ComfyUI Extension: ComfyUI Razv LLM Node

Authored by razvanmatei-sf

Created

Updated

0 stars

Custom ComfyUI node for integrating Claude API with image and text capabilities.

Custom Nodes (0)

    README

    ComfyUI Razv LLM Node

    Custom ComfyUI node for integrating Claude and Gemini APIs with image and text capabilities.

    Features

    • Text-based chat with latest Claude and Gemini models
    • Optional image input support for vision tasks
    • Support for Claude's "Extended thinking" capability across all models
    • Multiple Claude models: Opus 4.1, Sonnet 4.5, Haiku 4.5
    • Multiple Gemini models: 3 Pro Preview, 2.5 Pro, 2.5 Flash, 2.5 Flash Lite
    • Adjustable parameters (temperature, max tokens, system prompt, seed for Gemini)

    Installation

    Option 1: ComfyUI Manager (Recommended)

    1. Open ComfyUI Manager
    2. Search for "razv-llm"
    3. Click Install

    Option 2: Git Clone

    cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-llm.git
cd razv-llm
pip install -r requirements.txt

    Configuration

    Claude API

    Set your Anthropic API key using one of these methods:

    1. Environment variable: ANTHROPIC_API_KEY or CLAUDE_API_KEY
    2. Direct input in the node's API key field

    Gemini API

    Set your Google API key using one of these methods:

    1. Environment variable: GEMINI_API_KEY or GOOGLE_API_KEY
    2. Direct input in the node's API key field

    Usage

    1. Add the "RazvLLMChat" node to your workflow
    2. Connect an image (optional) and provide a text prompt
    3. Configure model and parameters as needed
    4. Execute to get Claude's response

    Node Parameters

    • prompt: Text input for the LLM
    • model: Choose from available models (see below)
    • temperature: Control response randomness (0.0 - 2.0)
    • max_tokens: Maximum response length (up to 200,000)
    • system_prompt: Optional system message to guide the model's behavior
    • api_key: Your API key (if not set via environment)
    • image: Optional image input for vision tasks
    • seed: Random seed for reproducible results (Gemini models only)
    • timeout: Request timeout in seconds (10-3600, default: 500)

    Available Models

    Claude Models (as of November 2025)

    • claude-sonnet-4-5 / claude-sonnet-4-5-20250929: Best balance of speed, cost, and intelligence. Recommended for most use cases. ($3/$15 per million tokens)
    • claude-opus-4-1 / claude-opus-4-1-20250805: Most capable model for complex tasks. Best coding model (72.5% SWE-bench). ($15/$75 per million tokens)
    • claude-haiku-4-5 / claude-haiku-4-5-20251001: Fast and cost-efficient for simple tasks. Near-frontier performance. ($1/$5 per million tokens)

    All Claude models support "Extended thinking" capability for deep reasoning and can work continuously for several hours on long-running tasks.

    Gemini Models (as of November 2025)

    • gemini-3-pro-preview: Most intelligent model with advanced reasoning capabilities. 1M context window. ($2-4/$12-18 per million tokens)
    • gemini-2.5-pro: Capable reasoning model
    • gemini-2.5-flash: Best price-performance balance
    • gemini-2.5-flash-lite: Fastest, most cost-efficient

    Requirements

    • ComfyUI
    • Python 3.8+
    • See requirements.txt for Python dependencies

    License

    MIT