Got it. I'll write it up in case it helps another. This will initially only cover CPU "sampling" (generating an image) until I get GPU sampling working. Sampling should run entirely offline.

install with pip

1. pip install --upgrade diffusers transformers scipy torch


2. sudo apt install git-lfs


3. clone the git repository at https://huggingface.co/runwayml/stable-diffusion-v1-5 (you have to log in or sign up first and accept their license agreement)

Then you can create a small Python script (inside your local working copy of the cloned git repo above) and run it to try sampling for yourself:

from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained('.')

prompt = "a photo of an astronaut riding a horse on mars"

image = pipe(prompt).images[0]  

image.save("astronaut_rides_horse.png")

easier and better method

https://github.com/invoke-ai/InvokeAI#installation

This provides a really nice web GUI, too.

onboard GPU note

My GPU shows up as Intel CometLake-S GT2 [UHD Graphics 630] from lspci | grep VGA or neofetch. screenfetch calls it Mesa Intel(R) UHD Graphics 630 (CML GT2). Either way I don't know how to use this GPU for sampling (or if it is even possible).