If anyone has a backup of the works of graphic-muscle from DeviantArt or knows where to find them, please let me know.
I got my grubby little hands on an archive. Thank you to the person who linked me this. ❤
Just like my previous attempt at making a LoRA - this is a male-focused style LoRA of an old artstyle of ripped-saurian, a.k.a. graphic-muscle on DA (circa about 2017-2014), with some characters that they used to draw.
General Usage: (for versions other than Legacy)
(I'll keep it short this time.)
It has a trigger word - "r-s", but when prompting for a muscular, masculine male character with body hair you it most often works without it. Try experimenting with and without the trigger word.
The LoRA has a tendency to draw characters either without legs or unfinished legs below knees.
On very rare occasions it will draw fishnets over a character's chest instead of body hair. I don't know why but maybe it can be fixed with putting 'fishnet' in the negative prompt.
It will most probably create a simple background unless specified otherwise.
It's supposed to be designed specifically for Pony Diffusion V6, but (at least from my knowledge) the general rule of thumb is that it should work with other SDXL checkpoints as well.
Specific Characters:
❗Disclaimer❗: Most of them will probably not work very well, if at all. The tags that do work are included in the list of trigger words. In case you have a separate character LoRA from the list below,
batman, wolverine, spiderman, venom, black_panther
fright_knight, meta_knight, lanzer, shovel_knight, king_knight, sir_kibble
bowser, donkey_kong
ganondorf, hawkman, captain_falcon, mewtwo, incineroar, machoke, zavok, tizoc, bullet_bill, pyron, hanzo, roadhog, reaper_overwatch, ginga (these will probably not work very well)
OCs:
avilocity (there might be some problems with the face), impetaur, saurbital, bronco
the-wall, grobot (these two will probably not work)
Description
Used a different tagging utility to caption the dataset (TaggUI and SmilingWolf/wd-swinv2-tagger-v3 instead of kohya_ss with SmilingWolf/wd-v1-4-convnextv2-tagger-v2), along with training it with more steps.
One drawback from this is that more characters look worse or fail to generate (so far I've discovered that avilocity and bronco don't work anymore).
FAQ
Comments (4)
what would you say you added or fixed in v2?
I used a tagger that, from my perspective, captioned the dataset better. Also did more steps in training. I think this should improve the quality of the LoRA
i think you can find his gumroad with sketches on kemono and is active on x and blue sky as 'a pecs predator' if that's useful
Thanks for telling me, but I already searched through everywhere. I already received an archive of their old DeviantArt drawings from someone, which I'll be using for a next version of the dataset one day.










