Skip to content
View in the app

A better way to browse. Learn more.

Unraid

A full-screen app on your home screen with push notifications, badges and more.

To install this app on iOS and iPadOS
  1. Tap the Share icon in Safari
  2. Scroll the menu and tap Add to Home Screen.
  3. Tap Add in the top-right corner.
To install this app on Android
  1. Tap the 3-dot menu (⋮) in the top-right corner of the browser.
  2. Tap Add to Home screen or Install app.
  3. Confirm by tapping Install.

[Support] Comfyui (Nvidia) Docker

Featured Replies

22 hours ago, Sebastian.d88 said:

Hi, I found the Problem. It is the old Nvidia driver 580 in the Unraid Nvidia Plugin. Actually I can't get the version 590 to work with my 5070ti. Currently, I'm passing the GPU through to a VMware and everything works fine.

Blackwell cards are only supported by the open source driver on unraid, so 580.

  • Replies 173
  • Views 91.8k
  • Created
  • Last Reply

Top Posters In This Topic

Most Popular Posts

  • Hi, i update my frontend with the the command comfyui shows you when you start the container with an old frontend. You find the command in the container protocol. In my case the command is /comfy/mnt/

  • Example runs

  • Hello, I have submitted to CA about a week ago. I understand the team has to do a review before allowing new content, so I hope it will be released shortly.   The wrapper code is public

Posted Images

I noticed a new TORCH_LOCK field was added to the template and set to required, the README makes it seem like this is optional.
If you disable the required flag it will fail to start but if you delete the parameter it comes up fine.
Right now this forces the user to set it correctly or delete it from the template which was confusing to me as that's not stated in the README

*EDIT: I also notice on fresh install the config.ini network_mode is set to public instead of personal_cloud so you can't install nodes
They also moved it so their wiki is outdated.. I found it under ./user/_manager/ not ./user/default/ComfyUI-Manager/

Edited by melmurp

Hi!

I'm getting this alerts, how do i update or get rid of them?

Version Compatibility Warning: Frontend version 1.25.11 is outdated. Backend requires 1.36.14 or higher. Visit https://docs.comfy.org/installation/update_comfyui#common-update-issues for update instructions.


Legacy ComfyUI-Manager data migrated. See terminal for details.

thank you!

Edited by renk1

21 minutes ago, renk1 said:

Hi!

I'm getting this alerts, how do i update or get rid of them?

Version Compatibility Warning: Frontend version 1.25.11 is outdated. Backend requires 1.36.14 or higher. Visit https://docs.comfy.org/installation/update_comfyui#common-update-issues for update instructions.

Legacy ComfyUI-Manager data migrated. See terminal for details.


thank you!

Can happens if you touched the requirements.txt file and it doesn't match git repo anymore... comfyui then refuses to upgrade properly
Quick fix is to update comfyui-frontend-package== to higher version then restart comfyui.
I think I just deleted the file then did an update and it pulled fresh

I didn't touch any file, I just updated and got those errors, but I re-ran the update and it seems to have fixed it.

Thanks!

Edited by renk1

  • Author

I noticed a new TORCH_LOCK field was added to the template and set to required, the README makes it seem like this is optional.
If you disable the required flag it will fail to start but if you delete the parameter it comes up fine.
Right now this forces the user to set it correctly or delete it from the template which was confusing to me as that's not stated in the README

*EDIT: I also notice on fresh install the config.ini network_mode is set to public instead of personal_cloud so you can't install nodes
They also moved it so their wiki is outdated.. I found it under ./user/_manager/ not ./user/default/ComfyUI-Manager/

This is something I was worried about AFTER I released the update.

I am going to remove it from the template for now and will fix it in a future update.

For fresh installs you still need to start the container twice before it can have modified the settings (1st load) and used them (2nd load).

On 1/25/2026 at 7:51 PM, melmurp said:

Blackwell cards are only supported by the open source driver on unraid, so 580.

Hi, thanks for your response. I need the driver version 590 to get this card run in the docker container. The card doesn't work in the Container with the open source driver 580. In tested it in my VM. With driver 580 I get the coda error too, but installing driver 590 solves this problem.

16 hours ago, Sebastian.d88 said:

Hi, thanks for your response. I need the driver version 590 to get this card run in the docker container. The card doesn't work in the Container with the open source driver 580. In tested it in my VM. With driver 580 I get the coda error too, but installing driver 590 solves this problem.

I have a 5070 TI working with this docker, which Cuda version are you on?

Support was added in 12.8 I think

Tue Jan 27 21:41:33 2026

+-----------------------------------------------------------------------------------------+

| NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0 |

+-----------------------------------------+------------------------+----------------------+

| Processes: |

| GPU GI CI PID Type Process name GPU Memory |

| ID ID Usage |

|=========================================================================================|

| 0 N/A N/A 2197523 C /comfy/mnt/venv/bin/python3 336MiB |

+-----------------------------------------------------------------------------------------+

Tensor compile is busted but that's an issue with cuda itself I believe.

I'm having an issue where comfyui randomly dies. I can get a couple generations in before the thing fails. Anyone run into this?

I see this in container logs:

loaded completely; 17458.07 MB usable, 4565.35 MB loaded, full load: True

0%| | 0/8 [00:00<?, ?it/s]/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [6,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [7,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [8,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [9,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [10,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:163: operator(): block: [5,0,0], thread: [11,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "scatter gather kernel index out of bounds" failed.

terminate called after throwing an instance of 'c10::AcceleratorError'

what(): CUDA error: device-side assert triggered

Search for `cudaErrorAssert' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.

CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.

For debugging consider passing CUDA_LAUNCH_BLOCKING=1

Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Exception raised from free_impl at /pytorch/c10/cuda/CUDAMallocAsyncAllocator.cpp:206 (most recent call first):

frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >) + 0x9d (0x15379f772fdd in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)

frame #1: <unknown function> + 0xc0e0 (0x15379fbc30e0 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10_cuda.so)

frame #2: <unknown function> + 0x49730 (0x15379fc00730 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10_cuda.so)

frame #3: <unknown function> + 0x4a734 (0x15379fc01734 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10_cuda.so)

frame #4: <unknown function> + 0x7e9d4 (0x15379f7549d4 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)

frame #5: c10::TensorImpl::~TensorImpl() + 0x9 (0x15379f74e369 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libc10.so)

frame #6: <unknown function> + 0x862f45 (0x1536f7262f45 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)

frame #7: <unknown function> + 0x862fe1 (0x1536f7262fe1 in /comfy/mnt/venv/lib/python3.12/site-packages/torch/lib/libtorch_python.so)

frame #8: python3() [0x59efc3]

<omitting python frames>

frame #10: python3() [0x54cd4d]

frame #12: python3() [0x54cd4d]

frame #15: python3() [0x5a33c8]

frame #18: python3() [0x54cd4d]

frame #20: python3() [0x54cd4d]

frame #23: python3() [0x5a33c8]

frame #26: python3() [0x54cd4d]

frame #28: python3() [0x54cd4d]

frame #31: python3() [0x5a33c8]

frame #34: python3() [0x54ce14]

frame #37: python3() [0x54ce14]

frame #41: python3() [0x5a33c8]

frame #44: python3() [0x54ce14]

frame #47: python3() [0x54ce14]

frame #50: python3() [0x54ce14]

frame #53: python3() [0x54cd4d]

frame #55: python3() [0x54ce14]

frame #59: python3() [0x5a33c8]

!! ERROR: ComfyUI failed or exited with an error

!! Exiting script (ID: 69)

/comfyui-nvidia_init.bash: line 933: 455 Aborted ${COMFY_CMDLINE_BASE} ${COMFY_CMDLINE_EXTRA}

!! ERROR: subscript failed

!! Exiting script (ID: 1)

I have a question about the docker, particularly if I can run other apps inside. Works great for training Flux loras.

I stumbled across this project, Ace Step 1.5

This tool generates music, with click of a button. Very well produced music.

GitHub
No image preview

GitHub - ace-step/ACE-Step-1.5: The most powerful local m...

The most powerful local music generation model that outperforms most commercial alternatives - ace-step/ACE-Step-1.5

My 3090 is inside my unraid server. I've been trying to create another docker container for me to install this app. Best way to run AceStep is with UV. It handles all the requirements and Python versions. This particular project uses Python 3.11 and it appears to conflict with ComfyUI docker. I really trying running it inside to see if it works.

I've tried to build a docker with help from AI. No success and it has been trail and error.

Is there any way you can point me to the right direction on what docker repo to use? A repo that sees and utilizes Nvidia hardware inside the docker, much like the ComfyUI docker?

AceStep to me, runs best on Linux, I have to use the smaller models because my main pc has a 3080 10 gig card. Ideally I'd prefer to run this on my Unraid server.

Any assistance would be of great help.

Thanks again, and also, love the ComfyUI docker. Its been awesome.


Edited by unraidrocks

  • Author
10 hours ago, unraidrocks said:

I have a question about the docker, particularly if I can run other apps inside. Works great for training Flux loras.

I stumbled across this project, Ace Step 1.5

This tool generates music, with click of a button. Very well produced music.

GitHub
No image preview

GitHub - ace-step/ACE-Step-1.5: The most powerful local m...

The most powerful local music generation model that outperforms most commercial alternatives - ace-step/ACE-Step-1.5

My 3090 is inside my unraid server. I've been trying to create another docker container for me to install this app. Best way to run AceStep is with UV. It handles all the requirements and Python versions. This particular project uses Python 3.11 and it appears to conflict with ComfyUI docker. I really trying running it inside to see if it works.

I've tried to build a docker with help from AI. No success and it has been trail and error.

Is there any way you can point me to the right direction on what docker repo to use? A repo that sees and utilizes Nvidia hardware inside the docker, much like the ComfyUI docker?

AceStep to me, runs best on Linux, I have to use the smaller models because my main pc has a 3080 10 gig card. Ideally I'd prefer to run this on my Unraid server.

Any assistance would be of great help.

Thanks again, and also, love the ComfyUI docker. Its been awesome.


According to the recent ComfyUI blog post, I believe that the answer is yes

No image preview

ACE-Step 1.5 is Now Available in ComfyUI

Commercial-grade music generation on consumer hardware

You would need to update your comfy using the manager first, but check out the "For ComfyUI Desktop & Local Users" section

Right on, that worked.

only issue is that ComfyUI workflow of music generator doesn't create, direct, orchestrate a prompt into detailed description or generate lyrics instantly. Have to manually add all those details into the workflow. Also have to have some music knowledge, (musician) to know what the hell I'm doing.

http://localhost:7860 (Gradio)

I was hoping for this to work, but it seems will require either it's dedicated docker or mess or break the ComfyUI I use for generating images.

Thanks again though. Didn't know they already have templates.

  • 1 month later...
  • Author
On 2/13/2026 at 1:35 PM, unraidrocks said:

Right on, that worked.

only issue is that ComfyUI workflow of music generator doesn't create, direct, orchestrate a prompt into detailed description or generate lyrics instantly. Have to manually add all those details into the workflow. Also have to have some music knowledge, (musician) to know what the hell I'm doing.

http://localhost:7860 (Gradio)

I was hoping for this to work, but it seems will require either it's dedicated docker or mess or break the ComfyUI I use for generating images.

Thanks again though. Didn't know they already have templates.

I think you might be able to achieve it
The tool only expose the port needed by Comfy but you could add an exported port for 7860 as long as the service is started on that port. You will need to rebuild the tool to add the expose port but that can be done from a Dockerfile "FROM" so luckily fairly easily

Hey there.

I needed to open up my comfyui and wasn't able to, kept crashing. After creating a fresh install to troubleshoot and working with an AI a bit I was able to get it working using a workaround.

Below are the findings and workaround in-case this helps someone else.
The Problem

On a host with a recent NVIDIA driver (595.45.04, CUDA 13.2) and an RTX 4070 Ti SUPER, the container was silently crashing on every start with !! ERROR: ComfyUI failed or exited with an error. No useful output in the Docker logs.

I hit this on both the latest and ubuntu24_cuda13.1-latest image tags.

The actual error (only visible by running ComfyUI manually inside the container) was:

Fatal Python error: Bus error
  File "/comfy/mnt/venv/lib/python3.12/site-packages/comfy_kitchen/backends/cuda/__init__.py", line 84 in <module>

Root Cause
The culprit is comfy_kitchen==0.2.8, which ComfyUI pulls in as a dependency. It ships a pre-compiled CUDA binary that crashes at import time on newer CUDA 13.x host drivers. PyTorch itself works fine — it's purely a comfy_kitchen compatibility issue.
Workaround
Uninstalling comfy_kitchen lets ComfyUI start normally. It gracefully falls back with the message "Failed to import comfy_kitchen, fp8 and fp4 support will not be available" — everything else works fine. Script removes comfy_kitchen on both UV and PIP methods.

To make this permanent across restarts, create a user_script.bash in your run directory aka comfy/mnt/user_script.bash:

bash

#!/bin/bash
echo "== Removing incompatible comfy_kitchen package"
/comfy/mnt/venv/bin/uv pip uninstall comfy_kitchen -y 2>/dev/null || \
/comfy/mnt/venv/bin/pip uninstall comfy_kitchen -y 2>/dev/null || \
echo "== comfy_kitchen not found or already removed, continuing"
exit 0

Make sure it's executable (chmod +x comfy/mnt/user_script.bash) run this on unraid terminal pointed at the right location.

Hope this helps someone.

Edited by SeldomNosePicker

  • 1 month later...

Hi all,

for the life of me I can´t get sageattention installed. And Iḿ not shy to admit that I am not the sharpest knive in the drawer regarding to Linux, env and compiling.

I tried pip3 install (error: externally-managed-environment), gitclone (ModuleNotFoundError: No module named 'torch'), creating a new venv (ERROR: Could not install packages due to an OSError: [Errno 13] Permission denied: '/SageAttention/.env_lin/lib/python3.12/site-packages/trampoline') but nothing worked out.

Can anybody point me to a good manual for this or help in any other way? Any hint is much appreciated!

Edited by FireMcFly

  • Author
On 4/29/2026 at 10:42 AM, FireMcFly said:

Hi all,

for the life of me I can´t get sageattention installed. And Iḿ not shy to admit that I am not the sharpest knive in the drawer regarding to Linux, env and compiling.

I tried pip3 install (error: externally-managed-environment), gitclone (ModuleNotFoundError: No module named 'torch'), creating a new venv (ERROR: Could not install packages due to an OSError: [Errno 13] Permission denied: '/SageAttention/.env_lin/lib/python3.12/site-packages/trampoline') but nothing worked out.

Can anybody point me to a good manual for this or help in any other way? Any hint is much appreciated!

There is a way to do this using a built-in solution from the container itself: the "UserScripts" folder.

https://github.com/mmartial/ComfyUI-Nvidia-Docker#533-userscripts_dir

You would need to add a mount to the location in your unraid template then copy the script you want to enable (here the SageAttention one) then make it executable. During the next restart, the initialization script will see it (check the logs) and attempt to install it for you.

I am giving a brief "how to" as the longer one is listed in the documentation.

Great hint, helped me alot. Somehow I overlook the whole userscript thing. Thanks:)

  • 3 weeks later...

Hi thanks for building this docker image.

Since v 0.20.1 i have the issue that comfyui wont update anymore, it failes when updating within comfy with the following error ;

[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/button.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[DEPRECATION WARNING] Detected import of deprecated legacy API: /scripts/ui/components/buttonGroup.js. This is likely caused by a custom node extension using outdated APIs. Please update your extensions or contact the extension author for an updated version.
[Workflow-Models-Downloader] Loaded 509 models from model-list.json
[WMD] Saved usage cache with 15 models
Error handling request from 192.168.178.71
Traceback (most recent call last):
File "/comfy/mnt/venv/lib/python3.12/site-packages/aiohttp/web_protocol.py", line 517, in handlerequest
resp = await request_handler(request)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/venv/lib/python3.12/site-packages/aiohttp/web_app.py", line 569, in handle
return await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/venv/lib/python3.12/site-packages/aiohttp/webmiddlewares.py", line 117, in impl
return await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/ComfyUI/middleware/cache_middleware.py", line 27, in cache_control
response: web.Response = await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/ComfyUI/server.py", line 86, in deprecation_warning
response: web.Response = await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/ComfyUI/server.py", line 180, in origin_only_middleware
response = await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/venv/lib/python3.12/site-packages/comfyui_manager/init.py", line 128, in manager_middleware
response: web.Response = await handler(request)
^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/venv/lib/python3.12/site-packages/comfyui_manager/legacy/manager_server.py", line 787, in queue_batch
await update_comfyui(None)
File "/comfy/mnt/venv/lib/python3.12/site-packages/comfyui_manager/legacy/manager_server.py", line 1600, in update_comfyui
rejection = manager_security.reject_simple_form_post(request)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/comfy/mnt/venv/lib/python3.12/site-packages/comfyui_manager/common/manager_security.py", line 65, in reject_simple_form_post
if request.content_type in SIMPLEFORM_CONTENT_TYPES:
^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'content_type'

Is there a way to fix this? I already removed the docker image and my user folders without any succes,

  • Author
9 hours ago, ionenl said:

Hi thanks for building this docker image.

Since v 0.20.1 i have the issue that comfyui wont update anymore, it failes when updating within comfy with the following error ;

[...]
File "/comfy/mnt/venv/lib/python3.12/site-packages/comfyui_manager/common/manager_security.py", line 65, in reject_simple_form_post
if request.content_type in SIMPLEFORM_CONTENT_TYPES:
^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'content_type'

Is there a way to fix this? I already removed the docker image and my user folders without any succes,

FYSA, I believe it is not just you, it is an issue with using the legacy manager UI apparently. I also have it occur to me.

The options are to disable the LEGACY webui temporarily or delete the run folder(what I did) so Comfy update.

I do not believe the patch as been merged yet https://github.com/Comfy-Org/ComfyUI-Manager/issues/2880

From it I expect the temporary solution is to set ENABLE_MANAGER_LEGACY_UI to false

Edited by martial

3 hours ago, martial said:

FYSA, I believe it is not just you, it is an issue with using the legacy manager UI apparently. I also have it occur to me.

The options are to disable the LEGACY webui temporarily or delete the run folder(what I did) so Comfy update.

I do not believe the patch as been merged yet https://github.com/Comfy-Org/ComfyUI-Manager/issues/2880

From it I expect the temporary solution is to set ENABLE_MANAGER_LEGACY_UI to false

Thanks, disabling the legacy webui and deleting the run folder worked for me!

Edited by ionenl

  • 2 weeks later...

I had this working a few weeks ago. Worked right out of the gate on the first install. I updated the nvidia plugin and it broke this docker. No ones fault. But i have tried 20+ times with different combinations or nvidia drivers and associated docker cuda versions. None or them work, i either get an error about torchaudio incompatibility, wrong cuda version or an error pops up saying a requirement is not installed(Just depends on the combination i put togeather). The torchaudio seems to be the most prevalent as i can install cuda 13.0 and everything else seems to load but torchaudio is at 12.8 and incompatible. Does anyone have a good way to get this installed? i would like to just go to an older version of the nvidia plugin but i am not smart enough to figure out how to go to an older version of the plugin.

Had something similiar in the past. Updated the Plugin and ComfyUi stopped working. It was the newest driver version that caused it. Go into the Plugin and try with an older driver version. Reboot required.

Screenshot_20260602_082520_Vivaldi.jpg

Also having issues with

Unraid 7.3.1
Nvidia Driver Plugin: 2026.05.07
Nvidia Driver Version: 595.80
Image Tag: comfyui-nvidia-docker:ubuntu24_cuda13.0-latest and ..._cuda13.2-latest

I get an error during container start:

== Pre-installing/Upgrading torch
== Installing torch using uv with backend: cu130
Installing: torch torchvision torchaudio
!! ERROR: Torch installation failed
!! Exiting script (ID: 99)
!! ERROR: subscript failed
!! Exiting script (ID: 1)

Edited by turtleinarock

18 hours ago, FireMcFly said:

Had something similiar in the past. Updated the Plugin and ComfyUi stopped working. It was the newest driver version that caused it. Go into the Plugin and try with an older driver version. Reboot required.

Screenshot_20260602_082520_Vivaldi.jpg

So i installed the same driver set you have, I also happen to have the 3090. My log shows failed and exit when launching. I was trying to load up the cuda 12.8 version of this docker. I get the below in my log as the last line. Not really sure why its registering cuda 13.0 when i ran the 12.8 docker. Either way it did not work. I am going to try some more configs with this driver. If i may ask what version did you install with that driver?

"RuntimeError: Detected that PyTorch and TorchAudio were compiled with different CUDA versions. PyTorch has CUDA version 13.0 whereas TorchAudio has CUDA version 12.8. Please install the TorchAudio version that matches your PyTorch version."

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

Account

Navigation

Search

Search

Configure browser push notifications

Chrome (Android)
  1. Tap the lock icon next to the address bar.
  2. Tap Permissions → Notifications.
  3. Adjust your preference.
Chrome (Desktop)
  1. Click the padlock icon in the address bar.
  2. Select Site settings.
  3. Find Notifications and adjust your preference.