This is some starter info from the last thread but with a small edition.
>>1632
Getting an AI UI running on your PC has really been streamlined these days.
>Requirements:
Have a GPU with at least 6GBs of VRAM (not to be confused with normal RAM). If you use Stable Diffusion Forge or ReForge, it's made to run off of weaker hardware so you can still use SDXL models. It's good to aim for 12+ GBs of VRAM. I use a 4070 Super since it's made for AI. It has 12GBs of VRAM and I can gen 1024x1400 images in 13 seconds. AMD cards work but they're more of a pain to setup.
https://github.com/LykosAI/StabilityMatrix
This setups everything for you in a very simple manner and is constantly updated with better features. It also has access to basically every AI UI you want. I use Reforge.
>Extensions
There are like 5 extensions I think are mandatory for generation:
<Adetailer
Automatically inpaints both hands and face while generating. Hand inpainting can by tricky but it works very well on the face. Using this means you can generate an image but inpaint with Adetailer to generate a face with a completely different art style. You can even use another checkpoint just for the Adetailer portion of the gen (this may crash your UI just to warn you). I use this to get the body shape and style from one artist while using another to get a certain kind of eyes. If you just want to have higher quality faces, then just leave the adetailer prompt blank. I recommend setting the steps higher and the pixels from 512x512 to 1024x1024 for a higher quality inpaint.
<Dynamic Prompts
Allows you to use wildcards. Wildcards are a .txt file with a list of words. When added to a prompt like:
Kemono_girls
It will randomly select a prompt from that .txt file. Everything in one line is a prompt. Different prompts are separated by a line. So you can have bat girl, bat ears, bat wings, in one line and it will use that whole line as a prompt. Having a dozen of these wildcards can give a lot of diversity to your prompts. It really helps to go on Civit AI to search for them. Here are a few of my old wildcards.
https://pixeldrain.com/u/T4w9gD26
<a1111-sd-webui-tagcomplete
Another must have extension. It auto completes words based on Danbooru tags. Makes it so much easier to make prompts for characters and tags with this helping you out. There's also an e621 version. Considering that models are based on danbooru tags this will make your prompt much better.
<sd-webui-lobe-theme
If you're using Forge, Reforge, or A1111 UIs, this upgrades the UI into something much more productive and useful.
<sd-webui-regional-prompter
Allows for region prompting which means you can set a prompt to show up in a certain area. Gives you way more control on what's being generated. Also makes it much easier to have multiple distinct characters in the same gen without their features being mixed.
<Vectorscope / Diffusion Color Grading (Diffusion-CG)
Vectorscope controls the brightness and contrast of an image while generating. It's great for making darker images without relying on Vpred models. Diffusion-CG helps with colors while generating. If you think your images are too grey or saturated despite changing your VAE. Try this.