supercorks/describe-image

Use this skill when you need to understand the content of an image file in the project, or when you need to generate a description for an image (e.g. for alt text or filename generation).

Script location: .github/skills/describe-image/

Prerequisites

• Node.js installed
• @google/genai and dotenv npm packages

For --google:

• GOOGLE_AI_API_KEY environment variable set
• If missing in current process env, the script attempts to read it from your zsh login/interactive environment (zsh -ilc)

For --llava:

• ollama installed and available in PATH
• Vision model pulled locally (default: llava)

Supported Formats

• JPEG (.jpg, .jpeg)
• PNG (.png)
• WebP (.webp)
• HEIC/HEIF (.heic, .heif)

Usage

To describe an image, run the following command:

node .github/skills/describe-image/index.js [options] <image_path>

Options

• --google Use Google Gemini (default provider)
• --llava Use local Ollama model
• --prompt "..." Custom prompt text
• --model <name> Override model (gemini-3-flash-preview for Google, llava for Ollama)
• -h, --help Show usage

Example:

node .github/skills/describe-image/index.js ./public/properties/shared/image_01.jpg
node .github/skills/describe-image/index.js --llava ./public/properties/shared/image_01.jpg
node .github/skills/describe-image/index.js --google --prompt "List visible buttons" ./public/properties/shared/image_01.jpg

Ollama install behavior

If --llava is used and ollama is not installed, the script prints installation steps (including Homebrew commands) and exits.

Output:

A modern living room with a white sectional sofa and large windows overlooking a city skyline.

The script outputs a concise 10-20 word description of the image content, focusing on the main subject and key features.