Extensions/ComfyUI-NanoBanana_node
ComfyUI Extension

ComfyUI-NanoBanana_node

A custom node for ComfyUI that allows you to interact with Nano Banana's Vertex API proxy, providing access to Gemini models with dynamic image inputs.

By darrell-goh·Created 8 months ago·Updated 6 months ago· 1
darrell-goh/ComfyUI-NanoBanana_node
Nodes2
On cloudLocal install
CategoryLLM
Stars1
Updated6 months ago

Nodes (2)

Nano Banana (Pro) Node with Multiple Outputs
LLM
Nano Banana (Pro) Node
LLM
Readme

ComfyUI Nano Banana Node

A custom node for ComfyUI that provides seamless integration with Google Vertex AI REST API format, featuring dynamic image inputs, chat mode functionality, and parallel image generation.

This project is derived from the design and structural patterns of the ComfyUI-Openrouter_node project by @gabe-init, from which elements such as dynamic image input handling and general node architecture principles were adapted. The original work is licensed under the MIT Licence, and its licence is included in this repository in accordance with its terms.

This project is an independent implementation focused solely on Nano Banana and Vertex AI integration and does not provide OpenRouter functionality.

Features

  • Two Specialized Nodes
    • Nano Banana (Pro) Node - Single output with chat mode and full feature set
    • Nano Banana (Pro) Node Multiple Outputs - Generate up to 5 variations in parallel
  • Dynamic Image Inputs - Automatically adds new image input slots as you connect images
    • Up to 14 images for Gemini 3 Pro Image models
    • Up to 6 images for Gemini 2.5 Flash Image models
  • Image Generation Support - Generate images with Gemini models through Vertex API with customizable resolution (1K/2K/4K) and aspect ratios
  • Parallel Generation - Create multiple image variations simultaneously with configurable temperature control
  • Chat Mode - Maintain conversation context across multiple messages with automatic session management (single output node only)
  • Multi-Image Support - Send multiple images in a single request to supported Gemini models
  • Environment Configuration - Uses .env file with python-dotenv for easy setup and credential management

Installation

  1. Clone this repository into your ComfyUI custom_nodes folder:
cd ComfyUI/custom_nodes
git clone https://github.com/gabe-init/ComfyUI-Openrouter_node
  1. Install dependencies using uv (recommended) or pip:
pip install -r requirements.txt
  1. Copy .env.template to .env and configure:
cp .env.template .env

  1. Edit .env with your configuration:

    For own proxy (simple endpoint):

    VERTEX_AI_API_KEY=your_api_key_here
VERTEX_AI_USE_SIMPLE_ENDPOINT=true
VERTEX_AI_ENDPOINT=https://your-nano-banana-endpoint.com
VERTEX_AI_MODELS=gemini-2.5-flash-image,gemini-3-pro-image-preview

    For direct Vertex AI access (standard endpoint):

    VERTEX_AI_API_KEY=your_vertex_ai_api_key
VERTEX_AI_USE_SIMPLE_ENDPOINT=false
VERTEX_AI_PROJECT=your-gcp-project-id
VERTEX_AI_LOCATION=your-endpoint-location
VERTEX_AI_MODELS=gemini-2.5-flash-image,gemini-3-pro-image-preview

  2. Restart ComfyUI

Usage

The Nano Banana nodes provide a simple interface to interact with Gemini models through the Vertex AI proxy service.

Nano Banana (Pro) Node

Standard node for single image generation with chat mode support.

Required Inputs:

  • system_prompt: The system prompt that sets the behavior of the LLM
  • user_message_box: The user message to send to the model
  • model: The Gemini model to use (configured via environment variables)
  • image_generation: Enable image generation for supported models
  • resolution: Image output resolution - 1K or 2K or 4K
  • aspect_ratio: Image aspect ratio - None (API default), Auto (calculate from input image), 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 5:4, 4:5, 21:9. When set to Auto, the aspect ratio is calculated from the first input image. When set to None, the parameter is omitted.
  • temperature: Controls the randomness of output (0.0 to 1.0)
  • timeout: Request timeout in seconds (30-600s, default: 300s) - increase for slower responses or 4K generation
  • chat_mode: Enable conversation mode to maintain context across messages

Optional Inputs:

  • image_1 through image_N: Dynamic image inputs that automatically appear as you connect images
    • Maximum 14 images for Gemini 3 Pro Image models
    • Maximum 6 images for Gemini 2.5 Flash Image models

Outputs:

  • STRING (Text Output): The text response from the model
  • IMAGE: An image tensor if the response contains a generated image
  • Stats: Token usage statistics (TPS, prompt tokens, completion tokens, temperature, model)
  • Thoughts: Model's reasoning/thinking content (if supported by the model)

Nano Banana (Pro) Node Multiple Outputs

Generate multiple image variations in parallel with a single execution.

Required Inputs:

  • system_prompt: The system prompt that sets the behavior of the LLM
  • user_message_box: The user message(s) to send to the model
  • model: The Gemini model to use
  • num_outputs: Number of parallel outputs to generate (1-5, default: 2)
  • use_same_prompt: Use the same prompt for all outputs (for variations) or separate prompts
  • prompt_separator: Delimiter to separate multiple prompts (default: ---)
  • image_generation: Enable image generation for supported models
  • resolution: Image output resolution - 2K or 4K
  • aspect_ratio: Image aspect ratio - None (API default), Auto (calculate from input image), 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 5:4, 4:5, 21:9. When set to Auto, the aspect ratio is calculated from the first input image. When set to None, the parameter is omitted.
  • temperature: Controls randomness (0.0 to 0.95) - automatically limited to 0.95 when using same prompt for variations
  • timeout: Request timeout in seconds (30-600s, default: 300s)

Optional Inputs:

  • image_1 through image_N: Dynamic image inputs (same limits as single output node)

Outputs:

  • IMAGE_1 through IMAGE_5: Up to 5 parallel generated images
  • Stats: Combined statistics for all parallel requests
  • Thoughts: Combined reasoning/thinking content from all parallel requests (if supported)

Usage Modes:

Variation Mode (use_same_prompt = True):

  • Enter one prompt in user_message_box
  • Set num_outputs to desired number of variations
  • Temperature automatically capped at 0.95 for diversity
  • Example: "Create a sunset scene" → generates 5 different sunset variations

Multiple Prompts Mode (use_same_prompt = False):

  • Enter multiple prompts separated by --- (or custom separator)
  • Each output uses a different prompt
  • Example: 
    Create a sunset scene over the ocean with vibrant colors.
---
Create a peaceful forest scene with morning mist.
---
Create a futuristic cityscape at night.

Examples

Image Generation

  1. Set a system prompt
  2. Enter a generation prompt (e.g., "Generate a beautiful sunset over mountains")
  3. Enable the "image_generation" option
  4. Select resolution (1K or 2K or 4K)
  5. Select aspect ratio (e.g., 16:9 for landscape, 9:16 for portrait, 1:1 for square, or Auto to let the API decide)
  6. Select an image-capable model (e.g., "gemini-3-pro-image-preview")
  7. Run the workflow
  8. The generated image will appear in the "image" output

Note: Resolution and aspect ratio for output

  • When Auto is selected, the API determines the aspect ratio automatically
  • 2K 1:1 = 2048×2048 pixels
  • 4K 16:9 = 4096×2304 pixels
  • See Nano Banana documentation for complete resolution table

Image Edit

  1. Connect an image output from another node to the "image_1" input
  2. Set a system prompt describing the edit style
  3. Enter an edit prompt (e.g., "Change the sky to purple and add clouds")
  4. Enable the "image_generation" option
  5. Select an image-capable model
  6. Run the workflow
  7. The edited image will appear in the "image" output

Image Combine

  1. Connect multiple image outputs to "image_1", "image_2", etc.
  2. Set a system prompt describing how to combine the images
  3. Enter a combination prompt (e.g., "Merge these images into a single cohesive composition")
  4. Enable the "image_generation" option
  5. Select an image-capable model
  6. Run the workflow
  7. The combined image will appear in the "image" output

Chat Mode

Chat Mode allows you to maintain conversation context across multiple messages:

  1. Enable Chat Mode: Toggle "chat_mode" to True
  2. Automatic Session Management:
    • Sessions are created automatically when you start a conversation
    • Continue the same session if sending messages within 1 hour
    • After 1 hour of inactivity, a new session is created
  3. Session Storage: Conversations are stored in a chats folder with timestamps

Managing Chat Sessions:

Use the included manage_chats.py utility:

# List all chat sessions
python manage_chats.py list

# View a specific session
python manage_chats.py view session_name

# Export a session to different formats (json, txt, md)
python manage_chats.py export session_name -f md -o output.md

# Clean up old sessions
python manage_chats.py clean -d 30

Configuration

Environment Variables

  • VERTEX_AI_API_KEY: Your Vertex API key (required)
  • VERTEX_AI_USE_SIMPLE_ENDPOINT: Set to true for own proxy
  • VERTEX_AI_ENDPOINT: Your Vertex AI endpoint URL (only activated for simple endpoint)
  • VERTEX_AI_MODELS: Comma-separated list of available models
  • VERTEX_AI_PROJECT: Google Cloud project ID (only for direct Vertex AI access)
  • VERTEX_AI_LOCATION: Google Cloud region (only for direct Vertex AI access)

Supported Models

Configure your available models in the .env file.

  • Common Gemini models include:
    • gemini-3-pro-image-preview - Advanced image capabilities (14 input images max)
    • gemini-2.5-flash-image - Fast image generation and understanding (6 input images max)
    • Other Gemini models as supported by your Vertex AI instance

Troubleshooting

  • Connection errors: Verify your VERTEX_AI_ENDPOINT and VERTEX_AI_API_KEY in .env
  • Model not available: Check that the model is included in VERTEX_AI_MODELS
  • Image generation not working:
    • Ensure you've enabled image_generation
    • Use a compatible model like gemini-3-pro-image-preview
    • Check resolution and aspect ratio settings
    • Review console output for [NanoBanana] debug messages
  • Too many input images error: Check model limits (6 for Flash, 14 for Pro)
  • Multiple outputs not generating variations:
    • Ensure temperature is below 1.0 (automatically set to 0.95 in variation mode)
    • Try using use_same_prompt = True for variations
  • Chat mode issues: Check that the chats folder has write permissions
  • Environment variables not loading: Ensure .env file is in the node directory and python-dotenv is installed

Version History

v1.3.0 (Current)

  • Usage metadata display: Processing time, token counts (output, thoughts, total) displayed on node during execution
  • Per-instance logging: Each node instance logs to its own file in logs/ directory with detailed execution data
  • Thoughts output: New output for model reasoning/thinking content (when supported by the model)
  • Parallel execution tracking: Logs individual requests and summary for parallel generation
  • Added 1K option for resolution

v1.2.1

  • Auto aspect ratio calculation: When set to Auto, aspect ratio is now calculated from the first input image using logarithmic comparison for accurate matching
  • On-node display: Calculated aspect ratio is displayed on the node during execution (e.g., "Auto → 16:9")
  • None option: Added None option to let the API decide the aspect ratio
  • Improved robustness of aspect ratio calculation with support for multiple input formats

v1.2.0

  • Added Nano Banana (Pro) Node Multiple Outputs for parallel generation
  • Support for up to 5 parallel image generations with configurable prompts
  • Temperature automatically capped at 0.95 when using same prompt for variations
  • Added Auto option for aspect ratio (lets API decide)
  • Model-specific image input limits (14 for Gemini 3 Pro, 6 for Gemini 2.5 Flash)
  • Dynamic output configuration using MAX_OUTPUTS constant

v1.1.0

  • Added resolution options (2K/4K) for image generation
  • Added aspect ratio selection (1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 5:4, 4:5, 21:9)
  • Added timeout configuration (30-600 seconds)

v1.0.0

  • Initial release
  • Dynamic image inputs (up to 10 images)
  • Chat mode with automatic session management
  • Temperature control
  • Environment-based configuration with .env file
  • Integration with Google Vertex AI REST API

License

MIT License - See LICENSE file for details

Credits

  • ComfyUI - The amazing ComfyUI framework
  • @gabe-init – Original OpenRouter node design that inspired the dynamic input and node structure