Easy Diffusion Setup Issues

Local AI Image Generation

A tree scene generated by Easy Diffusion on my laptop

I like the idea of being able to generate AI based images on my local hardware. This bypasses paywalls, account registrations, usage limits, and allows me to tweak the configurations to meet my needs. When I last tried to get Stable Diffusion running locally, I had encountered many problems due to the older hardware I am using for my desktop. I opted to wait for a while as I had other AI research that I could work on.

Marcel Gagné recently made reference to Easy Diffusion in one of his posts. The idea of running a single script and getting a functioning Stable Diffusion service running locally is very attractive and worth trying it out.

So after checking the requirements and installation process, I decided “fix” my hardware problem by switching to my laptop as my primary computer. The laptop has a GPU, whereas my desktop did not. I’ve been putting off this step but it is time. The desktop PC will change roles and become the brains for my laser cutting hobby.

Installing Easy Diffusion is pretty straight forward, and it worked as advertised. The instructions are good and clear. Having experience at the command line will help though. But in quick order you’ll have the web interface running, most likely.

The first image or two did work for me but were very slow. After a few runs, I investigated why it was slow. Simple enough — it was only using the CPU to process the images. And this led to the “big lessons”. Well, it boils down to one lesson really:

Make sure your video card drivers are installed and properly running BEFORE installing Easy Diffusion.

Because my laptop was a fresh install, Debian had defaulted to the Nouveau drivers which favored my AMD Ryzen’s built in video capabilities. Simple fix, I just needed to install the actual Nvidia drivers.

• You need the linux-headers package installed. This depends on your kernel version. Important note here — if you run an “apt update/upgrade” and the kernel gets updated — remember to reboot the computer to ensure the new kernel is actually in use. Otherwise the linux-header-$(uname -r) will refer to a different kernel version than the one in use.
• I had to configure Apt’s sources.list file…