nano-banana

name	nano-banana
description	Use this skill for AI image generation and editing powered by Google Gemini. Triggers include: any request to generate, create, draw, or make an image/picture/illustration/artwork from a text description; any request to edit, modify, transform, or alter an existing image; requests for style transfer or combining multiple reference images; requests to continue editing a previously generated image. Also use when the user mentions 'nano-banana', 'Gemini image', or asks for visual content creation. Do NOT use for screenshots, PDF rendering, chart/graph creation, or non-AI image tasks.
compatibility	{"os":["linux"],"deps":["node"]}

Nano Banana — AI Image Generation & Editing

Generate and edit images using Google Gemini 2.5 Flash via the nano-banana MCP server.

Available MCP Tools

All tools are prefixed with mcp_nano-banana_ when called. There are 6 tools:

1. mcp_nano-banana_generate_image — Generate from text

Create a new image from a text prompt.

Parameters:

Param	Type	Required	Description
prompt	string	✅	Text describing the image to generate

Example call:

{"prompt": "A serene Japanese garden with cherry blossoms at sunset, watercolor style"}

Behavior: Generates an image and saves it to ./generated_imgs/generated-[timestamp]-[id].png. Returns the file path and inline image content.

2. mcp_nano-banana_edit_image — Edit an existing image

Apply modifications to a specific image file on disk.

Parameters:

Param	Type	Required	Description
imagePath	string	✅	Full file path to the image to edit
prompt	string	✅	Description of modifications to make
referenceImages	string[]	❌	Optional array of file paths to reference images for style/content transfer

Example call:

{
  "imagePath": "./generated_imgs/generated-1234567890-abc.png",
  "prompt": "Make the sky more dramatic with orange and purple tones",
  "referenceImages": []
}

3. mcp_nano-banana_continue_editing — Continue editing last image

Continue editing the most recently generated/edited image in the session (no file path needed).

Parameters:

Param	Type	Required	Description
prompt	string	✅	Modifications/changes to make
referenceImages	string[]	❌	Optional reference images for style transfer

Example call:

{"prompt": "Add a rainbow in the background"}

4. mcp_nano-banana_get_last_image_info — Get last image info

Returns metadata about the last generated/edited image (path, file size, last modified).

Parameters: None.

5. mcp_nano-banana_configure_gemini_token — Set API key

Persist a Gemini API key to local config.

Parameters:

Param	Type	Required	Description
apiKey	string	✅	Gemini API key

6. mcp_nano-banana_get_configuration_status — Check config

Check whether the API key is configured and its source.

Parameters: None.

Workflow

Basic image generation

1. User asks for an image → call mcp_nano-banana_generate_image with a detailed prompt
2. The tool returns inline image content + saved file path
3. Send the saved file to the user via send_message with attachment_path

Iterative editing

1. Generate initial image with generate_image
2. User requests changes → use continue_editing (uses last image automatically)
3. Repeat as needed — each edit builds on the previous result

Edit existing image

1. User provides an image path → call edit_image with the path and modification prompt
2. For style transfer: include reference images in referenceImages array

Sending results to user

After any generation/edit, the image is saved to disk. Always send the file to the user:

send_message(attachment_path="/full/path/to/generated_imgs/filename.png", caption="描述")

Usage Guidance

• Prompt quality matters: Be specific about style, composition, colors, lighting. Translate vague user requests into detailed prompts.
• The model is Gemini 2.5 Flash Image: Good at photorealistic images, illustrations, and style transfer. Less suited for precise text rendering in images.
• File location: All generated images land in ./generated_imgs/ relative to the rayclaw working directory (typically /home/ubuntu/rayclaw/generated_imgs/).
• Session state: The MCP server tracks the last generated image per session, so continue_editing works across a conversation.
• API key: Configured via GEMINI_API_KEY env var in mcp.json. If the user reports auth errors, check with get_configuration_status.
• Rate limits: Gemini API has rate limits. If you get rate limit errors, wait a moment and retry.
• Image format: Output is always PNG.