Hey, thank you for everyone's likes, I just started a buymeacoffee account, since I had some serious hardware damages... It seems training does take a toll on the hardware. Please support~ thank you again o(////▽////)q https://www.buymeacoffee.com/reweik
V02 changes:
no main trigger, more stable, less comic (still have).
normally 1girl + 1boy should be enough to trigger it, if need help, use mother and son or siblings. if gender is swapped (occasionally), use tall female. or breasts, if the base model can't appoint the genders. there's a chance the boy will be taller, then use shota (it'll make the boy reaaaally small sometime). use size difference if the base model is really having a difficult time.
put solo in neg if the base model is biased toward solo, comic, panels if the lora triggers base model to do more comic than you'd like to see.
Concept of big sis and lil bro.
Basic prompts:
bigsislitbro (if the effect is not too strong on the base model, use this), girl and boy, 1girl, __description__ of girl, 1boy, __description__ of boy, action, size difference/ height difference if needed.
Necessary negs:
comic, monochrome, panels (because majority of dataset is manga pages), solo (to reduce the model's bias toward single person imgs)
I feel the civ doesn't like "size difference/ age difference/ height difference", had a difficult time to upload pictures.
Description
FAQ
Comments (8)
Awesome as always! Since big sis is here, will big bro eventually appear too?
I've seen a lot big bro loras, probably not []~(￣▽￣)~*, you can gender swap with this lora, just prompt tall male, short female, something like that. It is a fairly common concept in hentai
Doesn't work with the Regional Prompter extension. Nearly every image generated looks like a manga, even with the suggested negative prompt.
I advise the author to curate their data set to remove or crop the manga pages.
the dataset I was sent was 90% maga pagesヾ(≧▽≦*)o negs (comic, multiple views, panels:1.3)?
Hey, may I ask how you trained this Lora? What kind of pictures did you use for the size difference as well as what tagging was the best for you to achieve this? I tried a few settings for training one myself but most of the resulting Lora’s were barely flexible as well as often only show one character or two the same size.
a bunch of manga pages with big sister and little brother in it😂 to be frankly, the manga page indeed comes out way too often, thinking to re-do it... make sure the imgs you use for the dataset has most consist concept of 2 people in it and their body shapes must have obvious size difference. when prune the captions, make sure take out the body type tags.
@reweik Ty for the answer (: For the captions did you use keywords separated by words? Like a typical WD14 capturing, or did you describe the Szene in a continuous sentence? And do you describe both characters in detail? I tried a lot but for me the Lora’s turned out to not work properly. May you give me one example Tag of one of your pictures? Cause I am not sure if things like „standing next to boy/man“ or „seize/height difference“ should be included in the tags or left out
@Gwenyria Sorry, was busy making an easier to use one, just write typical WD14 captions, let the base model figure it out 😎
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.