Northstar (Alpha Flight) - Pony XL

a quick and dirty lora for the superhero Northstar (Jean-Paul Beaubier), from Marvel Comic's Alpha Flight. I say quick and dirty because it's trained on 18 images, (which is far less than I would normally use.)

Nonethless, in testing it came out surprisingly functional and not a complete hot mess, but I did note some observations below.

I will very likely make a version 2 once I can find/gen some additional images of sufficient quality to add to the dataset.

it's pony so it will do NSFW (woo French-Canadian Sausage/Cake!)

trigger: Northstar

additional keywords that might be helpful:

• Black Bodysuit/Bodysuit: self-explanatory. I used "Black Bodysuit" in all of my sample images below so I'm unsure if "Bodysuit" will have the same effect, but in tagging the dataset images, but were used.

• White Gloves: again, self-explanatory

• White footwear: self-explanatory (I used "footwear" instead of "boots" as sometimes this result in some checkpoints drawing, like, hiking boots.)

• pointy ears: probably not necessary, but there was a few (and in a data set this small 'a few' is a chunky percentage) where the tops of his ears were covered so omitting this might result in human ear (particularly if you have adetailer or anything else that tries to 'fix' faces.

• Bara: in my limited testing omitting "bara" resulted in a more twinky elfin Northstar then the hunky piece of Canadian beefsteak I was expecting. Will probably swap out the more twinky versions of him in the dataset the second time with "barafied" versions to avoid the skeeze.

• pulled up top: self-explanatory, dataset image where his top was pulled up exposing pecs, nipples abs. Should keep arms covered since none of the dataset images have him with bare arms (in uniform at least)

• pulled down bottoms: self-explanatory, dataset image where his pants were pulled down exposing penis, testicles or butt. Obviously prompt those things (unless you want to see a Northstar with ken doll genitals.)

• torn clothes/torn bodysuit: self-explanatory. there was 1 image in the dataset of 17 (so 5.88%) but it seems to work ok (see samples images)

Have fun using this for personal non-commercial use, blah blah etc, and I'd love to see what anyone generates using this.