- Venture The Unknown
- Posts
- DevLog: ComfyUI, the framework for AI generated images
DevLog: ComfyUI, the framework for AI generated images
Learn how to face swap yourself with someone with ComfyUI. Even if you don't have any GPUs!
Dev Log is a way for me to share what I’ve been working on, what I’ve tried & failed at. This is part of Dev Log series where I learn new stuff and document some of the pet projects.
Today’s log is about ComfyUI, a powerful AI generation workflow tool (generates images and videos). I just started and this log is mostly for setting things up.
The setup
Hardware:
Linux PopOS
NVIDIA GTX 4070 (may or may not be optional)
If you don’t have GPU, running on CPU is fine but beware, it takes significantly longer time to do anything related to AI generation.
Software:
GPU drivers (installation guide that I use)
Mainly I check available drivers using
sudo ubuntu-drivers list
For some reason I couldn’t use
ubuntu-drivers
command to install. But if you can, you should.I ended up using
nvidia-driver-535
by installing it through APTsudo apt install nvidia-driver
Python
I had python installed, if you don’t, Google how to install it.
The basic setup
Now, you should be able to follow instructions on ComfyUI Github. I did manual installation which I recommend if you’re a developer.
Once you have the base function up and running. You should install what’s called a custom node manager
Custom node manager handles:
Installation of missing models, clip, etc.
Managing versions of other custom nodes (including itself).
Installation of other custom nodes
Run basic SDXL setup
Here’s some interesting workflows to run (I haven’t personally run them, but I was just skimming through some interesting ones):
Basic SDXL flow (there’s a lot more here)
Learning materials
Run on the cloud
If you do not have any GPUs, you will want to use cloud services to run your app.
I am not affiliated with Modal but I’ve tried running ComfyUI on Modal the getting started sample works pretty well.
I’m trying to make it run with my own workflow. I still have some issues with getting it running.
I’ll be talking about this next couple of Tuesdays!