ComfyUI-Gemini-3

ComfyUI-Gemini-3-2

Full-featured, high-performance Google Gemini 3 integration for ComfyUI. This plugin implements all Gemini 3 capabilities with optimized performance.

🔧 Latest Update (2025-11-19): Fixed Gemini API error 400 "Invalid thinkingLevel field". The thinkingLevel parameter is now correctly placed in the API request. See QUICK_FIX_GUIDE.md for details.

🌟 Features

Core Gemini 3 Capabilities

• ✅ Dynamic Thinking Levels (Low/High) - Control reasoning depth vs latency
• ✅ Media Resolution Control - Fine-grained control over image/video/PDF processing
• ✅ Thought Signatures - Maintain reasoning context across API calls
• ✅ Function Calling - Strict validation with parallel and sequential support
• ✅ Structured Outputs - JSON schema validation with tools integration
• ✅ Context Caching - Reduce costs for repeated context (2048+ tokens)
• ✅ Batch API Support - Process multiple requests efficiently

Multimodal Support

• 📝 Text generation with advanced reasoning
• 🖼️ Image understanding and generation (up to 1120 tokens per image)
• 🎥 Video analysis (70-280 tokens per frame)
• 🎵 Audio understanding (speech, music)
• 📄 PDF document processing (optimized at medium resolution)
• 🔗 URL context and web search integration

Advanced Features

• 🧠 Multi-step reasoning with thought signature preservation
• 🔄 Streaming support with real-time updates
• 🛡️ Safety settings with configurable thresholds
• 📊 Token counting and usage tracking
• 🌡️ Temperature control (optimized for Gemini 3)
• 🔧 System instructions for behavior customization
• 🔍 Google Search integration for real-time information
• 💻 Code Execution for mathematical and data analysis tasks

📦 Installation

1. Clone into your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/yourusername/ComfyUI-Gemini-3-2.git

2. Install dependencies:

cd ComfyUI-Gemini-3-2
pip install -r requirements.txt

3. Set your API key:
   • Get your API key from Google AI Studio
   • Set environment variable: GEMINI_API_KEY=your_key_here
   • Or configure in the node settings

🎮 Nodes

基础节点 (Basic Nodes)

1. Gemini 3 文本生成 (Text Generation)

Generate text with advanced reasoning capabilities.

• Inputs: prompt, images, audio, video, PDFs, thinking level, media resolution, system instruction
• Outputs: text, thought signature, usage metadata
• Features: Multimodal input, customizable thinking levels, media resolution control

高级节点 (Advanced Nodes)

2. Gemini 3 函数调用 (Function Calling)

Execute function calls with strict validation and thought signature preservation.

• Inputs: prompt, function declarations, thought signatures, function responses
• Outputs: function calls (JSON), updated thought signature, text response
• Features: Parallel and sequential function calling, multi-step reasoning

3. Gemini 3 结构化输出 (Structured Output)

Generate JSON with schema validation and tool integration.

• Inputs: prompt, JSON schema, enable Google Search/Code Execution/URL Context
• Outputs: validated JSON output, usage metadata
• Features: Strict JSON schema validation, integrated with built-in tools

4. Gemini 3 聊天 (Chat)

Multi-turn conversation with automatic context management.

• Inputs: message, chat history, system instruction, max output tokens
• Outputs: response, updated chat history (with thought signatures), usage metadata
• Features: Automatic history management, thought signature preservation

流式节点 (Streaming Nodes)

5. Gemini 3 流式生成 (Streaming Generation)

Real-time streaming generation with progressive output.

• Inputs: prompt, thinking level, system instruction, max output tokens
• Outputs: full text, thought signature, stream chunks info
• Features: Real-time response streaming, chunk-by-chunk proce