How to run ComfyUI on RunPod?
Hi, I started with the whole Stable Diffusion recently and running ComfyUI on my M1 MacBook, which is quite slow for this use. I'd like to run the ComfyUI on RunPod, but I don't have any idea how to set it up and don't know if it's possible to download there custom nodes, my own checkpoints and loras from civit.ai etc. Could you please give me some newbie tips on how to install it there and what to do at first? Thank you a lot!
16 Replies
You don't need to install it, simply go to
Explore and select a ComfyUI template to deploy. You probably want one with Jupyter so that its easy to download your custom stuff.ok so when starting the server, I just need to pick the ComfyUI template, that's great! And what about downloading new checkpoints or loras? Is it possible there?
And how does it work with data storage? What are the differences between pod storage and network storage? Or what should be the settings?
And is it possible to change the GPU in Pod easily? When I choose some GPU, run Pod with it, install ComfyUI etc. and then will wan't to increase the power and choose a different GPU, can I do it, or do I need to start new Pod and install everything from scratch?
You can't change the GPU on a pod, you have to create a new pod to select a different GPU type
The normal storage volume is deleted when you delete your pod. Network storage allows you to attach it to different pods and persist the same data acrosss multiple pods as well as serverless. You don't have to reinstall everything when you create a new pod with network storage, but you have to install everything if you don't have network storage and create a new pod.
great, thank you for explanation 👍🏻 so do I need the normal storage or should I use the network storage instead?
Depends on your use case
stable diffusion image generation in ComfyUI.. what are the cases I would need the normal storage?
and is the storage changeable in the future on existing Pod, or is it the same as with GPU, so you need to set up a new pod with bigger storage?
You can increase your storage but can't decrease it
great
and what are the use cases I would need the normal storage instead of network?
or what is the "safe option" in storage settings?
If you ever want to shut down your pod (and therefore not pay to keep it running), but then come back to it later without having to redownload all of your specific checkpoints, loras, etc, then you should use network storage. If you need more storage than the pods give you by default, you should use network storage. If you're okay doing simple image generation with one or two checkpoints, minimal things you need to download each time you re-initialize your pod, then you can just start a pod and use its standard storage
This is incorrect, you can override the template defaults and change the amount of storage they assign by default.
so maybe the question is - do I need the normal storage at all, or should I use just the network storage?
Its up to you which one you prefer
I for sure need the network for at least checkpoints, loras etc. Is there some advantage in the normal one?
Syncing times and disk access times are faster without network storage
ok so probably the best version is to have at least 10gb normal storage and then some network and always copy the comfyui installation and all the checkpoints and loras from the network storage to the normal one for the "work session"?
I started first Pod and playing around.. could you just give me a hint how to download the checkpoints to the network storage? Or how do I recognize in what storage I'm in the terminal?
everything you download inside of the workspace folder is in the network storage (I think). The best way to download checkpoints inside of your pod is with wget--figure out the proper way to modify the download link to the checkpoint so you can use wget on it