[Announcement] The frontend will no longer be shipped in the main ComfyUI repo, it will be a separate pip package instead.

[comfyanonymous]

Due to the frontend being a separate project it doesn't make sense to continue shipping the frontend as part of the main repo.

To make our lives easier and prevent the main ComfyUI repo from becoming too large in size over time we have decided to start shipping the frontend as a separate pip package instead.

#7021

After this change is merged if you use the desktop build or the standalone package and update using the update/update_comfyui.bat script or the ComfyUI manager everything should continue working as usual.

If you update manually using "git pull" you will have to do the extra step of updating the frontend using: pip install -r requirements.txt

If you have any comments or concerns about this change, let us know.

[scoegrams]

Well this took me hours to setup the other day, now I cant get it to work. so yeah im pretty smashed by this. I could have used it how it was for years.

[simonlui]

Was there any reasons why you can't make the UI repo a Git submodule for this repo instead? I'm not against the core idea of separating out the frontend code but doing it in the way described with a PyPi pip package instead is a bit weird.

[comfyanonymous]

Git submodules would have the same problem since it's moving the problem to another git repo.

The reason for it being a pip package is that we already use python packages to manage core ComfyUI dependencies so it seemed like a much easier thing for people to migrate to than all the other solutions including git submodules.

We basically needed a way to ship potentially large frontend updates that now contain some binary files (images with workflows attached). Having them continue living in the git repo would be unsustainable.

We don't have to change anything in our packaging or updating scripts to make this work so we assume it will be a seamless transition for almost everyone.

[Maddoc1971]

"After this change is merged if you use the desktop build or the standalone package and update using the update/update_comfyui.bat script or the ComfyUI manager everything should continue working as usual."

Sorry, but that doesn't work....

[comfyanonymous]

This change hasn't been merged yet, it will be most likely merged in ~12 hours so if you have issues before that this isn't the problem.

[Maddoc1971]

Ah ok, thx...I just did it manually xD

[imaginaryhuman]

Ok folks. I use SwarmUI. Now SwarmUI is broken and will not work due to this apparent "front end missing" thing.

It is not enough to just say some kind of "its a separate package, do a pip install thing". I have no clue what you're talking about.

What EXACT steps does a person need to take in order to get this up and running again?

I'm on windows 11. Am I supposed to "go to the separate project" and somehow install that?

No idea what you mean by "After this change is merged".

Help!

[edit]... I seemed to be able to install the comfyui frontend using a github clone ... although had trouble with the pip install stuff... but somehow... then did an 'update-windows.bat' and after that it sort of seemed to work? I have no clue what i'm doing.

[steelan9199]

https://mp.weixin.qq.com/s/kEV64GvhpyzmbxW5MOWNug

[olegger58]

doesn't work.

[Conscienciagit]

Como eu faço para corrigir este erro (Your ComfyUI isn't git repo.) Vi que houve uma mudança no método de distribuição front-end para instalação de dependência via PyPI (PIP). Como faço isso?

[1blackbar]

I download portable with preinstalled nodes and several GB, we have 2025 with very fast internet speeeds, nobody cares about repo sizes.
But anyway i consider coming back to 2 month old prev comfy cause recent changes in memory management give me OOM on ... SDXL, yes on sdxl and rtx 3090... funny right....

[ltdrdata]

doesn't work.

Go to ComfyUI dir and git pull
And then launch ComfyUI again, it will show you more proper fixing command.

[ltdrdata]

I download portable with preinstalled nodes and several GB, we have 2025 with very fast internet speeeds, nobody cares about repo sizes.
But anyway i consider coming back to 2 month old prev comfy cause recent changes in memory management give me OOM on ... SDXL, yes on sdxl and rtx 3090... funny right....

We apologize for the abrupt changes.

However, if the front end continues to be updated within the repository as before, it's only a matter of time before it grows from tens of GB to TB.

Since this structure is not sustainable in the long run, we had to act quickly to address the issue.

[diGitTil]

#7021

After this change is merged if you use the desktop build or the standalone package and update using the update/update_comfyui.bat script or the ComfyUI manager everything should continue working as usual.

If you update manually using "git pull" you will have to do the extra step of updating the frontend using: pip install -r requirements.txt

If you have any comments or concerns about this change, let us know.

I sure do have comments!

I appreciate you devs are really deep into the matrix on this stuff, but us mere end users really struggle with troubleshooting if errors assume so much knowledge. If I might be so bold and suggest writing more instructional error messages that can help guide a less knowledgable person to self-service, I think we'd be super grateful.

What am I supposed to do here, what is thie "Request" and where do I lok into it? I'm so lost...

[UnrealGem]

This should be enough to push people over the edge. It's so aggravating for us less knowledgeable users to figure out all the errors but this update really broke it for me and now and I cant even update or reinstall ComfyUI without errors. Am I ready to tackle these new problems with more hours, possibly days trying to get it to start again? Not sure I have the energy or ambition to. I see so many complaints and others pushing for a new alternative and this might just be enough to get some new alternatives. I don't want to sound unappreciative, I truly am. All I am saying is that ComfyUI has such a steep learning curve on just the errors alone and now we have to deal with new ones, for the moment anyhow. Well good luck to all who are brave enough on to tackle these new problems.

[ltdrdata]

This should be enough to push people over the edge. It's so aggravating for us less knowledgeable users to figure out all the errors but this update really broke it for me and now and I cant even update or reinstall ComfyUI without errors. Am I ready to tackle these new problems with more hours, possibly days trying to get it to start again? Not sure I have the energy or ambition to. I see so many complaints and others pushing for a new alternative and this might just be enough to get some new alternatives. I don't want to sound unappreciative, I truly am. All I am saying is that ComfyUI has such a steep learning curve on just the errors alone and now we have to deal with new ones, for the moment anyhow. Well good luck to all who are brave enough on to tackle these new problems.

Resolving this issue is as simple as reinstalling the requirements.txt file.

Additionally, if an error occurs, the ComfyUI will display what command can be used to fix it if there is a problem.

Which part are you stuck on?

[murphylanga]

Unfortunately, you can no longer go back to the frontend v1.9.1. This is the last frontend where the preview with the checkpoint and Loread Loader worked!!!!!!!

This no longer works in the newer versions.

For those who can't get the "new" Frontend installation version to work in the portable version, try changing to the ComfyUI directory and in the cmd window enter:

D:\Special\ComfyUI_windows_portable\python_embeded\python.exe -m pip install -r requirements.txt
to integrate the frontend.

[Maddoc1971]

this
"D:\Special\ComfyUI_windows_portable\python_embeded\python.exe -m pip install -r requirements.txt"

is exactly what I did to integrate the frontend. For me it worked.

[GeorgeLee007]

just started ComfyUI in Google Colab and hit the issue.

Yet, the reminder message is clear enough. Thx

[Grimmvg]

waw this broke everything , spending hours already attempting to fix it , the earlier mentioned fixes don't work here. Damm i regret having pressed the update button. For artists / non coders, this is really a pain in the ass and the instructions are totally unclear. A gonna be a fun night thx

[Aenon1]

Making the process for the new ComfyUI UI automatic.

Opening a venv and running: D:\Ai\ComfyUI\ComfyUI\venv\Scripts\python.exe -m pip install -r requirements.txt does not have to be done manually.
Add:
python.exe -m pip install -r requirements.txt
to the startup bat file after the call to open the venv and boom, there it is.
Simple. If the call for the updates are not changed, this will work for them also.
Yes, it will run every time you boot Comfy and show that everything is already installed, but it's better than ComfyUI breaking every time that an update on the UI has been performed and will help those who have no idea about using command windows and the venv to run pip commands.

The directory structure will have to be changed to what you use, but here are the contents of my comfyUI.bat file:

cd/d "D:\Ai\ComfyUI\ComfyUI"

echo Checking for repository updates...
git pull

echo Activating virtual environment...
call venv\Scripts\activate.bat

python.exe -m pip install -r requirements.txt

echo Checking for pip updates...
python -m pip install --upgrade pip

echo Running the application...
python main.py --front-end-version Comfy-Org/ComfyUI_frontend@latest --auto-launch

pause
main.py

[mu-sr]

I am leaving the required solution, for people like me, who know only about cmd but have no understanding of Python's specifications or how ComfyUI works.

1. Open cmd and update Python to the latest version.
   python -m pip install --upgrade pip
2. Set the directory where you installed comfyui and the requirements.txt file, created by this update, is located.
   e.g.
   cd "ｄ:\ComfyUI\ComfyUI"
3. Enter the spell which many people are struggling with.
   pip install -r requirements.txt
   If successful, several dozen lines of text will be displayed.
4. Run the update_comfyui.bat file, which is likely located near the folder mentioned above. If several dozen lines of text are displayed here as well, run ComfyUI and it should start up just like before.

I sincerely hope that this hint can help in rescuing those who are trapped in confusion and despair. Sorry for poor English.

[Sweet77i]

May I ask how to solve this problem

[ltdrdata]

May I ask how to solve this problem

Change the requiremets.txt to the ..\ComfyUI\requirements.txt

[mu-sr]

or I suppose to check if there's requiremets.txt in your python embeded folder?
if not, you'll find it in comfyui folder.

[ltdrdata]

FYI, latest ComfyUI-Manager is auto fixing this issue.

Goto ComfyUI/custom_nodes/comfyui-manager in cmd
(Or ComfyUI/custom_nodes/ComfyUI-Manager)

And execute git pull.
And then relaunch the ComfyUI.

[Cloud71]

FYI, latest ComfyUI-Manager is auto fixing this issue.

Goto ComfyUI/custom_nodes/comfyui-manager in cmd (Or ComfyUI/custom_nodes/ComfyUI-Manager)

And execute git pull. And then relaunch the ComfyUI.

Thank you, it worked to me!

[obulks]

After the update, the ComfyUI/web/user.css file stopped working. To make it take effect, I need to place user.css in the path: python_embeded/Lib/site-packages/comfyui_frontend_package/static/user.css

[Kozatiju]

Manual update sadly broke everything for me aswell.

While updating I have these issues appearing at the end of the installing / updating process :

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
broken-source 0.8.0.dev0 requires numpy==1.26.4, but you have numpy 2.2.3 which is incompatible.
broken-source 0.8.0.dev0 requires pillow~=10.4.0, but you have pillow 11.1.0 which is incompatible.
broken-source 0.8.0.dev0 requires tqdm~=4.66.5, but you have tqdm 4.67.1 which is incompatible.
colour-science 0.4.4 requires numpy<2,>=1.22, but you have numpy 2.2.3 which is incompatible.
gradio 4.44.0 requires pillow<11.0,>=8.0, but you have pillow 11.1.0 which is incompatible.
ltx-video 0.1.2 requires huggingface-hub~=0.25.2, but you have huggingface-hub 0.29.2 which is incompatible.
mediapipe 0.10.11 requires protobuf<4,>=3.11, but you have protobuf 5.27.2 which is incompatible.
moderngl-window 2.4.6 requires numpy<2,>=1.16, but you have numpy 2.2.3 which is incompatible.
numba 0.59.1 requires numpy<1.27,>=1.22, but you have numpy 2.2.3 which is incompatible.
onnxruntime 1.18.1 requires numpy<2.0,>=1.21.6, but you have numpy 2.2.3 which is incompatible.
onnxruntime-gpu 1.18.1 requires numpy<2.0,>=1.21.6, but you have numpy 2.2.3 which is incompatible.
tb-nightly 2.17.0a20240424 requires protobuf!=4.24.0,<5.0.0,>=3.19.6, but you have protobuf 5.27.2 which is incompatible.
Successfully installed numpy-2.2.3

And when lauching Comfyui :

WARNING: The script f2py.exe is installed in 'C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Scripts' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
broken-source 0.8.0.dev0 requires pillow~=10.4.0, but you have pillow 11.1.0 which is incompatible.
broken-source 0.8.0.dev0 requires tqdm~=4.66.5, but you have tqdm 4.67.1 which is incompatible.
gradio 4.44.0 requires pillow<11.0,>=8.0, but you have pillow 11.1.0 which is incompatible.
ltx-video 0.1.2 requires huggingface-hub~=0.25.2, but you have huggingface-hub 0.29.2 which is incompatible.
mediapipe 0.10.11 requires protobuf<4,>=3.11, but you have protobuf 5.27.2 which is incompatible.
tb-nightly 2.17.0a20240424 requires protobuf!=4.24.0,<5.0.0,>=3.19.6, but you have protobuf 5.27.2 which is incompatible.
[ComfyUI-Manager] 'numpy' dependency were fixed
[ComfyUI-Manager] 'comfyui-frontend-package' dependency were fixed

At the best to finish :

Checkpoint files will always be loaded safely.
Traceback (most recent call last):
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\main.py", line 136, in
import execution
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\execution.py", line 13, in
import nodes
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\nodes.py", line 22, in
import comfy.diffusers_load
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\diffusers_load.py", line 3, in
import comfy.sd
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\sd.py", line 7, in
from comfy import model_management
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\model_management.py", line 189, in
total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
^^^^^^^^^^^^^^^^^^
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\model_management.py", line 146, in get_torch_device
return torch.device(torch.cuda.current_device())
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Lib\site-packages\torch\cuda_init_.py", line 971, in current_device
lazy_init()
File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Lib\site-packages\torch\cuda_init.py", line 310, in _lazy_init
raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

Any idea why the update has broken all these nodes ? It was working fine before. Any help would be much appreciated :-]

[jquintanilla4]

After this update, some of my older workflows don't work anymore. The frontend freezes whenever i try to load them up. The backend still works fine, so i have to close all tabs and restart the frontend with the default worflows.

[ltdrdata]

Manual update sadly broke everything for me aswell.

While updating I have these issues appearing at the end of the installing / updating process :

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. broken-source 0.8.0.dev0 requires numpy==1.26.4, but you have numpy 2.2.3 which is incompatible. broken-source 0.8.0.dev0 requires pillow~=10.4.0, but you have pillow 11.1.0 which is incompatible. broken-source 0.8.0.dev0 requires tqdm~=4.66.5, but you have tqdm 4.67.1 which is incompatible. colour-science 0.4.4 requires numpy<2,>=1.22, but you have numpy 2.2.3 which is incompatible. gradio 4.44.0 requires pillow<11.0,>=8.0, but you have pillow 11.1.0 which is incompatible. ltx-video 0.1.2 requires huggingface-hub~=0.25.2, but you have huggingface-hub 0.29.2 which is incompatible. mediapipe 0.10.11 requires protobuf<4,>=3.11, but you have protobuf 5.27.2 which is incompatible. moderngl-window 2.4.6 requires numpy<2,>=1.16, but you have numpy 2.2.3 which is incompatible. numba 0.59.1 requires numpy<1.27,>=1.22, but you have numpy 2.2.3 which is incompatible. onnxruntime 1.18.1 requires numpy<2.0,>=1.21.6, but you have numpy 2.2.3 which is incompatible. onnxruntime-gpu 1.18.1 requires numpy<2.0,>=1.21.6, but you have numpy 2.2.3 which is incompatible. tb-nightly 2.17.0a20240424 requires protobuf!=4.24.0,<5.0.0,>=3.19.6, but you have protobuf 5.27.2 which is incompatible. Successfully installed numpy-2.2.3

And when lauching Comfyui :

WARNING: The script f2py.exe is installed in 'C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Scripts' which is not on PATH. Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location. ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. broken-source 0.8.0.dev0 requires pillow~=10.4.0, but you have pillow 11.1.0 which is incompatible. broken-source 0.8.0.dev0 requires tqdm~=4.66.5, but you have tqdm 4.67.1 which is incompatible. gradio 4.44.0 requires pillow<11.0,>=8.0, but you have pillow 11.1.0 which is incompatible. ltx-video 0.1.2 requires huggingface-hub~=0.25.2, but you have huggingface-hub 0.29.2 which is incompatible. mediapipe 0.10.11 requires protobuf<4,>=3.11, but you have protobuf 5.27.2 which is incompatible. tb-nightly 2.17.0a20240424 requires protobuf!=4.24.0,<5.0.0,>=3.19.6, but you have protobuf 5.27.2 which is incompatible. [ComfyUI-Manager] 'numpy' dependency were fixed [ComfyUI-Manager] 'comfyui-frontend-package' dependency were fixed

At the best to finish :

Checkpoint files will always be loaded safely. Traceback (most recent call last): File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\main.py", line 136, in import execution File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\execution.py", line 13, in import nodes File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\nodes.py", line 22, in import comfy.diffusers_load File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\diffusers_load.py", line 3, in import comfy.sd File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\sd.py", line 7, in from comfy import model_management File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\model_management.py", line 189, in total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) ^^^^^^^^^^^^^^^^^^ File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\ComfyUI\comfy\model_management.py", line 146, in get_torch_device return torch.device(torch.cuda.current_device()) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Lib\site-packages\torch\cuda__init__.py", line 971, in current_device lazy_init() File "C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\Lib\site-packages\torch\cuda__init_.py", line 310, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled

Any idea why the update has broken all these nodes ? It was working fine before. Any help would be much appreciated :-]

You need to downgrade numpy to numpy<2 and reinstall torch.

Execute following commands.

C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip install "numpy<2"
C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip uninstall torch torchvision torchaudio
C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

[OtakuOW]

Help me :(

[OtakuOW]

[Kozatiju]

You need to downgrade numpy to numpy<2 and reinstall torch.

Execute following commands.

C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip install "numpy<2"
C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip uninstall torch torchvision torchaudio
C:\Users\Roger\Documents\ComfyUI\new_ComfyUI_windows_portable_nvidia_cu121_or_cpu\python_embeded\python.exe -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

Wow, thanks a lot ! Getting help from you is an honor :-)

A big thank you for the rest of your work, it is so HUGE and helpful for the community...

[Piscabo]

Why does my pip package get downgraded every time I do an update?

Installing collected packages: comfyui-frontend-package
Attempting uninstall: comfyui-frontend-package
Found existing installation: comfyui_frontend_package 1.12.8
Uninstalling comfyui_frontend_package-1.12.8:
Successfully uninstalled comfyui_frontend_package-1.12.8
Successfully installed comfyui-frontend-package-1.11.8

[OtakuOW]

new error: mat1 and mat2 shapes cannot be multiplied (308x768 and 4096x5120)