grsai-nano-banana_skill

Overview

This skill integrates the grsai nano-banana image models to generate high-quality images from text prompts or from input images. It supports multiple nano-banana model variants and configurable resolution, aspect ratio, and output paths. The skill is designed for automated command-line use and conversational workflows with asynchronous polling for long-running jobs.

How this skill works

The skill submits generation tasks to grsai's nano-banana endpoints and returns a task id. It then polls the result endpoint until the image is ready, downloads the output to a configured directory, and names files using a timestamp + model + description pattern. It accepts text prompts for text-to-image and an input-image URL for image-to-image transformations.

When to use it

• You need high-quality images from descriptive text prompts.
• You want to convert or stylize an existing image (image-to-image).
• You require programmatic generation via CLI or automation scripts.
• You need configurable resolution and aspect ratios for different outputs.
• You want asynchronous generation with automatic polling and saving.

Best practices

• Store the grsai API key in openclaw.json or an environment variable rather than embedding it in scripts.
• Choose the model that matches your target resolution (e.g., nano-banana-pro for higher res).
• Start with concise, descriptive prompts and iterate—avoid sensitive or policy-violating terms.
• Use image-to-image for stylistic edits and include a clear reference image URL.
• Set reasonable polling intervals and retry limits to handle peak-hour delays.

Example use cases

• Generate a 1:1 avatar by prompt: "hand-drawn cute Shiba Inu portrait" and save to ./generated.
• Convert a photo to oil painting style by providing an input-image URL and the "nano-banana-pro" model.
• Batch-generate social media visuals at 2K for multiple prompts using a shell script and the CLI wrapper.
• Automate product mockups by feeding reference product photos and prompting for backgrounds and lighting.
• Quickly prototype scene concepts by iterating prompts and model/resolution settings via conversational assistant.