nano-banana-use_skill

Overview

This skill lets you generate, edit, and compose images using Gemini Nano Banana image models via small, portable Python scripts. It supports both API key and Vertex AI authentication and exposes parameters for prompt, model, aspect ratio, and safety filter. Always confirm parameters with the user or explicitly state the defaults before running.

How this skill works

The skill runs lightweight Python scripts (invoked with uv) that call Gemini Nano Banana image endpoints (gemini-2.5-flash-image and gemini-3-pro-image-preview). Scripts accept a prompt plus options for model, aspect ratio, and safety filter, and can generate, edit, or compose images from one or more inputs. Authentication is handled either by setting an API key environment variable or by using Vertex AI environment variables if running within Google Cloud.

When to use it

• Quickly prototype image concepts from text prompts.
• Edit or retouch existing images using descriptive instructions.
• Compose a new image from multiple source images and a guiding prompt.
• Run image generation from local machines or CI systems with lightweight scripts.
• Integrate simple image creation into experiments without heavy SDK setup.

Best practices

• Always confirm or explicitly state the prompt, model, aspect-ratio, and safety-filter-level before executing a run.
• Prefer gemini-3-pro-image-preview as the default model unless you need a specific capability from gemini-2.5-flash-image.
• Set authentication via GOOGLE_API_KEY or GEMINI_API_KEY for local use; set GOOGLE_CLOUD_PROJECT, GOOGLE_CLOUD_LOCATION and GOOGLE_GENAI_USE_VERTEXAI=1 for Vertex AI.
• Use descriptive prompts and include style or composition details for predictable results.
• Choose an appropriate safety filter; default is BLOCK_MEDIUM_AND_ABOVE, but verify policy needs before lowering it.

Example use cases

• Generate concept art: "A futuristic banana city" -> save as city.png.
• Edit photos: provide original.png and prompt "Make the sky purple" -> edited.png.
• Compose stylized images: combine style.png and subject.jpg with a prompt to produce a painted composition.
• Batch-run experiments: script-driven generation across multiple prompts and aspect ratios for A/B testing.
• Local prototyping: run the scripts from a laptop using an API key environment variable.