Banana

Banana is a Happycapy skill that acts as an automated AI-driven Creative Director for image generation, editing, and batch workflows, built on top of Google Gemini Nano Banana models

What Is This?

Banana is a structured image generation skill that wraps Google Gemini's image capabilities inside a disciplined creative direction pipeline. Rather than forwarding raw user text to the model, it interprets intent, selects a domain mode, and constructs an optimized prompt using a five-component formula before any generation call is made. It supports text-to-image creation, image editing, multi-turn creative sessions, batch variation workflows, and brand preset management, all through a consistent command interface.

Why Use It?

Most raw text-to-image tools produce inconsistent results because they accept vague or unstructured prompts without any editorial layer. Banana solves this by enforcing a creative direction process on every request, ensuring that prompts are specific, visually grounded, and matched to the correct model and output size. This makes it especially useful for teams or individuals who need repeatable, brand-consistent visual assets without manually crafting complex prompts each time.

How to Use It?

Step 1 - Install and Configure

Before generating any images, run the setup command to install the MCP server and connect your API key.

/banana setup

This installs the @ycse/nanobanana-mcp package and walks you through configuring your Google Gemini API credentials. Without a valid billing-enabled API key, generation requests will fail with a precondition error and the skill will not retry automatically.

Step 2 - Choose Your Command and Describe Your Idea

Once configured, use the command that matches your workflow. For a single image, use the generate command with a plain description of what you need.

/banana generate a product shot of a glass water bottle on a white marble surface

For editing an existing image, provide the file path and your instructions.

/banana edit ./assets/hero.png change the background to a warm sunset gradient

For generating multiple variations at once, use the batch command and optionally specify how many versions you want.

/banana batch minimalist logo for a coffee brand 5

For ongoing creative sessions where style and character consistency matter across multiple turns, use chat mode.

/banana chat

You can also browse the built-in prompt database for inspiration by running /banana inspire with an optional category, or manage saved brand and style presets using /banana preset list, /banana preset create, and related subcommands.

Step 3 - Review Output and Manage Costs

After each successful generation, the skill saves the image file and logs the associated cost. You can review spending at any time using the cost commands.

/banana cost summary
/banana cost today
/banana cost estimate

If a generation is blocked for safety reasons, the skill will automatically rephrase the prompt and retry up to three times, asking for your approval before each attempt. If the API returns a rate limit error, it applies exponential backoff and retries automatically. Always confirm the returned file path exists before treating a generation as complete.

When to Use It?

Reach for Banana whenever a task involves creating or modifying visual assets, including blog hero images, social media graphics, product packshots, app icons, logos, editorial illustrations, background textures, or any situation where you would otherwise open a separate image generation tool. It is also the right choice when you need multiple variations of a concept quickly, when you want to maintain consistent brand styling across a project using presets, or when you need an intelligent layer between a vague creative brief and the final image output.

Important Notes

• A Google Gemini API key with billing enabled is required before any generation will work, and the /banana setup command must be completed first.
• Avoid using terms like "8K", "masterpiece", "ultra-realistic", or "high resolution" in your descriptions, as the skill treats these as banned keywords and relies on the imageSize parameter instead.
• Brand presets are stored locally in ~/.banana/presets/ and are not synced across machines, so team members need to recreate or share preset files manually.
• The skill reads two reference files before every generation call, which adds a small overhead but ensures model selection and prompt construction remain consistent across all request types.

FAQ

Q: What does the Banana skill do on Happycapy?

The Banana skill on Happycapy acts as an automated AI-driven Creative Director for image generation, using Google Gemini models to handle text-to-image creation, image editing, batch workflows, and brand preset management. It never passes raw user text to the API but instead interprets and enhances prompts using a structured 5-component formula for better results.

Q: How do I generate an image using the Banana skill?

On Happycapy, you can trigger the Banana skill with commands like '/banana generate <your idea>' or simply describe what you need, such as 'create a logo' or 'make a banner'. The AI agent will analyze your intent, ask clarifying questions if needed, construct an optimized prompt, select the right Gemini model, and return a generated image file.

Q: Can Banana handle editing existing images, not just generating new ones?

Yes, the Banana skill supports image editing through the '/banana edit <path> <instructions>' command. This Happycapy Skills feature lets the AI agent intelligently apply your editing instructions to an existing image, making it useful for refining visuals, adjusting styles, or updating brand assets without starting from scratch.

Q: What is the batch workflow feature in the Banana skill?

The '/banana batch <idea> [N]' command lets you generate multiple variations of an image in a single session, defaulting to 3 variations if no number is specified. This Happycapy Skills capability is ideal for A/B testing visuals, exploring creative directions, or producing a set of assets quickly using the underlying AI agent powered by Google Gemini.

Q: Does Banana support brand presets for consistent visual styles?

Yes, Happycapy's Banana skill includes a preset management system accessible via '/banana preset [list|create|show|delete]'. You can save brand colors, style preferences, and other parameters as named presets, and the AI agent will automatically apply them as defaults whenever you reference that brand, ensuring visual consistency across all generated assets.