Type | |
Stats | 167 271 53 |
Reviews | (32) |
Published | Jul 8, 2024 |
Base Model | |
Training | Steps: 3,288 Epochs: 13 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | quindef nakano nino, nakano miku, nakano yotsuba, nakano itsuki, nakano ichika multiple girls, siblings,5girls, sisters, long hair, short hair, brown hair pink hair,red hair,orange hair,two side up,blunt bangs, blue eyes, hair ornament,ribbon, bow, hair ribbon, star (symbol), bowtie, black ribbon, headphones, green bow, star hair ornament,green ribbon,butterfly hair ornament,headphones around neck, skirt, shirt, long sleeves, hair between eyes, school uniform, jacket, white shirt, ahoge, short sleeves, pantyhose, pleated skirt, open clothes, collared shirt, sweater, black jacket, black pantyhose, cardigan, green skirt, sweater vest, clothes around waist, red sweater, black cardigan, blue cardigan, yellow cardigan, cardigan around waist, |
Hash | AutoV2 45C3BA062C |
PLEASE READ BEFORE MINDLESSLY USING IT:
use weight 0.85-1 for more accurate results, lower weight for artstyle closer to your liking
After making multiple 2 character in one image models i decided to test how far i can go with it still working. It seems there is no limit with this very simple method. I usually do not share 2 char models due to added difficulty for the users but i think many might like this one.
In any case creating multiple characters in one image tends to take many attempts here are some tips to make producing good images faster:
If you use hires-fix, add a preview image method to see if the characters are right. i use a preview image node in comfyui to decide if i skip the image.
Learn fixing minor errors manually. I started to learn last week and I can finally release images i would otherwise not be able to release before without making a fool of me due to bad hands or legs misplacement.
third method: make many low resolution images until you find one image you like then use your image improvement steps on those you like
I did not let the model absorb the tags because i felt it is more reliable especially with multiple characters. so you have all the tags and flexibility to work with, which i prefer
the only trigger word is quindef, the other tags like sisters name were given by the anime tag program and may help as many checkpoints seem to understand and have data of the sisters.
Lastly:
Feel free to post your best results !