Would appreciate images, feedback, requests, reviews, and of course comments in the comments. Please check out my other models if you like this one.
DO NOT USE Emilia_\(re:zero\) it makes the results a lot worse -(mainly just the clothing so if going for a different outfit it may help)
V7 info:
Emilia from Re:Zero. best weights 0.6 - 1
Normal outfit: white_dress, pleated_white_skirt, white_thigh_boots
Normal optional: golden_eagle_on_sleeve
Cathood: fake_animal_ears, white_hood, hood_up
Nightgown: EmiNightGown, drunk_nightgown_outfit, purple_nightgown
Nightgown optional: side_braid
Swimsuit: alternate_hairstyle, high_ponytail, sidelocks
Swimsuit Optional: swimsuit ( not sure if this helps or makes it less accurate, seems 50/50)
Shorthair: white_dress, pleated_white_skirt, white_thigh_boots, alternate_hairstyle, short_hair
Shorthair optional: golden_eagle_on_sleeve
Overall: grey_hair
Overall optional: BLPPL, blue_pupils ringed_eyes ( BLPPL is an attempt i made at getting her unique eyes to trigger, seems random if it helps or not, some models seem to do better than others, haven't tried inpainting with it)
you can just do normal Emilia with short hair, probably didn't need to be a standalone thing
For other models just copy the prompts or use: white_dress, pleated_white_skirt, white_thigh_boots, crown_braid, blunt_bangs ( also the tags below this as well should work)
base hairstyle: blunt bangs, crown braid, white hair (or purple),
capable of nsfw
tags grabbed from a danbooru image: grey_hair, purple_eyes, thighhighs, hair_ornament, pointy_ears, blush, grey_hair, long_hair, cleavage, flower, x_hair_ornament, hair_flower, wide_sleeves, pleated_skirt, medium_breasts, skirt, bare_shoulders, white_thighhighs
might have rainbow highlight in hair, try putting it in negative prompt (although it still popped up usually)
model used: MightyCorneo | Stable Diffusion Checkpoint | Civitai
"bakedVae" version https://civarchive.com/models/23900/anylora-checkpoint
base Corneo - https://civarchive.com/models/4669/corneos-7th-heaven-mix
- accidently unpublished got worried for a sec lol
Description
IF COPYING PROMPTS THE Lora NAME WILL BE DIFFERENT, REPLACE IT WITH THE NAME OF THE ONE YOU DOWNLOADED
V6 is much more consistent and accurate, but might change the artstyle? (I can't really tell as my current models have similar artstyles)
can also do short hair. can't do her eyes, but might be able to inpaint them, if anybody has tips for training unique eyes they would be appreciated.
recommended weights 0.6 - 1.0
sometimes her base outfit will combine with your prompted one, if tthat happens lower wight, change prompt, or do batches. if using another Lora either lower or raise the weight, going up to 1.2 usually helps keep her appearance while using another high weighted Lora. ( it is also good if some part of her appearance is off)
her outfits besides her main where in the dataset but it seems to have trouble with them so i didn't include any examples but the tags are
black lost in memories swimsuit: LiN_memories_swimsuit_outfit, alternate_hairstyle, high_ponytail
nightgown: drunk_nightgown_outfit, purple_nightgown
cat hood: cat_hood_outfit, hood_up, cat_hood
her hair ornaments and puck's gem are baked i so getting rid of them might require lowering weight or luck
using her booru tags helps usually but isn't necessary: grey_hair, purple_eyes, thighhighs, hair_ornament, pointy_ears, blush, grey_hair, long_hair, cleavage, flower, x_hair_ornament, hair_flower, wide_sleeves, pleated_skirt, medium_breasts, skirt, bare_shoulders, white_thighhighs
FAQ
Comments (7)
Only works with Additional Networks?
Can you release a version that works as a SD standard Lora?
this is extremely late, but it should work with standard Lora, these except the newest version are all normal loras
oh, unless you mean base SD, which I don't have installed anymore since I don't use it. but mixing a bit of an anime model with the base model should allow it to work if you willing to change the style a bit
Morning clothes doesn`t work
yeah, I think i changed the dataset or it was just so infrequent. lowering the Lora weights and increasing the weights on the tags like (pink_dress:1.3) gives alright results
What model was this trained on?
NAI (NovelAI)