I'll give it a fair go, been testing these tags on Lacike's Mega Cock checkpoint, along with facial feature modifiers lately https://civitai.red/posts/30765232

This is actually something I covered in my guide on style tags since this set of styles is really nonindicitive till you go to the Danbooru wiki. "Realistic" is actually the tag for what is often called "semi-real" in the text to image context: By definition, it does not look like a photo, and is "realistic" in the sense of "not cartooned" or "not abstracted". "Photorealistic" is something that could be mistaken for a photograph, but is clearly not under close observation. "Photo (medium)" is for literal photos, though because of the nature of Illustrious's training data a large chunk is photos of artificial objects (e.g., anime figurines).

@NanashiAnon Very Interesting, thanks for the breakdown and for reminding me of your article, I actually had it bookmarked already so i must have come across it a while back and only glanced through it briefly intending to go back to it (then forgot to :P).

Can i assume that the 'photography (artwork)' tag is congruous to the 'photo (medium)' tag? I have a couple of tag CSV's that i use with ExTagComplete and that's high up on my Danbooru/E621 merged list as an E621 tag so i'm assuming it's that sites version of Danbooru's medium tag.

I do have Danbooru's tag groups wiki section constantly open for reference but i've not delved into the mediums group, generally using it for gestures/clothing/poses that i know but not by name like Akanbe (eyelid pull tongueout) and Wariza along with it's variants to name but a few. It's a great resource for that kind of thing to me.

@NeilSylver I have no idea. I only know Danbooru tagging.

@NanashiAnon There's a lot of crossover with slightly varying names. Of course, it was my previous checkpoint that supported those tags, my current one is pure Danbooru now so I should probably stop using the merged set to avoid non-working/unsupported tags :P

@NanashiAnon You wouldn't happen to have any knowledge of non-tag concepts that work in Illustrious? I've found some that work quite well despite not actually being tags, like soft features, cool/warm undertones and facial feature modifiers like 'broad nose' or Almond eyes' 'though those two may simply be a known concept/tag applied to a feature. I experimented with some in my photorealistic test posted above but it's very trial and error seeing what works.

@NeilSylver Beyond a guess those are things base SDXL understands and aren't covered by ILXL's extra training, no.

@NanashiAnon No worries, Just figured i'd ask in case. I did do a bit of digging into Illustrious documentation about it, never occurred to me to try SDXL documentation too. I've got a ridiculous amount of concept notes that i play with behind the scenes from things like rating labels (Rating:safe in the positive seems to perform better than 'censored' at curbing models inherent horniness imo) and source labels (anime/cartoon/film/3D, one of which is 'source:photo' which i'm going to try against photorealistic/photograph to see if it's comparable or gives different results.

@interfusor You're welcome :) Looking back, I wish i'd have tried a few different samplers when doing the random characters to see the differences. Maybe next time :P