openclaw

Summary

OpenClaw provides skills, configuration files, and workspace resources to extend and customize the OpenClaw AI agent framework. This workspace contains a collection of modular skills—including image generation via Google Gemini—as well as supporting scripts and documentation for configuration and usage.

Included Assets

• skills/README.md
  Overview of provided skills for OpenClaw.
• skills/gemini-image/SKILL.md
  Details about the gemini-image skill, including triggers, configuration, and usage examples.
• skills/gemini-image/scripts/generate.js
  Node.js script for generating images using Google's Gemini API. Supports text-to-image and image editing workflows.
• skills/gemini-image/references/prompts.md
  Tips and guidance on prompt engineering for effective image generation with Gemini.

How to Use

1. Configure API Access

Set your Gemini API key via environment variable or CLI flag:

• Environment variable:

  export GEMINI_API_KEY=your_api_key
  

• Or pass --key YOUR_API_KEY when running the script.

2. Generate Images

Navigate to the skills/gemini-image/scripts/ directory and run:

node generate.js "a cat sitting on a windowsill"

Supported options:

• --key Provide a Gemini API key
• --output-dir Set a custom output directory (default: skill_output/)
• --format Output format (png, jpeg, or webp)
• --size Image dimensions (default: 1024x1024)
• --input Path to an input image for editing
• --output Set explicit output filename

Example (custom JPEG output):

node generate.js "a city skyline at night" --format jpeg --output-dir ./images

3. Writing Effective Prompts

See skills/gemini-image/references/prompts.md for best practices on prompt structure, style selection, and descriptive keywords.

Notes

• Image generation is powered by Google Gemini (2.0 Flash Preview, "Nano Banana").
• The gemini-image skill supports text-to-image, configurable image size/format, and prompt-based image inpainting/outpainting (editing) using the CLI.
• Negative prompts are not supported; focus on describing the desired outcome.
• Images are saved to disk in the chosen output directory.
• For editing, use the --input flag to specify a base image, and supply a new prompt describing the intended final result.