How to setup a "portable Linux" version?

Hi

I am currently writing a plugin for ComfyUI for the Render Manager Royal Render.
I am working on Windows only so far.
For Windows I can easily download and test ComfyUI Portable (all included) and ComfyUI Desktop (separated Python + Data path)
As far as I see it, I can copy them from the fileserver to the local drive to use the on a new machine.

But how about Linux?
How would you setup a portable version that I can just copy?
(Before you ask: of course all machines in the network would have the same Linux version)

In short, use a docker. That’s the portable for Linux.

Unfortunately ComfyUI does not offer a docker solution yet. And unfortunately it is not this easy as with Windows, where you simply download the portable and are good to go. There are a few more steps involved. But far fewer than with a native installation, where you might even need to compile Torch yourself. I tried, and have then given up at it. I have the wrong Ubuntu version installed anyways.

Docker is also not completely trouble free. But no ComfyUI installation is. And you must install ROCm or Nvidia drivers on the host system before the container can use the GPU.

Docker does not provide the kernel driver.

I had a docker working on Ubuntu for a while. I use for example currently this one, the rocm 7 version:

Just ask ChatGPT for advice how to install and run. It’s not this complicated.

It has unfortunately quit working before a few days. The vae node now thinks there is no AMD gpu anymore, searches for Cuda and crashes. And since then i try to get it back working. Still at investigating.

Thanks,
I did not though about using docker.
I will try to provide some steps or scripts to easily create an image.

1 Like