🚀 ComfyUI Auto-Installer — 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
⚡ 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 — everything else (Python, uv, dependencies) is handled automatically.
✨ What's New in v5
Full Python rewrite — no more PowerShell dependency
Cross-platform — Windows, Linux, macOS, and Docker
TUI Manager — interactive terminal UI to launch, update, download models, and configure settings
VRAM-aware model catalog — 7 model families with quantization recommendations based on your GPU
GPU auto-detection — NVIDIA (CUDA 13.0/12.8), AMD (ROCm/DirectML), Apple Silicon (MPS)
SageAttention 2 + 3 — pre-compiled wheels including RTX 50XX Blackwell support
One-click update — update ComfyUI, all nodes, and dependencies with a single command
Model security scanner — detects malicious pickle code in .ckpt/.pt files
Junction architecture — models and outputs persist independently from ComfyUI updates
Docker ready — 4 image variants including a cloud version with JupyterLab for RunPod
📋 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.
🎨 Model Catalog (7 Families)
Interactive model downloader with VRAM-based recommendations (★ markers) and SHA-256 integrity checks. Each bundle offers multiple quantization variants (fp16, fp8, GGUF Q3→Q8). 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
🧩 34 Custom Nodes Included
Additive manifest — 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
⚙️ 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 SM80–SM100 (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.
🐳 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 — Ready to go with pre-installed PyTorch
latest-cloud: ~4.5 GB — + JupyterLab for RunPod / cloud
latest-lite: ~2 GB — Minimal (installs PyTorch on first run)
latest-lite-cloud: ~2 GB — Lite + JupyterLab
🔒 Security
No external script execution — all logic is internalized
Secure subprocess calls — no shell=True
HTTPS only — all URLs validated
SHA-256 integrity checks on all model downloads
Pickle model scanner — detects malicious code in .ckpt/.pt files
Zip-slip prevention on archive extraction
CI runs Bandit + pip-audit on every push
📂 Post-Installation
Three launcher scripts are generated:
UmeAiRT-Start-ComfyUI: Launch (Performance mode + SageAttention)
UmeAiRT-Start-ComfyUI_LowVRAM: Launch with --lowvram --fp8 for ≤8 GB VRAM
UmeAiRT-Manager: TUI manager (update, download, reinstall, settings)
🔗 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
New version based on Conda instead of venv.
FAQ
Comments (61)
Sadly got this error. V4.0 of the install script:
- ERREUR: La commande a échoué avec le code 1.
- Commande: C:\Users\W11\AppData\Local\aria2\aria2c.exe --console-log-level=warn --quiet=true -x 16 -s 16 -k 1M --dir="D:\AI\ComfyUI\scripts" --out="insightface-0.7.3-cp312-cp312-win_amd64.whl" "https://github.com/UmeAiRT/ComfyUI-Auto_installer/raw/main/whl/insightface-0.7.3-cp312-cp312-win_amd64.whl"
- Sortie de l'erreur:
- ERREUR FATALE lors de la tentative d'exécution: C:\Users\W11\AppData\Local\aria2\aria2c.exe --console-log-level=warn --quiet=true -x 16 -s 16 -k 1M --dir="D:\AI\ComfyUI\scripts" --out="insightface-0.7.3-cp312-cp312-win_amd64.whl" "https://github.com/UmeAiRT/ComfyUI-Auto_installer/raw/main/whl/insightface-0.7.3-cp312-cp312-win_amd64.whl". Erreur: L'exécution de la commande a échoué. Vérifiez les logs.
Une erreur fatale est survenue. Appuyez sur Entrée pour quitter.:
(UmeAiRT) D:\AI\ComfyUI>echo End of Phase 2. Press Enter to close this window.
End of Phase 2. Press Enter to close this window.
Have you tried running the script "as administrator"? Because the link works correctly :/
@UmeAiRT yup. Running as admin. If I go and download the script in question by hand it keeps going but then will have an error with another script and I need to repeat it over and over
@thisisarandomaccount2025 i have try a fix, can you retry?
@UmeAiRT Sure! Although I got it working by now
@thisisarandomaccount2025 how did you get it working?
@dlfoid23 basically every time the install script failed due to a failed download, I'd download the file it was complaining about, move it to the scripts folder and restart the script
The v4.0 install scripts works without a Error! Nice Work. But idk why the v4 version use more VRAM then 3.2 (same workflow)
v3.2 WAN2.2 Generation: 27,8gb
v4.0 WAN2.2 Generation: 31,9gb OOM
Appeared to completely install correctly but I can't run Comfy, I get this error 'To use the --use-sage-attention feature, the sageattention package must be installed first.'
NM, I just tried to rerun the script this time I sat and watched it. It's saying Xformers and Sage install failed. Oh well. Thanks anyway.
@yajukun Can you send me the install_logs file?
@UmeAiRT Hi, yeah, let me run it again, I deleted the folder yesterday. I think...this may be due to the fact that I am now using a 5090 Asus SCAR laptop? I miss my 4090 desktop, this 5090 has been nothing but trouble when it comes to Forge, ComfyUI, etc. I bet a lot of the issues people are having this year are due to 50 series chips.
@UmeAiRT Hi, I sent you the link using the sites chat function. Thanks!
@UmeAiRT
same. & when I try to install sage I get this error:
***********************************
C:\Users\user\AppData\Local\Miniconda3\envs\UmeAiRT>python.exe -m pip install sageattention
Collecting sageattention
Downloading sageattention-2.2.0.tar.gz (67 kB)
Installing build dependencies ... done
Getting requirements to build wheel ... error
error: subprocess-exited-with-error
× Getting requirements to build wheel did not run successfully.
│ exit code: 1
╰─> [18 lines of output]
Traceback (most recent call last):
File "C:\Users\user\AppData\Local\Miniconda3\envs\UmeAiRT\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 389, in <module>
main()
File "C:\Users\user\AppData\Local\Miniconda3\envs\UmeAiRT\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 373, in main
json_out["return_val"] = hook(**hook_input["kwargs"])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\user\AppData\Local\Miniconda3\envs\UmeAiRT\Lib\site-packages\pip\_vendor\pyproject_hooks\_in_process\_in_process.py", line 143, in get_requires_for_build_wheel
return hook(config_settings)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\user\AppData\Local\Temp\pip-build-env-eb3gaiol\overlay\Lib\site-packages\setuptools\build_meta.py", line 332, in get_requires_for_build_wheel
return self._get_build_requires(config_settings, requirements=[])
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\user\AppData\Local\Temp\pip-build-env-eb3gaiol\overlay\Lib\site-packages\setuptools\build_meta.py", line 302, in getbuild_requires
self.run_setup()
File "C:\Users\user\AppData\Local\Temp\pip-build-env-eb3gaiol\overlay\Lib\site-packages\setuptools\build_meta.py", line 318, in run_setup
exec(code, locals())
File "<string>", line 36, in <module>
ModuleNotFoundError: No module named 'torch'
[end of output]
note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'sageattention' when getting requirements to build wheel
@dlfoid23 You cant install sageattention with pip you have to build it from source
@UmeAiRT ok, I mistakenly thought the install did that. I had one of your previous installs working until I updated it. I can't seem to get any of them to function now.
This error is happening to me too. unable to install it. Previosly on the old version3.2 i was able too, but even the old one now automatically installs the new installer4.0. hoping for a fix
@wutnutt167 Yeah, I tried to DL the v3.2 WF to see if I could get that working. But like you said, it looks like they are all pointed to the same v4 resources. Tis ok, I already have a working ComfyUI with sage working with my 5090m already but it's a bit clunky. I trust @UmeAiRT WF more and he offers support. I swear this 5090m has been nothing but trouble however, it was the only way for me to get 24GB vram in a laptop. I get the feeling 50 series is already prone to issues and having the mobile version just makes things worse. Good luck.
@yajukun i got it to work. look at my comment i posted earlier today. might help you.
@wutnutt167 Hey, actually I saw that but I already had Cuda v12.9 installed. However, I also had Cuda 12.8 installed also and I didn't uninstall the old version. I guess I'll try to uninstall everything, remove the folder and reinstall 12.9. Thanks for the tip!
@wutnutt167 Already had 12.9 I may have to try installing sageattention manually. which is strange because V3.2 seemed to work without me having to do that. It may not fix it though.
@UmeAiRT Is there some way I can do this manually?
@dlfoid23 +++
For anyone getting Permissions error on v4.0, run the install script as admin.
getting this error
========================================================= | [Step 1/9] Setting up Miniconda and Conda Environment | =========================================================
Miniconda not found. Installing...-> File 'Miniconda3-latest-Windows-x86_64.exe' already exists. Skipping download.[INFO] Executing: C:\Users\USER\AppData\Local\Temp\Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /D="C:\Users\USER\AppData\Local\Miniconda3"
ERREUR: La commande a échoué avec le code 2.,
Commande: C:\Users\USER\AppData\Local\Temp\Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /D="C:\Users\USER\AppData\Local\Miniconda3",
Sortie de l'erreur:,
,
Welcome to Miniconda3 py313_25.9.1-3 ::error:: Directory 'C:\Users\USER\miniconda3' is not empty, please choose a different location.
ERREUR FATALE lors de la tentative d'exécution: C:\Users\USER\AppData\Local\Temp\Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /D="C:\Users\USER\AppData\Local\Miniconda3". Erreur: L'exécution de la commande a échoué. Vérifiez les logs.,
Une erreur fatale est survenue. Appuyez sur Entrée pour quitter.:
Ditto, hopefully we get an update
for those having errors like:
python.exe : ERROR: Could not find a version that satisfies the requirement torch==2.8.0+cu129 (from versions: none)
check if you are being geo blocked, especially if you are from some shithole country, like I am.
even if you aren't living in a shithole, your IP might still be blocked by the assholes at AWS or Cloudflare. or, maybe, even by assholes at Pytorch or HF.
TRY CHANGING YOUR IP ADDRESS!
P.S.
Also, even if you can see the wheel list (e.g. https://download.pytorch.org/whl/cu128) in your browser just fine, pip can still get blocked because the CDN treats pip requests differently.
Browser works → pip gets nothing. Don’t trust the browser check alone.
author of the script should also implement some workarounds/fallbacks in his script. ChatGPT gave me like 5 different workarounds: imitate a browser by using different user-agent, using curl/wget instead of pip, fallback to using pip download/install commands, etc)
When attempting to install ComfyUI, the heur:trojan.bat.alien.gen Trojan is detected.
I have the same.
@SVS13121975224 You can find the complete source code here: https://github.com/UmeAiRT/ComfyUI-Auto_installer
And a VirusTotal scan : https://www.virustotal.com/gui/file-analysis/M2ZjMWIzYzhhODkwNmUzODlmZGQ3YWU3ODBiNmNmYmM6MTc2NTY0NzE0Nw==
I think it's just the fact that it's a script that triggers the false positive.
v4.0 is portable comfyui? Why is it installing stuff outside of portable location?
Will TensorRT Interpolation be included in the future?
torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
ImportError: cannot import name 'triton_key' from 'triton.compiler.compiler' (C:\Users\sps02\AppData\Local\Miniconda3\envs\UmeAiRT\Lib\site-packages\triton\compiler\compiler.py)
Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"
How do i fix this?
Its because the script install Pytorch 2.8.0 but install triton-windows 3.5.* - so downgrade triton-windows to 3.4.*
Hello, I have this error. Any ideas what to do ?
EnvironmentNameNotFound: Could not find conda environment: UmeAiRT
You can list all discoverable environments with conda info --envs.
Failed to run 'conda activate UmeAiRT'.
FAILED to activate the Conda 'UmeAirT' environment.
Press any key to continue . . .
Loved it until your updates messed up my whole comfy installation. What is up with this conda bullshit? I am sorry but i am so pissed off
Same here. Everything was perfect until this update. Howerver I do undertstand these things are not very beginner-friendly to begin with and UmeAiRT has put a ton of hard work into this, so I appreciate it.
man this update ruined everthing i'm getting errors everywhere and i cant go back to the last 3.2 update cuz its the same, please upload the old instalation method its faster and more stable
this one is 3 times slower than before
need to have only cuda 12.9 installed, directories included and add path to cuda. works ok after that.
@videoadanai543 how can i add path to cuda
I tried to reintroduce the venv method in version 4.1, which is currently in beta.
@delta45424155 I don't see the connection between a guide to install SageAttention and my script?
Anyone having error with conda try these step it worked for me maybe it will for you too.
1. Go to C:\Users\user name\miniconda3\ and copy all files.
2. Go to C:\Users\Musab Khan\AppData\Local\ and create a new folder and rename it to Miniconda3 and paste all files in Miniconda3.
Don't know if step 3 is necessary but i did it.
3. Find _conda.exe in Miniconda3 and create a shortcut of it there and rename it to conda.exe
After these it worked for me maybe it will work for you too.
The Light install with the venv is really nice. If you could select a custom name for the Conda virtual env's would be cool too. Just installed. Thank you!
looks good, what's holdin you back from a pytorch 2.9, cuda 13 update 2, python 3.13.9?
Lot of thing dont work with CUDA 13 and Py 2.9 like Sageattention
Seems to work well enough...
pytorch version: 2.9.1+cu130
xformers version: 0.0.33.post2
Device: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Using sage attention
Python version: 3.13.9 (tags/v3.13.9:8183fa5, Oct 14 2025, 14:09:13) [MSC v.1944 64 bit (AMD64)]
ComfyUI version: 0.3.77
ComfyUI frontend version: 1.32.10
pytorch version: 2.9.1+cu130
@Fferrett Keep in mind that this type of installer is designed to work with many configurations and not just with the latest generation of graphics cards. And it's not enough to just get ComfyUI to start; you also need to get all the custom nodes I use in over 50 different workflows working.
@UmeAiRT
Working perfectly fine here on RTX4090 (upgraded my comfyui, originally installed with your script, which failed to install attentions anyway, so I decided to upgrade everything):
Python: 3.13.9 / ComfyUI: 0.4 + 1.32.10
pytorch: 2.10.0.dev20251207+cu130
Nunchaku: 1.1.0.dev20251111
Sageattention: v2.2.0-windows.post4
latest pytorch: "pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu130"
Nunchaku: https://github.com/nunchaku-tech/nunchaku/releases
Sageattention: https://github.com/woct0rdho/SageAttention/releases and "pip install triton-windows"
didn't try xformers and flash attention though. not sure if anyone would want them at all, when better alternatives exist.
@firsak I published an update today; the only thing I haven't modified yet is Python. But personally, the compilation isn't working; I had to use precompiled whl.
@UmeAiRT Some have claimed that Python newer than 3.10 is slower. I don't know if that's the case, though.
@ss9999 I haven't noticed any particular slowdown at the moment, and some modules are compatible with certain versions of Python, making it quickly complicated to find the "perfect" version.
Getting 500 connection error
- ERREUR: La commande a échoué avec le code 1.
- Commande: C:\Users\Razgrits\AppData\Local\Miniconda3\Scripts\conda.exe env create -f "B:\AI\comfy\scripts\environment.yml"
- Sortie de l'erreur:
3 channel Terms of Service accepted
Retrieving notices: done
Channels:
- conda-forge
- pytorch
- nvidia
- defaults
Platform: win-64
Collecting package metadata (repodata.json): failed
conda.exe :
At line:1 char:1
+ & "C:\Users\Razgrits\AppData\Local\Miniconda3\Scripts\conda.exe" env ...
+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ CategoryInfo : NotSpecified: (:String) [], RemoteException
+ FullyQualifiedErrorId : NativeCommandError
CondaHTTPError: HTTP 500 Internal Server Error for url <https://conda.anaconda.org/conda-forge/win-64/repodata.json>
Elapsed: 00:07.857717
CF-RAY: 9a9267074a2a0b7e-ATH
A remote server error occurred when trying to retrieve this URL.
A 500-type error (e.g. 500, 501, 502, 503, etc.) indicates the server failed to
fulfill a valid request. The problem may be spurious, and will resolve itself if you
try your request again. If the problem persists, consider notifying the maintainer
of the remote server.
- ERREUR FATALE lors de la tentative d'exécution: C:\Users\Razgrits\AppData\Local\Miniconda3\Scripts\conda.exe env create -f "B:\AI\comfy\scripts\environment.yml". Erreur: L'exécution de la commande a échoué. Vérifiez les logs.
Install a miniconda from the repo compatible with your python installation.
both methods of version 4.1 install but I always get this error when running the wanvideosampler node:
backend='inductor' raised:
ImportError: cannot import name 'triton_key' from 'triton.compiler.compiler' (E:\Workspace\Ume41\scripts\venv\Lib\site-packages\triton\compiler\compiler.py)
I never use this sampler, but this error seems to originate from the Triton module. The Windows version of this module may be incompatible with certain nodes.
The script install wrong Combination of PyTorch and Triton-Windows... I needed a full reinstall and found: Script install Torch 2.8.0 but triton-windows 2.5.1- just downgrade triton-windows to 2.4.*
@SysDeep I made a mistake between two updates. Everything is now running CUDA 13.0, PyTorch 2.9.1, and therefore Nunchaku 2.5. I just published the complete update.
@UmeAiRT Nice i will try it out :)