🚀 Complete Guide: Multi-Environment ComfyUI Setup with RTX 5090

This guide will help you create and manage multiple isolated ComfyUI environments, perfect for different workflows while sharing models and optimizing for your RTX 5090.

📋 Table of Contents

1. Understanding the Approach
2. Prerequisites
3. Quick Start
4. File Structure
5. Step-by-Step Setup
6. Managing Environments
7. Troubleshooting
8. Best Practices

🎯 Understanding the Approach

Why Multiple Environments?

• Dependency Isolation: Different custom nodes require different package versions
• Stability: Experimental features won't break your main installation
• Specialization: Optimized environments for specific tasks (training, audio, etc.)
• Testing: Try new features without risking your stable setup

How It Works

Base Directory/
├── comfyui_main/          # Stable daily use
├── comfyui_experimental/  # Latest features
├── comfyui_training/      # Model training
├── comfyui_audio/         # Audio-specific nodes
└── Shared Models/         # All installations use the same model files

⚙️ Prerequisites

Hardware Requirements

• NVIDIA RTX 5090 (or any CUDA-compatible GPU)
• 16GB+ RAM recommended
• 50GB+ free disk space (for models and environments)

Software Requirements

• Windows 10/11 with latest updates
• Python 3.10-3.11 (recommended for compatibility)
• Git installed and in PATH
• NVIDIA Drivers supporting CUDA 12.8+

Verify Your System

Run these commands in Command Prompt:

# Check Python
python --version

# Check Git
git --version

# Check NVIDIA
nvidia-smi

# Check CUDA (if installed)
nvcc --version

🚀 Quick Start

1. Download the Setup Files

Create a new folder for your AI projects and save these files:

create_comfyui_env.bat - Installation wizard comfyui_launcher.bat - Master launcher with menu check_installations.bat - Diagnostic tool

2. Run the Installation

# Navigate to your AI projects folder
cd C:\AI_Projects

# Run the installation wizard
create_comfyui_env.bat

3. Choose Your Installation Type

Select from the menu:

• 1. Main Stable - Daily use, most stable
• 2. Experimental - Latest features, may be unstable
• 3. Training - Optimized for model training
• 4. Custom - Advanced users

4. Launch ComfyUI

# Use the master launcher (recommended)
comfyui_launcher.bat

# Or direct launchers
start_comfyui_main.bat
start_comfyui_experimental.bat

📁 File Structure

Recommended Layout

C:\AI_Projects\
├── 📜 create_comfyui_env.bat          # Installation wizard
├── 📜 comfyui_launcher.bat            # Master menu launcher
├── 📜 check_installations.bat         # Diagnostic tool
├── 📜 update_comfyui_installations.bat # Batch updater
├── 🔧 comfyui_main\                   # Stable installation
│   ├── 🐍 venv\                       # Virtual environment
│   └── 💻 ComfyUI\                    # ComfyUI files
├── 🔬 comfyui_experimental\           # Testing installation  
│   ├── 🐍 venv\
│   └── 💻 ComfyUI\
├── 🎓 comfyui_training\               # Training installation
│   ├── 🐍 venv\
│   └── 💻 ComfyUI\
└── 🗂️ C:\Users\rick\ComfyUI\models\   # SHARED models directory
    ├── checkpoints\
    ├── loras\
    ├── vae\
    └── controlnet\

🔧 Step-by-Step Setup

Step 1: Prepare Your Workspace

# Create main AI projects folder
mkdir C:\AI_Projects
cd C:\AI_Projects

# Create shared models directory (if needed)
mkdir C:\Users\rick\ComfyUI\models

Step 2: First Installation (Main Stable)

1. Run create_comfyui_env.bat
2. Select Option 1: Main Stable Version
3. Wait for installation (10-20 minutes)
4. Verify installation completes successfully

Step 3: Verify RTX 5090 Detection

After installation, the script will automatically verify:

• ✅ PyTorch with CUDA 12.8+ support
• ✅ RTX 5090 GPU detection
• ✅ Compute capability sm_120
• ✅ All dependencies installed

Step 4: Additional Installations (Optional)

Repeat the process for other environment types:

• Experimental - For testing new features
• Training - For LoRA/model training
• Custom - For specialized workflows

Step 5: Organize Your Models

Place your models in the shared directory:

C:\Users\rick\ComfyUI\models\
├── checkpoints\      # .safetensors, .ckpt files
├── loras\           # LoRA files
├── vae\             # VAE models
├── controlnet\      # ControlNet models
├── upscale_models\  # ESRGAN, etc.
└── embeddings\      # Textual inversions

🎮 Managing Environments

Daily Use - Master Launcher (Recommended)

comfyui_launcher.bat

Benefits:

• Shows all available installations
• Automatic port assignment (no conflicts)
• Health checks before launching
• Management menu for updates

Quick Access - Individual Launchers

start_comfyui_main.bat
start_comfyui_experimental.bat
start_comfyui_training.bat

Running Multiple Instances Simultaneously

Each installation uses different ports:

• Main: Port 8188
• Experimental: Port 8189
• Training: Port 8190
• Custom: Port 8191+

Access them simultaneously:

• http://127.0.0.1:8188
• http://127.0.0.1:8189
• http://127.0.0.1:8190

🛠️ Maintenance

Updating All Installations

# From master launcher, choose 'M' then '2'
# Or run directly:
update_comfyui_installations.bat

Checking Installation Health

# From master launcher, choose 'M' then '3'  
# Or run directly:
check_installations.bat

Adding Custom Nodes

Custom nodes go in each installation's folder:

comfyui_main\ComfyUI\custom_nodes\
comfyui_experimental\ComfyUI\custom_nodes\

Tip: Install nodes separately in each environment to avoid conflicts.

🔧 Troubleshooting

Common Issues and Solutions

❌ "Installation not found" Error

Problem: Launcher can't find the installation directory.

Solution:

# Run diagnostic
check_installations.bat

# Verify directory exists
dir comfyui_*

❌ CUDA Not Detected

Problem: PyTorch can't find your RTX 5090.

Solution:

1. Update NVIDIA drivers to support CUDA 12.8+
2. Verify installation used correct PyTorch version:

   pip list | findstr torch

3. Reinstall with CUDA 12.8 support

❌ Port Already in Use

Problem: Another application is using the ComfyUI port.

Solution:

• Use different ports for each installation
• Check running processes:

  netstat -ano | findstr :8188

❌ Dependency Conflicts

Problem: Custom nodes downgrade packages (like huggingface-hub).

Solution:

• Use separate environments for conflicting nodes
• Install problematic nodes in experimental environment first

Diagnostic Commands

# Check GPU detection
python -c "import torch; print(f'CUDA: {torch.cuda.is_available()}, GPU: {torch.cuda.get_device_name(0) if torch.cuda.is_available() else None}')"

# Check installation structure
check_installations.bat

# Verify Python environment
where python
pip list | findstr torch

💡 Best Practices

1. Environment Strategy

• Main: Stable nodes only, daily workflow
• Experimental: New nodes, testing features
• Training: Dedicated to model training
• Audio/Video: Specialized for media processing

2. Model Management

• Keep all models in the shared directory
• Use symbolic links if you have models on different drives
• Regularly clean up unused models to save space

3. Backup Strategy

• Export your workflows regularly
• Backup your ComfyUI\user folder for custom settings
• Keep notes on which nodes work in which environment

4. Performance Optimization

• Use --highvram flag if you have sufficient GPU memory
• Monitor GPU temperature during long sessions
• Close other GPU-intensive applications when training

5. Organization Tips

C:\AI_Projects\
├── 📁 comfyui_main\          # Daily driver
├── 📁 comfyui_testing\       # Node testing
├── 📁 comfyui_training\      # Model training
├── 📁 comfyui_audio\         # Audio-specific
├── 📁 comfyui_video\         # Video-specific
└── 📁 project_archives\      # Old project backups

🎯 Workflow Examples

Daily Content Creation

# Use main stable environment
start_comfyui_main.bat
# Port 8188, stable nodes only

Testing New Nodes

# Use experimental environment  
start_comfyui_experimental.bat
# Port 8189, test new nodes here first

Model Training

# Use training environment
start_comfyui_training.bat
# Port 8190, all training dependencies available

Multiple Projects Simultaneously

# Run multiple instances for different clients/projects
start_comfyui_main.bat       # Client A - Port 8188
start_comfyui_experimental.bat # Client B - Port 8189

📞 Support

Getting Help

1. Check logs: ComfyUI\user\comfyui.log
2. Run diagnostics: check_installations.bat
3. Verify installation: Use management menu in launcher

Common Recovery Steps

1. If an environment gets corrupted, delete it and recreate
2. Models are safe in shared directory
3. Custom nodes can be reinstalled quickly

Performance Monitoring

# Monitor GPU usage
nvidia-smi -l 1

# Check system resources
tasklist | findstr python

This approach gives you maximum flexibility while maintaining stability. Start with one main environment and expand as needed!