Self hosting AI (if you can self host Opensim, you can do this!) (requires >4gb GPU)

Let's start with GenAI. At this point in time the main (but not only!) program people use is comfyui. The best way (to me) to approach comfyui (which is a node-based program and notoriously difficult to learn) is through another program that acts as a front end.

Krita is an open-source graphics program like GIMP (but better) and a guy named Acly made a plugin called "krita diffusion" that when installed, allows Krita to act as a front-end. After you install it Krita and the plugin, you just make a new image, type a prompt, possibly add some controlnets and away you go.

Stability Matrix was what I meant in my comment, and it's better for managing add-ons, such as getting base models and 'loras' (which add capabilities to your base model). That's a spot where Krita Diffusion is weak at. Stabiity Matrix offers the option of using 'packages' which basically means bundles of programs that are used to create or edit programs -you can also install a copy of comfyui there and approach it two ways (directly by launching a self-hosted web version) or through the Stability Matrix front ends. Additionally there is  two image generation front ends in the Stability Matrix interface -"image lab" and "inference" which can be used to make images and both use Stability Matrix's comfyui server (I don't know if they could use krita diffusion's server instead -but I suspect not, and it would be more limited anyway) I haven't found a satisfying way to make them work together, as a result I have both and something like 80gb for krita diffusion (most of that is optional, I include multiple models; sd1.5, sdxl,illustrius and flux 2 -it's possible to even simply choose sdxl at startup)

The topic of Models and Loras are too much to go into for this tutorial; what I'll say is that the base models give different capabilities (sd 1.5 is lightning fast, even at 4gb of ram, but often results in body horrors such as melted fingers and blurry faces,sdxl is better, has more options for style, but it's booru style prompting syntax is difficult to use).

It's also possible to install something like LMstudio's Bionic to get a text AI (like chatgpt, but much dumber) server which can help you with formulating prompts. if you, like me, have teensy tinsy vram you'll want to get your answer, pasted it somewhere, turn off the text ai and then switch on whatever you're using for graphics. Setting that up is a topic for a whole different post and honestly -it's no sin to use google.com/ai to help you set up your local server ;-) (it's what I did).

Links to get you started:
Krita image program (does NOT support AI, do not ask them about it):

https://krita.org/

Krita Diffusion krita plugin by Acly (does support AI, use their github or discord for support):
https://github.com/Acly/krita-ai-diffusion/

(side note, documentation can be found on their cloud site. You can buy credits very cheap there, but it's less flexible when it comes to using loras, etc and IMO is only really good for prototyping -meaning previewing how well your prompt works on a model like flux that would take half an hour or more to gen on your local hardware): https://www.interstice.cloud/

Stability Matrix (an all-in-one package to use different GenAI servers, including comfyui)

https://lykos.ai/

For text ai servers (mini-chatgpts, essentially):

https://lmstudio.ai/docs/bionic 

Comments will be moderated -heavily; and closed if they become a pain in my ass to moderate. That said, legit questions are very welcome but be aware I'm still trying to figure this out myself, so my knowlege is limited!