๐ ComfyUI Auto-Installer โ v5 (Python Rewrite)
Version 5 is a full rewrite from the ground up in Python, replacing all the PowerShell scripts from previous versions. It's cross-platform, faster, smarter, and now ships with a TUI manager, Docker images, and GPU-optimized inference out of the box.
If you are upgrading from the PowerShell version (v4.x), a one-command migration preserves all your models, outputs, and custom nodes: irm https://get.umeai.art/migrate.ps1 | iex
โก Quick Start (One-Liner)
Windows (PowerShell):
irm https://get.umeai.art/comfyui.ps1 | iexLinux / macOS:
curl -fsSL https://get.umeai.art/comfyui.sh | sh
Only requires Git โ everything else (Python, uv, dependencies) is handled automatically.
โจ What's New in v5
Full Python rewrite โ no more PowerShell dependency
Cross-platform โ Windows, Linux, macOS, and Docker
TUI Manager โ interactive terminal UI to launch, update, download models, and configure settings
VRAM-aware model catalog โ 7 model families with quantization recommendations based on your GPU
GPU auto-detection โ NVIDIA (CUDA 13.0/12.8), AMD (ROCm/DirectML), Apple Silicon (MPS)
SageAttention 2 + 3 โ pre-compiled wheels including RTX 50XX Blackwell support
One-click update โ update ComfyUI, all nodes, and dependencies with a single command
Model security scanner โ detects malicious pickle code in .ckpt/.pt files
Junction architecture โ models and outputs persist independently from ComfyUI updates
Docker ready โ 4 image variants including a cloud version with JupyterLab for RunPod
๐ Prerequisites
Git
GPU: NVIDIA (CUDA 12.x+), AMD (Radeon RX 6000+), or Apple Silicon (M1+)
Internet connection
Note: Python is automatically installed via uv if not present. No manual Python setup required.
๐จ Model Catalog (7 Families)
Interactive model downloader with VRAM-based recommendations (โ markers) and SHA-256 integrity checks. Each bundle offers multiple quantization variants (fp16, fp8, GGUF Q3โQ8). Downloads are accelerated via aria2c with HuggingFace + ModelScope fallback:
FLUX (Image): Dev, Fill
Z-IMAGE (Image): Turbo
WAN 2.1 (Video): T2V, I2V 480p
WAN 2.2 (Video): I2V, Fun Inpaint, Fun Camera
HiDream (Image): Dev
QWEN (Image Edit): Image Edit
LTX-2 (Video + Audio): Dev
๐งฉ 34 Custom Nodes Included
Additive manifest โ never removes user-installed nodes.
Core (always installed): ComfyUI-Manager
UmeAiRT Tier: ComfyUI-UmeAiRT-Sync, ComfyUI-UmeAiRT-Toolkit, ComfyUI-Crystools, ComfyUI-nunchaku
Full Tier (all of the above +): ComfyUI-Impact-Pack, ComfyUI-Impact-Subpack, ComfyUI-GGUF, ComfyUI-mxToolkit, ComfyUI-Custom-Scripts, ComfyUI-KJNodes, ComfyUI-WanVideoWrapper, ComfyUI-VideoHelperSuite, ComfyUI-Frame-Interpolation, rgthree-comfy, ComfyUI-Easy-Use, ComfyUI-HunyuanVideoMultiLora, ComfyUI-Florence2, ComfyUI-MultiGPU, ComfyUI-WanStartEndFramesNative, ComfyUI-Image-Saver, ComfyUI_UltimateSDUpscale, comfyui_controlnet_aux, x-flux-comfyui, ComfyUI-Detail-Daemon, wlsh_nodes, ComfyUI_essentials, ComfyUI-wanBlockswap, Derfuu_ComfyUI_ModdedNodes, ComfyUI_LayerStyle, ComfyUI-Upscaler-Tensorrt, comfyui-vrgamedevgirl, comfyui-int-and-float, was-node-suite-comfyui
โ๏ธ GPU Optimizations (Auto-Installed)
PyTorch 2.10: CUDA 13.0/12.8, ROCm 7.1, DirectML, MPS
xformers: Memory-efficient attention
Triton: triton-windows / triton (Linux)
SageAttention 2: Unified ABI3 wheels (Windows), per-arch SM80โSM100 (Linux)
SageAttention 3: RTX 50XX Blackwell native (Windows + Linux)
FlashAttention: Linux + NVIDIA only
Nunchaku & InsightFace: Pre-compiled wheels
Additional Python packages auto-installed: facexlib, onnxruntime-gpu, nvidia-ml-py, cupy-cuda13x, imageio-ffmpeg, hf_xet, cython, rotary_embedding_torch, blend_modes, segment_anything, gguf, and more.
๐ณ Docker Support
Requires Docker and an NVIDIA GPU: docker run --gpus all -p 8188:8188 -v comfyui-data:/data registry.gitlab.com/umeairt-studio/comfyui-auto_installer-python:latest
latest: ~4 GB โ Ready to go with pre-installed PyTorch
latest-cloud: ~4.5 GB โ + JupyterLab for RunPod / cloud
latest-lite: ~2 GB โ Minimal (installs PyTorch on first run)
latest-lite-cloud: ~2 GB โ Lite + JupyterLab
๐ Security
No external script execution โ all logic is internalized
Secure subprocess calls โ no shell=True
HTTPS only โ all URLs validated
SHA-256 integrity checks on all model downloads
Pickle model scanner โ detects malicious code in .ckpt/.pt files
Zip-slip prevention on archive extraction
CI runs Bandit + pip-audit on every push
๐ Post-Installation
Three launcher scripts are generated:
UmeAiRT-Start-ComfyUI: Launch (Performance mode + SageAttention)
UmeAiRT-Start-ComfyUI_LowVRAM: Launch with --lowvram --fp8 for โค8 GB VRAM
UmeAiRT-Manager: TUI manager (update, download, reinstall, settings)
๐ Links
Source code: GitLab (https://gitlab.com/UmeAiRT-Studio/ComfyUI-Auto_installer-Python)
Mirror: Codeberg (https://codeberg.org/UmeAiRT)
Ecosystem: UmeAiRT Studio (https://umeai.art)
Description
fix: locale compatibility & Miniconda version pin (v5.1.27)
FAQ
Comments (29)
Hello, this message is to make a very important recommendation. Yesterday, when I tried to update my Comfy version with the Umeairt updater, it threw an error and completely corrupted my installation. So, I decided to reinstall over my existing installation. While it was doing this, the installation recognized what I already had installed and started recognizing my saved files. Unfortunately, my computer shut down and the installation didn't finish (very bad luck). When I ran it again, it warned me that I had a pending installation and asked if I wanted to delete it. I thought it would delete what the previous installation had done and start over, but it deleted everything I had (models, LORAs, workflows, everythingโI was devastated). The worst part is that when I deleted it via the command prompt, the deleted files weren't sent to the Recycle Bin. Aside from the time it takes to download everything again, it doesn't affect me since I was using the instance for testing, but for someone else, it could be a big problem. That's why I would like to see a notification about this deletion. Any explanation with precautions for reinstalling or repairing already installed instances?
I'm truly sorry for the loss of your data. It's difficult to predict every possible outage, and you'll agree that a PC crash is hard to anticipate. The best approach is always to perform a clean installation in a new folder and simply move the model and output folders, for example, to preserve them.
I added this message to confirm the deletion, but perhaps it wasn't explicit enough. Do you have any suggestions?
lol v5.2 and still not supprted on rtx 4050
lol it work on my 3060 mobile. If you dont give me any information on the problem i cant help you
Works (Actually ROCKS) on my 4060 TI 16GB and tries to work on my 5070 12GB in multi-gpu but still tweaking it. Easy Peasy :)
failed at step 7, some kind of error on my amd gpu lol. truly automatic XD
Can you send me the logs so I can see where the error is coming from? I don't have an AMD GPU to run a full test.
@UmeAiRTย yeah, on AMD GPU Install all time failed [confirmed], i don't have log, so i can't send it to you. i try one time to set up you Comfy installer...in last 2 days ago before re-setup my windows
@UmeAiRTย
here you go
====================================
| [Step 1/13] System Configuration |
====================================
-> Long path support already enabled.
======================================
| [Step 2/13] Checking Prerequisites |
======================================
-> Git: git version 2.53.0.windows.2
- Downloading aria2 (download accelerator)...
-> Downloading "aria2-1.37.0-win-64bit-build1.zip"
aria2.zip โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ 2.5/2.5 MB 405.4 kB/s 0:00:00
-> Extracting aria2...
-> aria2 installed to scripts/aria2/.
===========================================
| [Step 3/13] Creating Python Environment |
===========================================
- Install path: C:\Users\sanef\ComfyUI
-> Virtual environment already exists.
-> Venv python: C:\Users\sanef\ComfyUI\scripts\venv\Scripts\python.exe
==========================================
| [Step 4/13] Provisioning Configuration |
==========================================
-> dependencies.json: copied
-> Downloading "model_manifest.json"
-> model_manifest.json: downloaded from Assets repo
-> Downloading "tools_manifest.json"
-> tools_manifest.json: downloaded from Assets repo
- 1 config file(s) provisioned.
==================================
| [Step 5/13] Setting Up ComfyUI |
==================================
- Configuring Git for long paths...
-> Git long paths configured.
-> ComfyUI directory already exists.
=============================================
| [Step 6/13] External Folders Architecture |
=============================================
-> Linked custom_nodes โ custom_nodes (External)
-> Linked models โ models (External)
-> Linked output โ output (External)
-> Linked input โ input (External)
-> Linked user โ user (External)
=================================
| [Step 7/13] Core Dependencies |
=================================
- Checking for AMD GPU...
-> AMD GPU detected: AMD Radeon RX 6800 XT
-> AMD GPU detected.
-> Using Windows AMD configuration: directml
- Installing PyTorch (torch, torchvision, torchaudio, torch-directml) ...
- Command failed with code 1.
- Command: C:\Users\sanef\ComfyUI\scripts\.bootstrap\Scripts\uv.exe pip install --python
C:\Users\sanef\ComfyUI\scripts\venv\Scripts\python.exe torch torchvision torchaudio torch-directml
โญโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ Traceback (most recent call last) โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฎ
โ C:\Users\sanef\AppData\Local\Temp\ComfyUI-Auto_installer\src\cli.py:205 in install โ
โ โ
โ 202 โ โ ) from e โ
โ 203 โ โ
โ 204 โ path = cleanpath(path) โ
โ โฑ 205 โ run_install( โ
โ 206 โ โ path, install_type_enum, โ
โ 207 โ โ verbose=verbose, node_tier=node_tier_enum, โ
โ 208 โ โ cuda_version=cuda_version, skip_nodes=skip_nodes, โ
โ โ
โ C:\Users\sanef\AppData\Local\Temp\ComfyUI-Auto_installer\src\installer\install.py:226 in run_install โ
โ โ
โ 223 โ โ
โ 224 โ cuda_tag = detect_and_select_gpu(platform, deps, log, cuda_override=cuda_version) โ
โ 225 โ โ
โ โฑ 226 โ install_core_dependencies(python_exe, comfy_path, deps, log, cuda_tag=cuda_tag) โ
โ 227 โ โ
โ 228 โ # โโ Step 8: Python Packages โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ
โ 229 โ log.step("Installing Python Packages") โ
โ โ
โ C:\Users\sanef\AppData\Local\Temp\ComfyUI-Auto_installer\src\installer\dependencies.py:79 in โ
โ install_core_dependencies โ
โ โ
โ 76 โ โ โ
โ 77 โ โ torch_pkgs = torch_cfg.packages.split() โ
โ 78 โ โ log.item(f"Installing PyTorch ({', '.join(torch_pkgs)}) [{cuda_tag}]...") โ
โ โฑ 79 โ โ uv_install( โ
โ 80 โ โ โ python_exe, โ
โ 81 โ โ โ torch_pkgs, โ
โ 82 โ โ โ index_url=torch_cfg.index_url, โ
โ โ
โ C:\Users\sanef\AppData\Local\Temp\ComfyUI-Auto_installer\src\utils\packaging.py:176 in uv_install โ
โ โ
โ 173 โ โ
โ 174 โ log.info(f"uv {' '.join(args)}") โ
โ 175 โ โ
โ โฑ 176 โ return run_and_log( โ
โ 177 โ โ uv_path, โ
โ 178 โ โ args, โ
โ 179 โ โ ignore_errors=ignore_errors, โ
โ โ
โ C:\Users\sanef\AppData\Local\Temp\ComfyUI-Auto_installer\src\utils\commands.py:99 in run_and_log โ
โ โ
โ 96 โ โ โ if result.stderr: โ
โ 97 โ โ โ โ for line in result.stderr.strip().split("\n")[:20]: โ
โ 98 โ โ โ โ โ log.error(line, level=3) โ
โ โฑ 99 โ โ โ raise CommandError(cmd_display, result.returncode, result.stderr) โ
โ 100 โ โ โ
โ 101 โ โ return result โ
โ 102 โ
โฐโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโฏ
CommandError: Command failed (code 1): C:\Users\sanef\ComfyUI\scripts\.bootstrap\Scripts\uv.exe pip install --python
C:\Users\sanef\ComfyUI\scripts\venv\Scripts\python.exe torch torchvision torchaudio torch-directml
@Ghost_Nameย same error? check mine
@sanefer0364ย yeah, same man
@sanefer0364ย @Ghost_Nameย I tried to fix itโcould you try again? I don't have an AMD GPU, so it's hard for me to test specific features. If you have the time and are willing, send me a message on Discord at โ_umeโ so we can try to test AMD mode with a real-world test.
When I try to use your UmeAiRT-Manager.bat to launch Comfy, in the list where I need to select the VRAM Mode (Performance/Normal/Low), nothing happens after I press Enter. This happens regardless of which option I choose.
im currently using cuda 13.0 but the manager show its running on cuda 12.8, is there a way to force it to run on cuda 13? It must be why some custom nodes fails to import.
You may have multiple versions of CUDA installed, which can create conflicts. Try uninstalling all versions of CUDA and then reinstalling only version 13.0.
- 34 nodes in manifest
-> ComfyUI-Manager: cloning...
- Command timed out after 300s: git clone
https://github.com/ltdrdata/ComfyUI-Manager.git
F:\AiGenerator\COMFYUI\ComfyUI\custom_nodes\ComfyUI-Manager
-> ComfyUI-Manager: retry 2/3 (shallow)...
- Command failed with code 128.
- Command: git clone --depth 1 https://github.com/ltdrdata/ComfyUI-Manager.git
F:\AiGenerator\COMFYUI\ComfyUI\custom_nodes\ComfyUI-Manager
-> ComfyUI-Manager: retry 3/3 (shallow)...
- Command failed with code 128.
- Command: git clone --depth 1 https://github.com/ltdrdata/ComfyUI-Manager.git
F:\AiGenerator\COMFYUI\ComfyUI\custom_nodes\ComfyUI-Manager
-> ComfyUI-Manager: Clone FAILED after 3 attempts (Command failed (code
128): git clone --depth 1 https://github.com/ltdrdata/ComfyUI-Manager.git
F:\AiGenerator\COMFYUI\ComfyUI\custom_nodes\ComfyUI-Manager)
It's a network problem: he tried 3 times to clone comfyui manager from github and he didn't succeed
Switch from 4.4 to 5.2 - Holy ... Super fast Installation, No Errors - Works fine with my 5090! Better VRAM Management, Faster BF16 WAN2.2 Generation... Thanks for this Masterpiece.
I worked hard on version 5.0. Glad to hear it works well. I added special optimizations for 50XX series cards.
Hello, thanks for your work. It run's for me. But one question: In step 6 (External folders architecture) i get a runtime error because the drive isn't a NTFS volume. I want to install ComfyUI on my SD-Card (FAT32). Alternativ i installed ComfyUI on my c: - Drive. But why is a NTFS volume necessary?
Hello, and yes, on Windows NTFS is essential for my installation because I create links between folders, and this function doesn't work with FAT32. Furthermore, with FAT32, the file size limit is only 4GB, and most template files exceed this size
Wow, dude. Started using your install scripts a little over a year ago and you continue to one-up yourself. Thank you.
Best installer ever, thank you!!!
I have an old installation from the beginning of the year made with the previous version of this installer. Can I use this one to install in a different directory and it will not impact my old comfy ui installation?