ComfyUI-LLMs-Toolkit

Language / 语言切换

---

Inject powerful Large Language Models into your ComfyUI workflows!

A professional-grade ComfyUI custom node suite supporting world-leading Large Language Models

---

Project Overview

ComfyUI-LLMs-Toolkit is a high-performance ComfyUI extension designed for AI creators. Through a lightweight API-driven approach, you can easily integrate world-class Large Language Models like DeepSeek, Qwen, GPT, etc., even with limited computational resources, democratizing AI creation.

Why Choose Us?

• Zero Hardware Barrier: No high-end GPU required, enjoy cutting-edge AI capabilities with just APIs
• Global Model Support: Integrated mainstream LLMs worldwide, one-stop solution
• High-Performance Architecture: Optimized concurrent processing, dramatically improving workflow efficiency
• Developer Friendly: Environment variable configuration, Docker support, developer-first approach

---

Core Features

Powerful LLM Ecosystem

Category	Supported Models	Key Capabilities
Chinese Leaders	DeepSeek-V3, Qwen-Max, GLM-4	Chinese understanding, code generation, mathematical reasoning
International Giants	GPT-4, Claude-3, Gemini	Multimodal processing, creative writing, complex reasoning
Professional Vertical	Doubao, Spark, Moonshot	Dialogue generation, long text, role-playing

Technical Highlights

• Lightning Deployment: One-click installation, get started in 5 minutes
• Concurrency Optimization: Multi-threading processing, supports batch requests
• Security First: Environment variable configuration, secure API key management
• Hot Reload: Dynamic configuration updates, no need to restart ComfyUI
• Smart Caching: Response caching mechanism, reducing API call costs
• Containerized: Docker support, consistent deployment environment

---

Quick Start

System Requirements

• Python: >= 3.8
• ComfyUI: Latest version
• Memory: >= 4GB RAM
• Network: Stable internet connection

Lightning Installation

# Method 1: Git Clone (Recommended)
cd ComfyUI/custom_nodes/
git clone https://github.com/HuangYuChuh/ComfyUI-LLMs-Toolkit.git
cd ComfyUI-LLMs-Toolkit

# Install Dependencies
pip install -r requirements.txt

# Method 2: Docker Deployment
docker pull your-dockerhub/comfyui-llms-toolkit:latest
docker run -d --name comfyui-llms -p 8188:8188 your-dockerhub/comfyui-llms-toolkit

Environment Configuration

1. Create Configuration File

# Copy environment variable template
cp config/env.example .env

2. Configure API Keys

Edit the .env file and select the models you want to use:

# DeepSeek (Recommended for beginners)
DEEPSEEK_API_KEY=sk-your_deepseek_key_here
DEEPSEEK_BASE_URL=https://api.deepseek.com
DEEPSEEK_MODEL_NAME=deepseek-chat

# Qwen (Chinese optimized)
QWEN_API_KEY=your_qwen_key_here
QWEN_BASE_URL=https://dashscope.aliyuncs.com/api/v1
QWEN_MODEL_NAME=qwen-max

# OpenAI GPT (International standard)
OPENAI_API_KEY=sk-your_openai_key_here
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_MODEL_NAME=gpt-4o-mini

# More configurations see config/env.example

3. API Key Acquisition Guide

Provider	Get Address	Free Credits	Rating
DeepSeek	platform.deepseek.com	¥500 free credits	⭐⭐⭐⭐⭐
Qwen	dashscope.aliyun.com	1M tokens/month	⭐⭐⭐⭐
OpenAI	platform.openai.com	$5 trial credits	⭐⭐⭐⭐⭐
GLM	open.bigmodel.cn	5M tokens/month	⭐⭐⭐⭐

First Run

1. Restart ComfyUI:

   # Restart ComfyUI to load new nodes
   cd /path/to/ComfyUI
   python main.py
   

2. Find Nodes:

   • Right-click in ComfyUI interface
   • Navigate to Add Node -> LLMs Toolkit
   • Select the node type you need

3. Start Creating:

   • Drag in LLM nodes
   • Configure model parameters
   • Connect inputs and outputs
   • Enjoy your AI creation journey!

---

Supported Model Matrix

Chinese AI Champions

Model	Provider	Core Advantages	Price Level	Config Prefix
DeepSeek-V3	DeepSeek	Mathematical reasoning, Code generation	Ultra-low	DEEPSEEK_
Qwen-Max	Alibaba	Multimodal, Chinese optimization	Medium	QWEN_
GLM-4	Zhipu AI	Logical reasoning, Knowledge Q&A	Low	GLM_
Doubao-Pro	ByteDance	Dialogue generation, Creative writing	Medium	DOUBAO_
Spark-Max	iFLYTEK	Language understanding, Text analysis	Low	SPARK_
Moonshot-V1	Moonshot AI	Long text, Deep understanding	Medium	MOONSHOT_

International AI Giants

Model	Provider	Core Advantages	Price Level	Config Prefix
GPT-4o	OpenAI	General intelligence, Multimodal	High	OPENAI_
Claude-3.5	Anthropic	Safe dialogue, Long text	High	CLAUDE_
Gemini-Pro	Google	Search enhanced, Multilingual	Medium	GEMINI_

---

FAQ

Q: How to get API keys?

A: All major vendors provide free trial credits:

• DeepSeek: Register and get ¥500 credits, suitable for beginners
• Qwen: Large quota available after Alibaba Cloud account verification
• OpenAI: New users get $5 free credits, requires international payment method

Q: Does it support local models?

A: Currently mainly supports API calls, local model support is on the development roadmap:

• v2.0: Planned support for Ollama local models
• v2.1: Support for GGUF format models
• v3.0: Support for custom model fine-tuning

Q: What to do when encountering connection errors?

A: Common solutions:

1. Check network: Ensure access to corresponding API services
2. Verify keys: Confirm API keys are correct and have balance
3. Check configuration: Confirm BASE_URL format is correct
4. View logs: Check ComfyUI console output

---

📅 Changelog

[1.1.0] - 2026-03-01

✨ Added

• DeepSeek Reasoning Support: Supported extracting deep thinking processes (reasoning_content) from DeepSeek R1 models, exposing it to a dedicated reasoning output pin in the generic node.
• o1/o3 Model System Role Compatibility: Integrated a downgrade workaround for strict 'o1/o3' models by spoofing an assistant acceptance response.

🛠 Changed

• Shared API Client Refactor: Extracted HTTP request logic to api_client.py for smarter retrying equipped with Exponential Backoff & Jitter.
• Graceful Error Degradation: The API node now outputs readable text error traces instead of throwing Python Exceptions, preventing ComfyUI workflows from halting abruptly.
• User-Agent Masquerade: Added browser-standard User-Agent headers to successfully bypass HTTP 403 (Cloudflare Error 1010) on proxy platforms.

🐛 Fixed

• Multi-turn Memory Fix: Fixed openai_compatible.py to properly store assistant responses when enable_memory is toggled on.
• Node Parameter Deserialization Bug UI: Switched from dynamic widgets to naive <canvas> element drawings to fix the notorious parameter shift bug caused by token_usage_display.

🗑 Removed

• Redundant Token Output Pins: Permanently removed the input_tokens and output_tokens pins from the main LLM node since these metrics are already elegantly rendered directly onto the node's foreground canvas.

---

Open Source License

This project is licensed under GNU General Public License v2.0.

This means:

• Free to use: Both personal and commercial use
• Free to modify: You can modify the source code
• Free to distribute: You can distribute original or modified versions
• Open source obligation: Modified versions must be open source

---

If this project helps you, please give us a Star!

Contact Us

---

Made with love for the ComfyUI community

Making AI creation simpler, making technology more humane