Sign In

Setup & Basic Usage of XenoEngine SD Web UI - Free noGPU Full Featured Huggingface Space

Jun 20, 2023
tool guide
Setup & Basic Usage of XenoEngine SD Web UI - Free noGPU Full Featured Huggingface Space

In this article, we'll walk you through the first time setup of your Huggingface Space and then some tips for first usage. If you don't know what this is all about, check the Introduction Article and Table of Contents for this series.

Articles Index

Copy And Install XenoEngine SD WebUI

  1. Visit the XenoEngine SD WebUI Space - but do not try to run anything yet! (It can work in a shared space like this - but with 10-15 minutes render time per image, adding yourself to the queue could have you waiting for a LONG time if there are other people using it.)

  2. Log in or Create a Huggingface account.

  3. In the top right corner next to the "Settings" tab, find the 3 Dots Hamburger Menu icon - click it and select "Duplicate This Space".

  4. Make sure that your username is set as the "Owner".

  5. Give the space a name - anything you like - this will become the URL/Location of the space.

  6. Set it to "Private" (so you don't have other people jumping into your queue).

  7. Leave the Hardware set to the "FREE" option. (Though not fully tested, this should work fine with upgrades - and it looks like you can quickly and fairly easily swap them out as needed for now, leave it off and let it boot up so you can see what you're looking at).

  8. Wait. For a While. Every time you restart this space, it will create an install of Stable Diffusion, all the extensions, and boot it all up for you.

  9. In the Build Logs (Click the logs button at the top if you don't see the overlay) look for the following. (NOTE: Civitai Helper will be booting up at this point, so the log will keep adding things - once the below lines have passed, you can generally start looking around and working, though. You'll know it's totally done when it gives you a "Civitai Updated X Images" line.

    Running on local URL:  http://0.0.0.0:7860
    To create a public link, set share=True in launch() 
  10. Start Creating!

First Time Usage Notes

  • It can take a while to start. A clean "Factory Reboot" can be as long as 20 minutes depending upon how smoothly all the pulls and pushes are happening. A normal restart can take less time (but still take a while) - on a restart, it just checks to see if a file/extension is there and it moves on if it already exists. (Models are NOT safe during a restart - they'll vanish unless put away or reacquired.

  • The code doesn't have good timeout fallbacks at this point (something I'm currently reading about to try to improve) so if, for example, Civitai is having issues like it does so often, bootup may fail - just restart (though waiting a bit and confirming that the offending site is back to working is a good idea).

Basic Usage Tips And Tricks

I'll get into more tips and tricks over the coming days and weeks. For now, play around and kick the tires - and see what things you can make. Here are some things to help you out the first time in there - whether you're brand new to Stable Diffusion or you're an old pro but need some tricks on working within the limits of Huggingface and No GPU Rendering.

Extensions To Get To Know

I've not, yet, found a good way to efficiently mirror output files, config files, and models across persistent storage. In other words, this can't connect to Google drive or some other place to automatically save things. I've installed various extensions to help out, though.

  • Civitai Helper Tab (Docs/Info): This is a handy way to grab a model you've found on Civitai and then quickly and easily install it. The stuff you install this way will vanish on the next run, though - so use this primarily for things you just want for the current session.

  • Batchlinks Downloader (Docs/Info): Grabs models from any other source and supports all model types. This is also handy because you can simply keep a text file of the models you want to use each time - or you can create combinations of models for different tasks. For example, you may want a set of Models and LoRas for "Realistic" work - while that set differs a lot when you're making "Anime" stuff. Paste in the file for your "preset list" and let 'er rip.

    NOTE: If there are models you want to use EVERY time you start (for example, your favorite checkpoint) you can edit the "on_start.sh" file - there are comments in there to help you. (I'll write up something better about this in future articles).

  • Hugging Face Tab (Docs/Info): This is an interface to let you quickly send files to store on Huggingface - useful for when you are merging models and so on. Create your model, then push it to HF with relative ease.

  • Filer Tab (Docs/Info): This makes saving things easier. If you're not downloading the good images when you render them - you can jump over here and grab the batch. (TIP: You should do this often - saving as you go, or grabbing batches when you get something really good - if the space crashes or otherwise has issues, a restart kills off all unsaved images). This is also what you'll use to copy changes to config files and so on - more on that in later articles.

I've left certain extensions there, but they may not work or may have issues with certain features. For example, the "Training" tab is probably useless - though I haven't tried it. It takes an hour or so on a fast machine to get a hard round of training in - I can't imagine how long (if it would be possible at all) to train without a GPU.

Another example are Jinga Templates in Dynamic Prompts. While most of that seems to work okay (though wildcards are a bit funky - I'm looking into it) Jinga Templates seem to rely on the GPU to work - and since we don't have one, it's going to toss an error.

Explore... the worst that happens is that it crashes and you need to restart the system.

4

Comments