I recently started generating AI images, but I’m having trouble with some issues. I’d appreciate some help from someone knowledgeable and kind.
1. I can’t install ComfyUi custom nodes. Even when I use comfyui manager, I can’t install the missing custom nodes.
2. I can’t use Web Ui Forge. I can generate images right after installing Forge, but it starts to malfunction as I use it, and eventually I can’t generate images anymore. I’ve reinstalled it many times, but nothing has changed.
I think that PyTorch is the one that is commonly used by both and tends to break easily. In particular, it seems that some of the custom nodes in ComfyUI may change the version of PyTorch when installed. Reinstall PyTorch as needed. I think that you can avoid problems with Forge by using the all-in-one package. This is because the Forge version of PyTorch is included in the Forge folder.
Also, if you are not using a CUDA environment, there have been reports of the xformers software malfunctioning. In any case, if you read the error message and search the internet using the last part of the error message as it is, you will often find a solution. If you know the error, it will give you a clue to the solution.
I tried to reinstall PyTorch as you told me to, but in order to install PyTorch, I have to install CUDA, and before installing CUDA I have to install something called cuDNN, which means there are too many steps. Is this really necessary?
In the case of the portable version of ComfyUI or the all-in-one version of Forge, the CUDA environment in the folder is actually used, so it is already installed, and in that case, installation is not required.
However, if you want to install and use it normally, rather than in a virtual environment, you will need both CUDA and cuDNN. Also, it seems that some of the custom nodes assume the existence of a CUDA toolkit outside of the virtual environment, so in your case, you may need it for the custom node.
Installing CUDA and PyTorch is very difficult even for people who are used to doing it, so be sure to find and refer to an appropriate know-how page first. It’s something you only have to do once, but it’s really very troublesome…
I think you’ll have to search to find out which version of CUDA is best, which version of Python or PyTorch is best for avoiding errors in Windows, etc., and I think you’ll definitely not understand…
even if you do. Let’s just think that it’s OK as long as it works in the end.