在 ComfyUI 中稳定调用 Google Gemini API 的节点，提供向后兼容与稳定性，支持错误回退、文本/JSON响应、多图（最多三张/批次）、自定义系统指令和可调安全过滤。
💡 在 ComfyUI 流程中稳定调用 Gemini 并处理文本与图像多模态响应。
🍴 7 Forks💻 Python🔄 2026-03-23
https://pan.quark.cn/s/e58c8376a81b
📦 requirements.txt
torch
pillow
numpy
google-generativeai>=0.8.3
📄 README
ComfyUI-Gemini
These nodes integrate the Gemini API into ComfyUI, allowing you to send prompts and images to Gemini AI models.
Features
Error Fallback Value: Specify a fallback value to use if there’s an error accessing the Gemini API.
Response Type Selection: Choose between receiving responses in plain text or json format.
Image Support: Include up to three images(or batches) with your prompt to provide visual context.
Custom System Instructions: Set a system instruction to guide the AI’s responses.
Adjustable Safety Settings: Control the content filtering level to block inappropriate content.
Setting Up the Gemini API Key
Obtain API Key:
Sign up for access to the Gemini API through Google Cloud Platform.
Create a new API key with permissions to access the Gemini models.
Set the API Key:
Option 1: Set it in your environment variables:
On Windows:
“`cmd
set GOOGLE_API_KEY=your_api_key_here
“`
On Linux/macOS:
“`bash
export GOOGLE_API_KEY=your_api_key_here
“`
Option 2: Provide it directly in the node’s api_key input.
Inputs Explained
prompt (STRING):
Your main question or instruction for the AI.
Example: “What are the health benefits of green tea?”
safety_settings (CHOICE):
Controls the filtering of the response content.
Options:
BLOCK_NONE: No filtering.
BLOCK_ONLY_HIGH: Blocks only high-risk content.
BLOCK_MEDIUM_AND_ABOVE: Blocks medium and high-risk content.
response_type (CHOICE):
Determines the format of the AI’s response.
Options:
text: Plain text response.
json: Response in JSON format.
model (CHOICE):
Selects the Gemini model to use.
Options:
gemma-3-12b-it: Very small model that you can run locally or remotely on the Google infrastructure.
gemma-3-27b-it: Medium size model that you can run locally or remotely on the Google infrastructure.
gemini-2.0-flash-lite-001: Old version of the very fast and free model.
gemini-2.0-flash-001: Old version of the fast and free model.
gemini-2.5-flash: Latest stable version of the fast and free model.
gemini-2.5-pro: Latest stable Pro model with advanced capabilities.
api_key (STRING, Optional):
Your Gemini API key.
Recommended to set via environment variable.
proxy (STRING, Optional):
Proxy server URL if you need to route requests through a proxy. See: Why use Proxy?
image_1, image_2, image_3 (IMAGE, Optional):
Images or image batches to include with your prompt.
Useful for tasks like image captioning or visual question answering.
system_instruction (STRING, Optional):
Sets a system-level instruction to influence the AI’s behavior.
Example: “You are a helpful assistant that provides concise answers.”
error_fallback_value (STRING, Optional, Lazy Input):
Value to return if an error occurs when accessing Gemini.
If not set, the node will raise an exception on error.
temperature (FLOAT, Optional, Lazy Input):
Controls the randomness of the output. Any negative value means to use default value for model.
num_predict (INT, Optional, Lazy Input):
The maximum number of tokens to include in a candidate, value 0 means to use default value for model.