ComfyUI Installation Script for Linux & Windows

Automated installation scripts for ComfyUI on Linux and Windows systems that handle everything from Python environment setup to custom node installation. These scripts provide a streamlined, dependency-conflict-free installation process with full control over package versions and installation steps.

🌟 Features

Automated Environment Setup: Installs pyenv, Python, and creates an isolated virtual environment
Interactive Prompts: Collects ComfyUI version, frontend version, and alias name at runtime
Version Control: Pin specific versions of ComfyUI, PyTorch, NumPy, Transformers, and other critical packages
Frontend Pinning: Each alias/launcher auto-installs the correct frontend version on launch
GPU Acceleration: Supports CUDA 13.0, CUDA 12.8, CUDA 12.6, ROCm 7.1, and CPU-only installations
Performance Optimization: Includes optional Nunchaku, Flash Attention, and Sage Attention
Custom Node Collection: Automatically clones and configures 45+ popular custom nodes
Lowercase Cloning: Clones custom nodes with lowercase directory names to match ComfyUI-Manager convention
Selective Installation: Choose which components to install via interactive menu
Smart Dependency Management: Prevents version conflicts by enforcing package versions
Symlink Support: Centralize models, input, output, user data, and custom_nodes across multiple installations
Multi-Install Support: Run multiple ComfyUI versions side-by-side with per-version aliases and launchers
Shell Integration: Adds user-chosen aliases (e.g., comfy2, comfy3) with non-destructive, additive handling
Multiple Shell Support: Auto-detects and configures bash, zsh, and fish shells

📋 Prerequisites

Linux

Operating System: Linux (Ubuntu, Debian, Fedora, Arch, openSUSE, etc.)
Permissions: Sudo access for installing system dependencies
Disk Space: ~10-20GB for full installation (varies with custom nodes)
GPU (optional): NVIDIA GPU with CUDA support for acceleration features
Git: For cloning repositories

Windows

Operating System: Windows 10 or Windows 11
PowerShell: Version 5.1+ (included with Windows 10/11) or PowerShell 7+
Disk Space: ~10-20GB for full installation
GPU (optional): NVIDIA GPU with CUDA support for acceleration features
Git: Git for Windows installed and in PATH (download)
Visual Studio Build Tools (optional): Required only if building packages from source (e.g., flash-attn)

🚀 Quick Start

Linux

# Download the script
git clone https://github.com/r-vage/ComfyUI-Installation-Script-for-Linux.git
cd ComfyUI-Installation-Script-for-Linux

# Make executable
chmod +x install_comfy_env.sh

# Run the script
./install_comfy_env.sh

Windows

# Download the script
git clone https://github.com/r-vage/ComfyUI-Installation-Script-for-Linux.git
cd ComfyUI-Installation-Script-for-Linux

# Allow script execution (if not already enabled)
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

# Run the script
.\install_comfy_env_win.ps1

Note: The Windows script uses directory junctions (similar to symlinks) which work without Administrator privileges. If you need true symlinks, run PowerShell as Administrator.

The script will display your configuration and prompt you to select which installation steps to run.

⚙️ Configuration

Before running the script, you can customize these variables at the top of install_comfy_env.sh (Linux) or install_comfy_env_win.ps1 (Windows):

Python Configuration

PYTHON_VERSION="3.12.10"          # Python version (3.10.x, 3.11.x, 3.12.x, 3.13.x, 3.14.x recommended)
VENV_PATH="/mnt/data/AI/comfy_env"  # Virtual environment location
PYENV_ROOT="${PYENV_ROOT:-$HOME/.pyenv}"  # pyenv installation directory

Windows equivalent: $PYTHON_VERSION = "3.12.10" — venv path defaults to $BASE_PATH\comfy_env (e.g. D:\AI\comfy_env)

Recommended Python versions: 3.10.x, 3.11.x, 3.12.x, 3.13.x, 3.14.x have prebuilt wheels for PyTorch, Nunchaku, and Flash Attention. Other versions will require compilation from source.

PyTorch Configuration

PYTORCH_VERSION="2.9"             # PyTorch major.minor for wheel URLs
PYTORCH_FULL_VERSION="2.9.1+cu128" # Full version string
PYTORCH_INDEX_URL="https://download.pytorch.org/whl/cu128"  # Wheel repository

Available options:

CUDA 13.0: cu130 (latest)
CUDA 12.8: cu128 (default)
CUDA 12.6: cu126
ROCm 7.1: rocm7.1 (AMD GPUs)
CPU only: cpu (no GPU)

Example CUDA 13.0 configuration:

PYTORCH_FULL_VERSION="2.9.1+cu130"
PYTORCH_INDEX_URL="https://download.pytorch.org/whl/cu130"

Example ROCm 7.1 configuration (AMD GPUs):

PYTORCH_FULL_VERSION="2.9.1+rocm7.1"
PYTORCH_INDEX_URL="https://download.pytorch.org/whl/rocm7.1"

Example CPU configuration:

PYTORCH_FULL_VERSION="2.9.1+cpu"
PYTORCH_INDEX_URL="https://download.pytorch.org/whl/cpu"

Critical Package Versions

NUMPY_VERSION="2.2.6"             # NumPy version (2.2.x for PyTorch 2.9+)
TRANSFORMERS_VERSION="5.3.0"      # Transformers (5.x for Qwen3-VL/Mistral3)

These versions are enforced at the end of installation to override any conflicting dependencies from custom nodes.

ComfyUI Installation (Interactive)

When you run the script, it prompts for three values:

Enter values for this installation (press Enter for defaults):

  ComfyUI version [0.19.3]:     # → clones/checks out this tag (e.g., v0.19.3)
  Frontend version [1.42.11]:   # → pinned in aliases and launchers
  Launch alias [comfyui]:       # → shell alias name (e.g., comfy2, comfy3)

The folder name is derived automatically: version 0.19.3 → ComfyUI_0.19.3.

Static configuration (Python, PyTorch, NumPy, Transformers) stays at the top of the script and rarely changes.

Multiple ComfyUI Installations:
Simply run the script again with different values at the prompts:

# First run:   version 0.19.3, alias comfyui   → /mnt/data/AI/ComfyUI_0.19.3
# Second run:  version 0.19.0, alias comfy2     → /mnt/data/AI/ComfyUI_0.19.0
# Third run:   version 0.18.2, alias comfy3     → /mnt/data/AI/ComfyUI_0.18.2

Each installation gets its own alias and launcher script (start_comfy2.sh, start_comfy3.bat).
All installations share the same venv, models, input, output, user data, and custom nodes via symlinks.
The envact alias is always shared (one venv for all installs).

Symlink / Junction Configuration

CREATE_SYMLINKS=true                # Enable/disable symlink creation
USER_MODELS_PATH="/mnt/data/AI/models"          # Centralized models directory
USER_INPUT_PATH="/mnt/data/AI/input"            # Centralized input directory
USER_OUTPUT_PATH="/mnt/data/AI/output"          # Centralized output directory
USER_USERDATA_PATH="/mnt/data/AI/user"          # Centralized user directory (settings, workflows, templates)
USER_CUSTOM_NODES_PATH="/mnt/data/AI/custom_nodes"  # Centralized custom_nodes directory

Windows equivalent: $CREATE_SYMLINKS = $true — paths default to D:\AI\models, D:\AI\input, D:\AI\output, D:\AI\user, D:\AI\custom_nodes (uses directory junctions instead of symlinks)

Symlinks allow you to:

Share models across multiple ComfyUI installations
Share input images across multiple ComfyUI installations
Share custom nodes across multiple ComfyUI installations (useful when testing different ComfyUI versions)
Share user data (frontend settings, saved workflows, node templates) across all versions — cross-version safe because comfy.settings.json is merged against frontend defaults at runtime
Store models on a different drive/partition
Keep outputs in a centralized location

Set CREATE_SYMLINKS=false to use the default per-installation directories.

Tip — directories outside BASE_PATH:
By default, the centralized paths (USER_MODELS_PATH, USER_INPUT_PATH, USER_OUTPUT_PATH, USER_USERDATA_PATH, USER_CUSTOM_NODES_PATH) are derived from BASE_PATH. You can point any of them to a completely different location — for example a larger disk — by editing the "Derived Paths" section in the script:

Linux:

USER_MODELS_PATH="/mnt/ssd2/models"             # Models on a fast SSD
USER_OUTPUT_PATH="/home/$USER/comfyui_output"   # Output in home directory
USER_USERDATA_PATH="/home/$USER/comfyui_user"   # User data in home directory

Windows:

$USER_MODELS_PATH = "E:\models"                         # Models on drive E:
$USER_OUTPUT_PATH = "C:\Users\$env:USERNAME\comfyui_out" # Output in user profile

The script creates the target directories automatically and symlinks/junctions them into the ComfyUI tree.

Optional Features

INSTALL_NUNCHAKU=true               # Nunchaku acceleration (requires NVIDIA GPU)

Set to false to skip Nunchaku installation if you don't have an NVIDIA GPU or don't need this optimization.

📦 What Gets Installed

Installation Steps

The script is divided into 11 steps that you can run selectively:

Python Environment - pyenv, Python version, virtual environment
PyTorch - PyTorch, torchvision, torchaudio with CUDA/CPU support
Nunchaku - Acceleration library (optional, NVIDIA GPU only)
Face Recognition - facexlib, insightface, onnxruntime-gpu, facenet_pytorch
ComfyUI Core - ComfyUI base installation and requirements
Custom Nodes - 45+ popular custom nodes (see list below)
Custom Node Dependencies - Install requirements for all custom nodes
Performance Libraries - llama-cpp-python, flash-attn, sageattention
Upgrade/Pin Packages - Upgrade specific packages to latest compatible versions
Enforce Versions - Force exact versions of PyTorch, NumPy, Transformers, ComfyUI Frontend
Shell Aliases - Add user-chosen launch alias and envact alias to shell config

Custom Nodes Included (45 nodes)

Core Extensions

ComfyUI_SmartLML - Smart Language Model Loader
ComfyUI_Eclipse - Extended toolkit
ComfyUI-Manager - Node manager

UI & Workflow Tools

ComfyUI-Crystools-MonitorOnly - System monitoring
ComfyUI-Custom-Scripts - UI enhancements
rgthree-comfy [FORK] - Workflow utilities
ComfyUI-Easy-Use - Simplified nodes
ComfyUI_essentials - Essential utilities
ComfyUI_essentials_mb - Additional essentials
cg-image-filter - Image filter/selector

Model Support & Optimization

ComfyUI-GGUF - GGUF model support
ComfyUI-nunchaku - Nunchaku integration (optional)
ComfyUI-TeaCache - Caching optimizations
ComfyUI_Patches_ll - Performance patches
ComfyUI_IPAdapter_plus - IP-Adapter support

Sampling & Scheduling

RES4LYF [FORK] - Advanced samplers, schedulers, and noise types
sd-dynamic-thresholding - Dynamic CFG thresholding
ComfyUI-Raffle - semi-random prompt generator for danbooru tags
SeedVarianceEnhancer - adds diversity to Z-Image Turbo outputs
ComfyUI-DyPE - DyPE. Artifact-free 4K+ image generation

ControlNet & Advanced Control

ComfyUI-Advanced-ControlNet - Advanced ControlNet features
comfyui_controlnet_aux - ControlNet preprocessors

Image Processing & Effects

ComfyUI-Impact-Pack - Image enhancement suite
ComfyUI-Impact-Subpack - Impact Pack extensions
ComfyUI_LayerStyle - Layer effects
ComfyUI_LayerStyle_Advance - Advanced layer effects
ComfyUI-Detail-Daemon - Detail enhancement
ComfyUI-KJNodes - Various utilities
Comfyui_TTP_Toolset - Additional tools
ComfyUI_UltimateSDUpscale - Tiled upscaling
ComfyUI-VAE-Utils - VAE utilities
was-node-suite-comfyui - WAS toolkit
comfy_mtb - MTB's custom nodes

Specialized Models

ComfyUI-Florence2 - Florence2 model support
ComfyUI-SUPIR - Super resolution
ComfyUI_BiRefNet_ll - Background removal
ComfyUI_PuLID_Flux_ll - Face ID for Flux
ComfyUI-ReActor - Face swapping

Video Processing

ComfyUI-VideoHelperSuite - Video utilities
ComfyUI-Frame-Interpolation - Video frame interpolation
ComfyUI-GIMM-VFI - Video frame interpolation
ComfyUI-WanVideoWrapper - Wan video generation
ComfyUI-SeedVR2_VideoUpscaler - Video upscaling
ComfyUI-WanMoeKSampler - Wan MoE sampling

🎯 Usage

Interactive Installation

Run the script and select steps interactively:

./install_comfy_env.sh

You'll see a menu:

Select installation steps (numbers, ranges, or 'a' for all — e.g., 6-10 or 1 5 6-8):
  1) Python environment (pyenv + venv)
  2) PyTorch and base dependencies
  ...
  a) All steps (default)

Your selection [a]:

Examples:

a - Install everything (default)
1 2 5 - Only setup Python, PyTorch, and ComfyUI core
6-7 - Only clone and setup custom nodes (if you already have ComfyUI)
5 9-11 - ComfyUI core + upgrade/pin/aliases
10 - Only enforce package versions (useful after manual package changes)

Running ComfyUI

After installation, you have several options (examples assume alias comfy2 and version 0.19.0):

Linux

Option 1: Launcher script (recommended)

/mnt/data/AI/start_comfy2.sh

Option 2: Shell alias (if Step 11 was run)

comfy2    # Auto-installs correct frontend, activates env, launches ComfyUI

Option 3: Manual

source /mnt/data/AI/comfy_env/bin/activate
cd /mnt/data/AI/ComfyUI_0.19.0
python main.py

Windows

Option 1: Double-click launcher

D:\AI\start_comfy2.bat

Option 2: PowerShell launcher

D:\AI\start_comfy2.ps1

Option 3: PowerShell alias (if Step 11 was run, after reloading profile)

comfy2

Option 4: Manual

& "D:\AI\comfy_env\Scripts\Activate.ps1"
cd "D:\AI\ComfyUI_0.19.0"
python main.py

Each alias and launcher automatically pins the correct frontend version via uv pip install -q comfyui-frontend-package==VERSION (instant no-op when already installed).

Activating Virtual Environment Only

# Shared across all installations (Step 11)
envact

# Or manually
source /mnt/data/AI/comfy_env/bin/activate

🔧 Customizing Custom Nodes

To add/remove custom nodes, edit the [6/10] Clone Custom Nodes section around line 550:

# Add your custom node:
clone_if_missing "https://github.com/yourusername/your-custom-node.git"

# Remove a node by commenting it out or deleting the line:
# clone_if_missing "https://github.com/some/node-you-dont-want.git"

Then run:

./install_comfy_env.sh

And select steps 6-7 to update custom nodes and their dependencies.

🐛 Troubleshooting

pyenv Installation Fails

If pyenv dependencies fail to install, you may need to manually install build dependencies for your distribution:

Ubuntu/Debian:

sudo apt-get update
sudo apt-get install build-essential libssl-dev zlib1g-dev \
libbz2-dev libreadline-dev libsqlite3-dev curl git \
libncursesw5-dev xz-utils tk-dev libxml2-dev libxmlsec1-dev libffi-dev liblzma-dev

Fedora:

sudo dnf install make gcc patch zlib-devel bzip2 bzip2-devel readline-devel sqlite sqlite-devel openssl-devel tk-devel libffi-devel xz-devel libuuid-devel gdbm-libs libnsl2

Nunchaku or Flash Attention Installation Fails

These require:

NVIDIA GPU
CUDA 12.6+ or compatible version
Prebuilt wheels for your Python version

If you don't have an NVIDIA GPU, set INSTALL_NUNCHAKU=false and skip step 3.

Custom Node Dependencies Conflict

If a custom node installs incompatible package versions, run step 10 to enforce configured versions:

./install_comfy_env.sh  # Select step 10 only

Shell Aliases Not Working

Linux

After step 11, you need to reload your shell configuration:

# For bash
source ~/.bashrc

# For zsh
source ~/.zshrc

# For fish
source ~/.config/fish/config.fish

# Or just restart your terminal

Windows

After step 11, reload your PowerShell profile:

. $PROFILE

# Or just restart PowerShell

Windows: Execution Policy Error

If you get a script execution error:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Windows: pyenv-win Not Found After Installation

Close and reopen PowerShell. If still not found, ensure these are in your PATH:

%USERPROFILE%\.pyenv\pyenv-win\bin
%USERPROFILE%\.pyenv\pyenv-win\shims

Windows: Flash Attention Not Available

Flash Attention rarely has prebuilt Windows wheels. You can either:

Skip it (it's optional — ComfyUI works fine without it)
Install Visual Studio Build Tools and compile from source
Use Sage Attention as an alternative

Out of Disk Space

The full installation requires 10-20GB. To reduce space:

Skip custom nodes (don't run step 6-7)
Remove unused custom nodes manually from custom_nodes/ directory
Use symlinks to store models on a different drive

🤝 Contributing

Contributions are welcome! Here's how you can help:

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Make your changes:
- Add new custom nodes to the installation list
- Update package versions
- Add support for new Linux distributions
- Improve error handling
Test your changes on a fresh system if possible
Commit your changes (git commit -m 'Add amazing feature')
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

Reporting Issues

When reporting issues, please include:

Your Linux distribution and version
Python version being installed
Full error output
Which installation step failed

📝 License

This script is provided as-is for the ComfyUI community. Feel free to modify and distribute.

📜 Changelog

See CHANGELOG.md for the version history and recent changes.

🙏 Acknowledgments

ComfyUI - The amazing stable diffusion GUI
pyenv - Python version management (Linux)
pyenv-win - Python version management (Windows)
All the custom node developers for their incredible work

📞 Support

If you encounter issues:

Check the Troubleshooting section
Search existing Issues
Open a new issue with detailed information

Happy ComfyUI-ing! 🎨

Version 0.8.2

since im not using windows anymore there will be no further updates for the portable versions 🤷🏻.

pyTorch 2.9+cu128
python: 3.12.10
Frontend: 1.36.13
removed crystools -> replaced with crystools-monitor only
Even though some API imports have been fixed, the deprecated messages are still displayed.
the reason for adding the forked versions instead of the originals is that some of the nodes are currently not working with the latest comfyui version.

Additional Installed Libs

triton 3.51.1
llama_cpp_python 0.3.16
flash attention 2 (v2.8.3+torch2.9+cu128-cp312)
nunchaku-1.0.2+torch2.9-cp312-cp312
sage attention 2.2.0
transformers==4.56.2
insightface, facexlib, facenet_pytorch, comfy-cli, bitsandbytes, tensorflow, hf_xet, requests, pilgram, tf-keras, setuptools_scm, qwen-vl-utils, huggingface-hub, accelerate, Pillow, opencv-python, safetensors, psutil, PyYAML, aiohttp, av, gguf, ultralytics, pytorch-extension
python headers for torch compile/triton

Installed Custom Nodes

comfy-mtb
comfyui_birefnet_ll
comfyui_controlnet_aux
ComfyUI_Eclipse 1.2.1
comfyui_essentials
comfyui_essentials_mb
comfyui_layerstyle
ComfyUI_LayerStyle_Advance
ComfyUI_Patches_ll (forked and fixed Imports)
comfyui_pulid_flux_ll
comfyui_ttp_toolset
comfyui-advanced-controlnet
ComfyUI-Crystools-MonitorOnly (forked and fixed small error in browser log)
ComfyUI-Custom-Scripts (forked and fixed API Imports)
comfyui-detail-daemon
comfyui-easy-use
comfyui-florence2
comfyui-frame-interpolation
ComfyUI-GGUF
ComfyUI-GIMM-VFI
comfyui-impact-pack
comfyui-kjnodes
ComfyUI-Manager (removed the anoying fetch msg output)
ComfyUI-nunchaku (forked and fixed Imports)
ComfyUI-ReActor
comfyui-supir
ComfyUI-TeaCache (forked and fixed Imports)
comfyui-videohelpersuite
ComfyUI-WanVideoWrapper
raffle
RES4LYF (forked and partially fixed (community pull requests))
rgthree-comfy (forked and fixed API Imports)
sd-dynamic-thresholding
was-ns

Version 0.3.75

(You will see some warnings from various nodes because the authors of comfyui are changing the API. -> [DEPRECATION WARNING])

Frontend Version: 1.32.3 (fix)
- i've uploaded a patch for the latest frontend version if you want to upgrade to 1.34.3:
  https://civarchive.com/articles/23392/patch-for-the-comfyui-frontend
pyTorch 2.9+cu128
python: 3.12.10

Additional Installed Libs

triton 3.51.1
llama_cpp_python 0.3.16
flash attention 2 (v2.8.3+torch2.9+cu128-cp312)
nunchaku-1.0.2+torch2.9-cp312-cp312
sage attention 2.2.0
transformers==4.56.2
facenet_pytorch, pytorch-extension, facexlib, insightface, comfy-cli, bitsandbytes, tensorflow, hf_xet, requests, pilgram, tf-keras, nvidia-ml-py
python headers for torch compile/triton

Installed Custom Nodes

comfy-mtb
comfyui_birefnet_ll
comfyui_controlnet_aux
ComfyUI_Eclipse (1.0.60)
comfyui_essentials
ComfyUI_essentials_mb
comfyui_layerstyle
ComfyUI_LayerStyle_Advance
comfyui_patches_ll
comfyui_pulid_flux_ll
comfyui_ttp_toolset
comfyui-advanced-controlnet
ComfyUI-Crystools
comfyui-custom-scripts
comfyui-depthanythingv2
comfyui-detail-daemon
comfyui-easy-use
comfyui-florence2
comfyui-frame-interpolation
ComfyUI-GGUF
ComfyUI-GIMM-VFI
comfyui-impact-pack
comfyui-impact-subpack
comfyui-inspire-pack
comfyui-jdcn
comfyui-kjnodes
ComfyUI-Manager
comfyui-multigpu
ComfyUI-nunchaku
comfyui-reactor
comfyui-supir
comfyui-videohelpersuite
ComfyUI-WanMoeKSampler
ComfyUI-WanVideoWrapper
facerestore_cf
raffle
RES4LYF
rgthree-comfy
sd-dynamic-thresholding
sdxl_prompt_styler
teacache
was-ns
whiterabbit

Files with SHA256: 546b61930fcd6f4cd4867dbcc702e86972fed5f4b683eef12a544b32386e0377