ComfyUI Extension: ComfyUI_Nano_Banana

Authored by ru4ls

Created 3 months ago

Updated 10 days ago

11 stars

A set of custom nodes for ComfyUI that leverage the Gemini 2.5 Flash Image Preview API to generate images from text prompts, single images, and multiple images.

Custom Nodes (0)

README

ComfyUI_Nano_Banana

A set of custom nodes for ComfyUI that leverage both Google Vertex AI and Google Generative AI SDK to generate images from text prompts, single images, and multiple images with configurable aspect ratios and resolutions using the Gemini Image model.

What's New

Version 6.0.1 - Fix for MALFORMED_FUNCTION_CALL Issue

This patch update fixes the MALFORMED_FUNCTION_CALL error that occurred when using the Google GenAI SDK with the gemini-3-pro-image-preview model and Automatic Function Calling (AFC).

Fixes:

Fixed Issue #12: The MALFORMED_FUNCTION_CALL error has been resolved by explicitly disabling Automatic Function Calling (AFC) configuration in the Google GenAI SDK, preventing malformed function call errors when using the gemini-3-pro-image-preview model. The update also includes enhanced error handling.

Version 6.0 - The Multi-Turn Chat, Deprecation & Interactive Image Generation Update

This major update introduces a Multi-Turn Chat node that enables conversational image generation and editing with preserved context across multiple interactions, and deprecates legacy nodes in favor of a cleaner architecture.

New Features:

Multi-Turn Chat Node! A new "Nano Banana Multi-Turn Chat" node that supports conversational image generation and editing. Maintains conversation history and allows iterative image modifications by referencing previous images as context for new generations. Includes reset functionality to start fresh conversations.

Enhanced Conversation Context The node preserves conversation flow across multiple node executions, allowing for iterative improvements and refinements to generated images. Accepts initial images to start conversations and builds upon them in subsequent turns.

Changes:

Deprecation of Legacy Nodes The NanoBanana and NanoBananaGrounding nodes are now deprecated in favor of the unified NanoBananaAIO node. All functionality from these nodes has been incorporated into the AIO node, resulting in a cleaner architecture with reduced code duplication and improved maintainability.

For a complete history of changes, see the CHANGELOG.md file.

Installation

Clone this repository into your custom_nodes folder.

cd ComfyUI/custom_nodes
git clone https://github.com/ru4ls/ComfyUI_Nano_Banana.git

Install the required dependencies:

pip install -r ComfyUI_Nano_Banana/requirements.txt

Configuration Setup

You can use either Google Generative AI API approach (simpler) or Google Vertex AI approach (more powerful) depending on your needs.

Option 1: Google Generative AI API

To use the simpler API approach, you need a Google AI API key. You can obtain one from the Google AI Studio. Please note that the Gemini API is a paid service and may incur costs.

Copy the .env.api.template file to create your .env file:
```
cp .env.api.template .env
```
Edit the .env file and replace YOUR_API_KEY with your actual API key:
```
GOOGLE_API_KEY="your-actual-api-key-here"
```

Option 2: Google Vertex AI

For access to the full functionality including the thinking process output and enhanced grounding capabilities, use the Vertex AI approach with your Google Cloud Project.

To use this you need a Google Cloud Project with the Vertex AI API enabled.

Enable the Vertex AI API: Follow the instructions in the Google Cloud documentation to enable the API for your project.
Authenticate Your Environment: This node uses Application Default Credentials (ADC) to securely authenticate with Google Cloud. Run the following gcloud command in your terminal to log in and set up your credentials. This is a one-time setup.
```
gcloud auth application-default login
```
The node authenticates directly through the installed Python libraries and does not depend on the gcloud.cmd executable being available in your system's PATH at runtime.

Create a .env file: Copy the .env.vertexai.template file to create your .env file:

cp .env.vertexai.template .env

Then edit the .env file and add your Google Cloud project details:

PROJECT_ID="your-gcp-project-id"
LOCATION="your-gcp-location"  # e.g., us-central1 (Note: gemini-3-pro models automatically use global endpoint)

Automatic Approach Detection

The system automatically detects and uses the available credentials:

If both PROJECT_ID and LOCATION are set, it uses the Vertex AI approach
If only GOOGLE_API_KEY is set, it uses the API approach
If neither is available, an error is shown

Nodes

Nano Banana (DEPRECATED)

This node is now deprecated. Please use the "Nano Banana All-in-One" node instead, which includes all the functionality of this node plus additional features like multiple image generation and grounding capabilities.

This node previously provided a flexible interface for image generation with support for multiple aspect ratios and image sizes, supporting text-to-image and image-to-image workflows with up to three reference images using the official Google Generative AI SDK. New features included model thought process visibility.

Nano Banana Grounding (DEPRECATED)

This node previously enabled image generation that was grounded in real-time Google Search results, with proper citations and source references. It allowed for fact-based image generation with verifiable information from the web.

Nano Banana All-in-One (AIO)

This unified node combines all features from the existing nodes into a single, powerful interface. It dynamically adapts its behavior based on the image_count parameter: generating a single image (like NanoBananaGrounding) or multiple images (1-10) with the same powerful grounding, search, and thinking capabilities. This is the recommended node for new workflows.

Inputs:

model_name (STRING): The Gemini model to use. Currently using: gemini-3-pro-image-preview for advanced capabilities (default: gemini-3-pro-image-preview).
prompt (STRING): The text prompt for image generation or manipulation.
image_count (INT): Number of images to generate (1-10). When set to 1, behaves like NanoBananaGrounding; when >1, generates multiple sequential images (default: 1).
use_search (BOOLEAN): Toggle to enable or disable Google Search functionality (default: True).
image_1 to image_6 (IMAGE, optional): Up to six reference images. Provide at least one image for image-to-image generation.
aspect_ratio (STRING): The output aspect ratio for the generated image. Options include: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 (default: 1:1).
image_size (STRING): The output image quality/size. Options include: 1K, 2K, 4K (default: 2K).
temperature (FLOAT, optional): Controls the creative randomness of the output. Higher values (e.g., 1.2) are more creative, lower values (e.g., 0.5) are more deterministic.

Available Aspect Ratios & Resolutions:

1:1 - 1024x1024 (square)
2:3 - 832x1248 (portrait)
3:2 - 1248x832 (landscape)
3:4 - 864x1184 (portrait)
4:3 - 1184x864 (landscape)
4:5 - 896x1152 (portrait)
5:4 - 1152x896 (landscape)
9:16 - 768x1344 (vertical/video)
16:9 - 1344x768 (horizontal/video)
21:9 - 1536x672 (ultrawide)

Outputs:

images (IMAGE): Batch of generated images (single image when image_count=1, multiple images when image_count>1).
thinking (STRING): The AI's thought process and reasoning (only available when using Vertex AI approach; shows helpful message for API users).
grounding_sources (STRING): Citation information with source URLs and search queries used to generate the response.

Note: When using the Google Generative AI API approach (as opposed to VertexAI), the thinking and grounding_sources outputs will include helpful messages about using Vertex AI for full capabilities.

Nano Banana Multi-Turn Chat

This node supports conversational image generation and editing with preserved context across multiple interactions. Maintains conversation history and allows iterative image modifications by referencing previous images as context for new generations. Includes reset functionality to start fresh conversations.

Inputs:

model_name (STRING): The Gemini model to use. Currently using: gemini-3-pro-image-preview for advanced capabilities (default: gemini-3-pro-image-preview).
prompt (STRING): The text prompt for image generation or modification based on previous conversation context.
reset_chat (BOOLEAN): Toggle to reset the conversation history and start a fresh chat session (default: False).
aspect_ratio (STRING): The output aspect ratio for the generated image. Options include: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 (default: 1:1).
image_size (STRING): The output image quality/size. Options include: 1K, 2K, 4K (default: 2K).
temperature (FLOAT): Controls the creative randomness of the output. Higher values (e.g., 1.2) are more creative, lower values (e.g., 0.5) are more deterministic (default: 1.0).
image_input (IMAGE, optional): Initial image to start the conversation with. Use this to provide an initial image for the first interaction in a conversation.

Available Aspect Ratios & Resolutions:

1:1 - 1024x1024 (square)
2:3 - 832x1248 (portrait)
3:2 - 1248x832 (landscape)
3:4 - 864x1184 (portrait)
4:3 - 1184x864 (landscape)
4:5 - 896x1152 (portrait)
5:4 - 1152x896 (landscape)
9:16 - 768x1344 (vertical/video)
16:9 - 1344x768 (horizontal/video)
21:9 - 1536x672 (ultrawide)

Outputs:

image (IMAGE): The generated image based on the current prompt and conversation context.
response_text (STRING): The AI's response text to the current prompt.
metadata (STRING): Generation metadata including finish reason and safety ratings.
chat_history (STRING): Complete conversation history with all prompts and responses.

Example Usage

Text to Image Generation (with configurable aspect ratio)

Add the NanoBanana node to your workflow.
Select your desired aspect_ratio from the dropdown (e.g., 16:9 for wide landscape, 9:16 for vertical, etc.).
Enter a prompt.
Ensure no image_ inputs are connected.
Connect the output image to a PreviewImage or SaveImage node to see the result.

Sample Prompt: "A cinematic close-up of a transparent glass chess piece (a knight) aimed at the camera. Inside the glass piece, a tiny glowing galaxy swirls with purple and gold nebulae. The background is a dimly lit library with dust motes dancing in a single shaft of volumetric golden hour light striking the chess piece, creating caustic light refractions on the wooden table."

Image Editing and Image Fusion Generation (with configurable aspect ratio and 1 to 6 reference images)

Add the NanoBanana node to your workflow.
Select your desired aspect_ratio from the dropdown (the original images will be adapted to this output aspect ratio).
Connect one or more LoadImage nodes (up to 6) to the image_1 to image_6 inputs.
Enter a prompt describing the desired changes or outcome.
Connect the output image to a PreviewImage or SaveImage node to see the result.

Sample Prompt: "A high-speed freeze-frame photograph of the glass chess knight shattering into thousands of sharp, crystalline shards. The galaxy inside is bursting outward, spilling purple and gold nebulae mist into the room. The glass fragments are suspended in mid-air, each one refracting the golden sunlight and the internal galaxy light. The wooden table is covered in glittering debris. Maintain the realistic depth of field and volumetric dust."

Grounding with Search Results Generation

Add the NanoBananaGrounding node to your workflow.
Enter a prompt that requires current data or information from the web (e.g., weather forecasts, current events, trending topics).
Toggle the use_search parameter to True to enable Google Search functionality.
Optionally connect reference images if needed.
Set the desired aspect_ratio and image_size based on your needs.
Connect the two outputs: image and grounding_sources to appropriate display nodes.
The grounding_sources output will contain citations and links to the sources used in generating the response.

Sample Prompt: "Search for and visualize the current weather forecast for the next 5 days in Jakarta in a clean, modern glass hud style with the city as a background weather chart. Add a realistic visual of what I could wear each day."

Example Workflow:

The node will perform a Google search based on your prompt
Generate an image based on the search results
List all sources and citations used in the generation process

Grounding with Search and Multi image Results Generation

Add the NanoBanana AIO node to your workflow.
Enter a prompt that requires current data or information from the web (e.g., weather forecasts, current events, trending topics) along with image variation guide (e.g., image 1 of 4 ..., image 2 of 4 ...).
Set image_count to desired number.
Toggle the use_search parameter to True to enable Google Search functionality.
Optionally connect reference images if needed.
Set the desired aspect_ratio and image_size based on your needs.
Connect the two outputs: image and grounding_sources to appropriate display nodes.
The grounding_sources output will contain citations and links to the sources used in generating the response.

Sample Prompt: "Using provided image ensure style consistency, composition and how data displayed. Search for and visualize the current weather forecast for the next 5 days in [CITY], with the city iconic spot as a background weather chart.

CITY image 1 of 4 Jakarta. image 2 of 4 Tokyo. image 3 of 4 London. image 4 of 4 Amsterdam."

Example Workflow:

The node will perform a Google search based on your prompt
Generate a set of images along with image reference style based on the search results
List all sources and citations used in the generation process

Multi-Turn Chat Conversation (with preserved context)

Add the NanoBananaMultiTurnChat node to your workflow.
Enter your initial prompt to generate the first image.
Optionally connect an initial image_input to start the conversation with a specific image.
Set your desired aspect_ratio and image_size parameters.
Execute the node to generate the initial image and response.
For subsequent interactions, use the same node instance with the same parameters but change the prompt to continue the conversation and modify the image iteratively.
Use reset_chat to start a fresh conversation when needed.
Connect the output image to a PreviewImage or SaveImage node to see the results.
The chat_history output shows the complete conversation history.

Example Workflow:

First execution: "Create an image of a clear perfume bottle sitting on a vanity"
Second execution: "Change the color of the liquid inside the glass bottle to a vibrant royal blue"
Third execution: "Extreme close-up on the glass texture and silver cap of the blue perfume bottle. The framing is cropped tightly and weighted to the left"

License

This project is licensed under the MIT License - see the LICENSE file for details.