githubusercontent.com/ozten/imagen/main/scripts/install.sh | bash
   

4. After installation, verify it worked: imagen --version

If API keys are NOT configured:

Check for API keys:

# Check env vars
echo "${GEMINI_API_KEY:+gemini_ok}" "${OPENAI_API_KEY:+openai_ok}"

# Check config file
cat ~/.config/imagen/config.toml 2>/dev/null

If no keys are found for the provider needed by the chosen model:

1. Tell the user which API key is needed (Gemini for nano-banana, OpenAI for gpt-* models)
2. Ask them to set up keys via environment variable or config file:
   • Environment variable: export GEMINI_API_KEY="your-key" or export OPENAI_API_KEY="your-key"
   • Config file (~/.config/imagen/config.toml):

     [keys]
     gemini = "your-gemini-api-key"
     openai = "your-openai-api-key"
     

3. Do NOT proceed with generation until at least one relevant key is configured

Step 1: Generate the image

Determine the prompt, output path, model, aspect ratio, and other options, then run:

imagen "PROMPT" \
    --model MODEL \
    --aspect-ratio ASPECT_RATIO \
    --size SIZE \
    --format FORMAT \
    --output OUTPUT_PATH

Only include flags that differ from defaults.

Step 2: Report results

Report the output file path and parameters used. If generation fails, check:

• API key is configured for the chosen provider
• Prompt doesn't violate content policies
• Model alias is valid

Batch Generation

For multiple images, either use -n for the same prompt or run imagen multiple times with different prompts:

# Same prompt, multiple variations
imagen "Abstract art" -n 3 -o abstract.jpg

# Different prompts in sequence
imagen "Panel 1: A sunny beach" -o panel_1.jpg
imagen "Panel 2: Same beach at sunset" -o panel_2.jpg

Integration

Other skills and pipelines can delegate image generation to this skill by invoking the imagen CLI directly.