Don't have a Linux machine myself, so here's how I got SD WebUI running on GCE.

Just a quick log for anyone else who doesn’t have a local GPU. I’m worried Colab might eventually restrict SD, so I spun up a VM on Google Compute Engine to keep it running. Went with Ubuntu 22.04, which comes with Python 3.10.6 and git pre-installed—saves the hassle.

When setting up the machine, pick n1-standard-4, and for the GPU slot, just attach one NVIDIA T4—that’s enough for SD, don’t go straight for a big card and burn cash. Make sure the boot disk is set to 22.04 LTS, and check the firewall to allow HTTP, otherwise you won’t be able to access the WebUI from outside later. Once it’s created, just SSH into the terminal. Installing the drivers and WebUI is standard stuff, plenty of guides out there. My setup mainly solves the “no device but want to run locally” problem—pay-as-you-go on the cloud, shut it down when not in use to save money. If you’re planning to run it long-term, check out preemptible instance pricing too.

T4 is a bit rough for SDXL, but it handles 1.5 just fine.

Honestly, cloud GPUs aren’t that cheap when you break it down by the hour.

A lot of people forget to allow HTTP through the firewall.

Preemptible instances are a godsend, only catch is they get reclaimed every now and then.

If you’ve got a local card, local is still the way to go—low latency.

Cloud GPU by the hour really doesn’t pay off, run it a few months and you’d have paid for a used card, unless you’re only using it occasionally :sweat_smile: