create-image

name: create_image description: | Creates image from user prompt. Use when the user wants to generate, create, or edit images using AI. Triggers: "create an image of", "generate a picture", "draw", "make an illustration", "visualize", "edit this image", "modify the photo", "change the background". Supports text-to-image generation and image editing with masks. NOT for image analysis or description—only for creating or modifying visual content.

Create Image

Overview

This skill generates images from natural language descriptions or edits existing images using AI image generation models. It provides a unified interface supporting multiple vendors:

• Google Gemini (default) - Fast drafts and high-quality pro output
• OpenAI - HD quality with optional mask-based editing

Quick Start

# Create output directory first
mkdir -p ./output

# Generate an image (Google, default)
python3 .claude/skills/create_image/image_gen.py "A sunset over mountains" -o ./output/sunset.png

# Generate with OpenAI
python3 .claude/skills/create_image/image_gen.py "A sunset over mountains" --vendor openai -o ./output/sunset.png

# High quality
python3 .claude/skills/create_image/image_gen.py "Detailed portrait" --hq -o ./output/portrait.png

# Edit an existing image
python3 .claude/skills/create_image/image_gen.py "Make the shirt green" --reference ./photo.jpg -o ./output/edited.png

Vendors

API Keys

CLI Reference

Basic Usage

python3 .claude/skills/create_image/image_gen.py "<prompt>" [options]

# Prompt from file
python3 .claude/skills/create_image/image_gen.py -p <prompt-file> [options]

Note: Output files should be written to ./output/ (writable workspace directory).

Options

Prompt Sources

You can provide the prompt in two ways:

1. Inline (positional argument): python3 .claude/skills/create_image/image_gen.py "your prompt here"
2. From file: python3 .claude/skills/create_image/image_gen.py -p ./prompt.txt

The file option is useful for:

• Long, detailed prompts
• Reusable prompt templates
• Multi-line prompts with formatting

Supported Aspect Ratios

1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 21:9

Programmatic API

generate_image()

from image_gen import generate_image

result = generate_image(
    prompt="A sunset over mountains",
    output_path="./output/sunset.png",  # optional
    aspect_ratio="16:9",                # default: "1:1"
    vendor="google",                    # or "openai"
    high_quality=False,                 # True for pro/HD
    model=None,                         # override default model
    api_key=None,                       # override env variable
)

edit_image()

from image_gen import edit_image

result = edit_image(
    prompt="Make the shirt green",
    reference_image="./photo.jpg",
    output_path="./output/edited.png",  # optional
    vendor="google",                    # or "openai"
    high_quality=False,                 # True for pro/HD
    model=None,                         # override default model
    mask_image=None,                    # OpenAI only: path to mask
    api_key=None,                       # override env variable
)

Return Value

Both functions return the same dictionary:

{
    "success": bool,           # True if successful
    "image_path": str | None,  # Path where image was saved
    "text": str | None,        # Text response (Google only)
    "error": str | None,       # Error message if failed
    "image": PIL.Image.Image | None,  # PIL Image object
    "model": str,              # Model that was used
    "vendor": str,             # Vendor that was used
}

Instructions for Claude

Step 1: Determine Parameters

From the user's request, extract:

• Prompt: The image description or edit instruction (required)
  • For long prompts, save to a file and use -p
• Vendor: User preference, or default to google
• Mode: Generate (no reference) or Edit (reference provided)
• Quality: Standard (default) or high (--hq)
• Aspect ratio: Based on intended use (default: 1:1)
• Output path: Where to save the image

Step 2: Choose Vendor

Step 3: Execute

# Create output directory
mkdir -p ./output

# Generate image
python3 .claude/skills/create_image/image_gen.py "<prompt>" --vendor <vendor> -r <aspect-ratio> -o ./output/<filename>.png [--hq]

# Edit image
python3 .claude/skills/create_image/image_gen.py "<prompt>" --reference ./<image> -o ./output/<filename>.png

Programmatic (via inline python):

python3 << 'EOF'
import sys
sys.path.insert(0, '.claude/skills/create_image')
from image_gen import generate_image, edit_image

result = generate_image(prompt="...", vendor="google", aspect_ratio="16:9", output_path="./output/image.png")
EOF

Step 4: Handle Result

if result["success"]:
    print(f"Saved to: {result['image_path']}")
else:
    print(f"Error: {result['error']}")

Examples

Text-to-Image Generation

mkdir -p ./output

# Inline prompt (short)
python3 .claude/skills/create_image/image_gen.py "A cartoon cat wizard" -o ./output/wizard_cat.png

# Widescreen landscape
python3 .claude/skills/create_image/image_gen.py "Mountain panorama at sunset" -r 16:9 -o ./output/mountains.png

# High quality with Google Pro
python3 .claude/skills/create_image/image_gen.py "Detailed portrait" --hq -r 3:4 -o ./output/portrait.png

# Using OpenAI
python3 .claude/skills/create_image/image_gen.py "Steampunk clockwork" --vendor openai -o ./output/steampunk.png

Image Editing

# Edit with inline prompt
python3 .claude/skills/create_image/image_gen.py "Make the background a beach" --reference ./portrait.jpg -o ./output/beach.png

# Edit with Google Pro
python3 .claude/skills/create_image/image_gen.py "Change shirt to green" --reference ./person.jpg --hq -o ./output/green.png

# Targeted edit with OpenAI mask
python3 .claude/skills/create_image/image_gen.py "Replace background with space" --vendor openai --reference ./portrait.png --mask ./bg_mask.png -o ./output/space.png

Mask Image Guidelines (OpenAI only)

For targeted image editing with OpenAI:

• Use PNG format
• Same dimensions as reference image
• Transparent areas: Where the model may change pixels
• Opaque areas: Where the original must be preserved

Error Handling