Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
Well, I present you the LOCON version of my Lucy McMillan LORA , who fights for first place with Sharon Apple for the title of best girl in Macross Plus. To be honest i was just going to Make Myung's Lora, Sharon...
Source du modèle
Description de la source
Well, I present you the LOCON version of my Lucy McMillan LORA , who fights for first place with Sharon Apple for the title of best girl in Macross Plus. To be honest i was just going to Make Myung's Lora, Sharon Apple's appearance is simply too variable(Begin either a hologram or a box shaped murderous AI) to do anything but then I decided that if i was going to do a screen capping run i might as well collect Lucy's dataset too. Oh boy... re watching the OVA made me remember how much passive-aggressive and generally batshit insane Myung is. Kudos to Guld for dying and dodging that bullet. So yeah in the grand scheme of things Lucy is a better woman, hell, Sharon too.
As expected, images were scarce and her outfits either generic or ugly(her operator uniform). So I didn't explicitly tag them. What is on the other hand explicitly different is her hairstyle in her loose and half_updo ones.
Her freckles mostly work, you can add more but i wouldn't recommend going too high as that tag seems to pull the model to photorealistic and at some saturation point it gives me Trypophobia(odd that, not sure how they trained the concept to have that effect).
Sources
1 sourceArtefacts du modèle
1 artefactExtraits de sources
1 extraitSource context: 165 downloads · Type LoCon · Base model SD 1.5
So Instead of outfits I explicitly trained her with hairstyle with a bow variant:
half_updo_red_hair_bow: Basically her ponytails-ish hair variant.
Works fine with: lucy_mcmillan, half_updo_red_hair_bow, red_choker
That will give you the closest to her classic cannon appearance, her loose hair being the true alternate form.
As For dataset it is almost entirely screencaps and some images from V1 to try to stabilize it. So no dataset for you as always instead i will add the tag summary so people can experiment.
