Didn't find much info about this, so i thought I'd write a quick blog post.
There's quite some info on the topic of how to actually make an UI that changes graphics settings at runtime, so this won't focus on that. Here's a video from GorkaGames who has good blueprint tutorials on Youtube, which should help you get that covered.
So to the main part, when you have all the settings set up but you want to change what happens when you select low, medium or high settings.
So in the Config folder of your code projects engine folder you have a file called BaseScalability.ini, which looks something like this at the beginning of the file.
What you first want to do is to figure out at how low settings your game still looks like it should. Here's a screenshot from our settings menu after testing in build what worked for us:
To my understanding how unreal handles these settings is in a waterfall kind of sense, it always starts at the engine's BaseXYZ.ini file, then goes to the projects Config folder to see if there's a override called DefaultXYZ.ini, then I think it goes to the game's Config folders platform specific overrides which would probably be under Project/Config/Windows etc. Here's the Lyra Scalability documentation article I used to find out this info (It has a lot of stuff there, which is unrelated to this issue, but useful nonetheless).
The next step after this is to create this file, in this location and copypaste all the content from the BaseScalability.ini file into it.
What i did then was just to make sure the lower options are just duplicates of the lowest option that still kept the game visually similar figured out in the previous step.
So for example my Reflections setting needed to be at a minimum on the Medium level of the what is set up in the BaseScalability config, like here:
I didn't test if you could remove settings completely in the DefaultScalability config file and what would happen if you did. Would it just bring the lowest setting from the Base file or, would it not show up at all.
This solution solved the problem i was facing to the extent that was necessary right now, so I'll do the experiment potentially later in production.
Thanks for reading and hopefully this saved you some time if you ran into a similar issue.