This is a Google Gemini API integration node for ComfyUI, supporting text generation and image generation functions. With this node, you can directly use Google's Gemini 2.0 series models in your ComfyUI workflow. Special Note: Regarding the error [ERROR]API call error: 'NoneType' object has no attribute 'parts', it means that the image or prompt you uploaded violates the "Generative AI Prohibited Use Policy". Please test with general scene or product images first to ensure compliance.

Updates

2024.3.19: Updated to support multi-image processing.

Features

Supports Gemini 2.x and 3.x series models
Supports text-to-text generation
Supports image-to-text generation (image understanding)
Supports text-to-image generation (Note: this functionality is implemented in separate .py files, not directly within this node.)
Built-in proxy support for convenient use by users in China
Automatic dependency checking and installation
Comprehensive error handling and logging

Installation Method

Ensure you have ComfyUI installed.

Clone or download this repository into ComfyUI's custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI_Gemini_Expanded_API.git

Install dependencies:

cd ComfyUI_Gemini_Expanded_API
pip install -r requirements.txt

Restart ComfyUI.

How to Use

1. Configure Gemini API or Enterprise/Vertex AI

Option A(Gemini API)

First, you need to obtain a Google Gemini API key:

Visit Google AI Studio or Google Cloud Console
Create an API key (In Google Cloud Console it is recommended to set key restriction to the Generative Language API)
In ComfyUI, locate and use the Configure Gemini API Key node to enter your API key.

Option B (Gemini Enterprise Agent Platform, formerly Vertex AI)

Start by visiting Google Cloud Console Vertex AI API and enable it.
Visit the Vertex AI Model Garden and select a Google Gemini model and then "view code" button.
Read the instructions for the chosen model at "Try (Python)" for environment authentication or "Try while using express mode (Python)" for API-key authentication.
For environment authentication, set GOOGLE_CLOUD_PROJECT=your-project-id, GOOGLE_CLOUD_LOCATION=global, and GOOGLE_GENAI_USE_ENTERPRISE=true before launching ComfyUI. Existing installations may continue using GOOGLE_GENAI_USE_VERTEXAI=true; the node supports both names, with GOOGLE_GENAI_USE_ENTERPRISE taking precedence if both are set.
In ComfyUI, locate and use the Configure Gemini API Key node and set use_vertexai_env to true.
For Enterprise express mode, enter its API key and enable vertexai_express; project and location are not required for this mode.

2. Text Generation

Use the Expanded Gemini Text/Image node for text generation:

Connect the API key configuration node to the config input.
Enter your prompt text in prompt.
Adjust generation parameters (temperature, top_p, top_k, etc.).
If needed, connect images through image_inputs. For video, pass a native ComfyUI video through Configure Gemini Video Input, then connect its output to video.
Configured video is normalized to MP4/H.264 and sent with embedded audio at the selected sampling FPS. pad_at_start optionally prepends black video and silence. duration_aware_padding derives that padding from the video's fractional-second duration and rounds it down to native video frames for comparison testing. The normalized video must be smaller than 100 MB.
Enable use_cache on the configuration node to reuse unchanged prompt, system instruction, image, and video context through Gemini after the local result cache misses.

3. Proxy Settings

If you are in China or other regions requiring a proxy:

Set use_proxy to True.
Set proxy_host (default is 127.0.0.1).
Set proxy_port (default is 7890).

Parameter Description

API Key Configuration Node

api_key: Google Gemini API key
use_cache: Enable Gemini API context caching. Disabled by default.
cache_ttl_minutes: Number of minutes Gemini retains cached context.
cache_seed: Generation seed used while use_cache is enabled.

Video Input Configuration Node

video: Native ComfyUI video input.
fps: Gemini sampling rate from 1 through 24 FPS.
pad_at_start: Prepend matching-resolution black video and silence for half one sampling interval before encoding.
duration_aware_padding: With start padding enabled, use the video's fractional-second duration rounded down to native video frames instead of the standard padding duration.

Text Generation Node

Required Parameters

config: API key configuration
prompt: Prompt text
model: Select model (gemini-2.0-flash, gemini-2.0-flash-exp, gemini-2.0-pro)
temperature: Generation temperature (0.0-1.0), controls the randomness and creativity of the output.
top_p: Nucleus sampling parameter (0.0-1.0)
top_k: Number of candidate tokens to consider (1-100). Higher values mean more diversity.
max_output_tokens: Maximum output tokens (1-8192)
include_images: Whether to include images in the response (True/False)

Optional Parameters

video: Optional GEMINI_VIDEO_CONFIG input. Its video is sent before images and prompt with embedded audio.
image_inputs: Optional ordered image autogrow inputs for image understanding.
use_proxy: Whether to use a proxy (True/False)
proxy_host: Proxy host address
proxy_port: Proxy port

Output

Text generation node output:

text: Generated text
image: If image generation is enabled, outputs the image.

Precautions/Notes

According to Google's "Generative AI Prohibited Use Policy", Gemini API has the following restrictions:
Must not be used to generate content that violates laws and regulations.
Must not be used to generate harmful, fraudulent, pornographic, or violent content.
Must not be used to generate content that infringes on others' privacy or intellectual property rights.
Image generation may have additional restrictions, and certain types of images might not be generatable.
Using this node requires a stable network connection or effective proxy settings.
API requests may be affected by Google server load.
Large requests may require longer processing times.
The image generation feature requires a model that explicitly supports it (e.g., gemini-2.0-flash-exp).

Troubleshooting

If you encounter network connection problems, please check your proxy settings.
If API requests fail, please check if your API key is valid.
If dependency installation fails, please manually install the required dependency packages.

Acknowledgements

Thanks to Google for providing the Gemini API service. Thanks to tatookan for creating the original custom node repository for me to expand upon.