used the civitai onsite lora trainer for this one which is a pretty convenient tool for low vram brainlets like me (note: do not let it pick the prompts for the three test images itself, it managed to choose the least relevant tags posssible from the dataset and just generate garbage)
not sure if its actually better than the last version because its pretty iffy when it comes to hands, feet and eyes, requires a ton of generations to get something decent
they werent trained on it but i strongly recommend using the prompts: pupils are black dots, asymmetrical hair
and the negatives:long hair,ponytail,pigtails,hoodie,2girls,3girls,sidecut,undercut,facial piercings,monochrome, blonde hair
used the civitai onsite lora trainer for this one which is a pretty convenient tool for low vram brainlets like me (note: do not let it pick the prompts for the three test images itself, it managed to choose the least relevant tags posssible from the dataset and just generate garbage)
not sure if its actually better than the last version because its pretty iffy when it comes to hands, feet and eyes, requires a ton of generations to get something decent
they werent trained on it but i strongly recommend using the prompts: pupils are black dots, asymmetrical hair
and the negatives:long hair,ponytail,pigtails,hoodie,2girls,3girls,sidecut,undercut,facial piercings,monochrome, blonde hair
because it really likes going offscript
good luck