Guide: Setting up SD as a shared service on our office server.

Hey everyone! 👋 We just rolled out Stable Diffusion as a shared service on our office server last week, and it’s been a total game-changer for our marketing and design teams. No more waiting for local renders or dealing with mismatched model versions. I thought I'd share our setup guide since a few folks in the Automation channel were asking.

Here’s the basic stack we used:
- **Hardware:** A dedicated Ubuntu server with an RTX 4090 (24GB VRAM is key for batch runs).
- **Core Software:** Automatic1111 WebUI (forked) for the base installation.
- **Access Method:** Set up with Tailscale for secure VPN access, so our remote team can use it too.
- **Shared Assets:** A central network drive for models, Loras, and output folders.

The biggest wins for us were:
* **Standardized models:** Everyone uses the same checkpoint for brand consistency.
* **Queue system:** The built-in queue prevents GPU overload during office hours.
* **Template workflows:** We saved our common prompts and settings as presets for things like blog images and ad creatives.

A few pitfalls we hit (and fixed!):
- **Permissions:** Had to carefully set user permissions on the shared output folders to avoid... creative deletions.
- **Updates:** We now have a weekly update schedule so new extensions don’t break mid-project.
- **Default settings:** We locked down a default sampling method and resolution in the config to keep things fast.

It’s essentially like having our own internal AI image service now. The design team loves it for rapid mockups, and I’m using it for generating landing page variants to A/B test. If you're thinking of setting one up, I highly recommend it for team collaboration!

Happy to answer any questions about the config or our workflow!

~ jenny