/vt/ - Virtual Youtubers

>>41572019
yeah, that's one way to go about it I guess. base NAI negatives aren't super fantastic even on good_shit, not just other mixes, nor is "masterpiece" a very good idea. it deepfries even harder than what you see now on the colab when I use my own model mixes...

anyways, I updated some stuff. there's a toggle box for using "localtunnel" - localtunnel is fucking broken right now, so don't use it, but when it WAS working, it re-enabled the progress bar and previews for me ;_;

also, the console output in the notebook will show "Connected" with a little green checkmark and green text for progress updates while images generate copied from TheLastBen's notebook

and it uses rsync to copy the embeddings and model when you first start it up - for the embeddings - that shows you all of their names - for the model - it shows a progress bar, which Linux's copy command "cp" does not...

I also made xformers a toggleable option, since voldy finally updated webui to use a newer version of torch, you can just use the regular xformers installation and not the special wheel we've been installing in the past.

I tested it and nothing is broken, so feel free to re-copy it if you use it. I also updated the config.json/ui-config.json that it loads by default, and I might further update styles.csv so you can pick from a few different variations on "good" negative prompts, but that's already in the code, so as I change those files in my drive, the next time you launch, you'll have different default settings, like clip skip 2, some samplers hidden by default LMS, DPM fast, DPM adaptive, LMS Karras, PLMS and the danbooru autocomplete tag thing should show 10 results. and it will load the NAI vae automatically since that's the new "default"