Loading
Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
The Ai Hentai girl Projekt Melody in her original appearance from the Vtuber group VShojo . Works best around 0.8-1.0 weight . Training done on NAI.
Source du modèle
Description de la source
I have a request form
The Ai Hentai girl Projekt Melody in her original appearance from the Vtuber group VShojo . Works best around 0.8-1.0 weight . Training done on NAI.
Trigger : " ProjektMelodyOriginal , bandaid on nose, short hair, ponytail, hair ribbon, hairclip " always needed, " medium breasts " suggested for accuracy.
Suggestions/notes :
Leotard Outfit: " black choker, beltbra, underboob, harness, leotard, navel, white gloves, elbow gloves, white thighhighs "
Beltbra + Shorts Outfit: " black choker, strapless, beltbra, underboob, white gloves, elbow gloves, white shorts, micro shorts, white thighhighs "
Crop top Outfit: " halterneck, strapless, white shirt, print shirt, off-shoulder shirt, crop top, short sleeves, micro shorts, white shorts, white sneakers "
Sources
1 sourceArtefacts du modèle
1 artefactProjektMelodyOriginal_V1-Manityro.safetensors
safetensors · 58,5 MB · SHA-256 372ac93546ec…f5f9 · Civitai
TéléchargerExtraits de sources
1 extraitSource context: 590 downloads · Type LoCon · Base model SD 1.5
" halterneck, strapless " can be removed to mix it up.
" Elbow gloves, white gloves " can be added to get the look with them under the sleeves.
" White shirt " or " Black shirt " work for color changing.
Remove " White sneakers " if you want upper body shots.
Swimsuit Outfit: " black bikini, swimsuit, halterneck "
Sometimes gens funny bikini bottoms as the data wasn't the best.
She is very flexible clothing wise and NSFW able with almost every piece being removeable one by one so please go wild and have fun :)
Been a while due to some technical issues of my own doing but now I am back to make some more loras. Melody is one of my favs out there so I really wanted to make a great lora of her as the other ones of her left a lot to be desired. I split her original appearance from the newer Vtuber model to avoid bleeding issues so I will be making another lora for her newer looks down the line. I hope you all enjoy.
Feedback and reviews are always appreciated.
Nerdy training numbers:
Trained on D8Dreambooth trainer
Optimizer: 16bit AdamW
Training resolution: 768
Unet LR: 0.0001
Tnec LR: 0.00001
Unet weight decay: 0.024
Tenc weight decay: 0.030
35 Epochs - 6960 Steps
Trained on 175 images.
