There are 2 versions. The fine-grained used more tags during training. But as shown in the samples, that also means just prompting ViviAll, might make a Vivi with red hair. I prefer the first model, but its your choice.
V1-finegrained works with pysss View LoRA details!
Here are the tags for the fine-grained version:
Generally
ViviAll, pink_hair,1st outfit
ViviAll, 3d, purple_hair, pink_hair, multicolored_hair, hair_intakes, hair_bow, bare_shoulders, ahoge, blue_eyes, smile, ViviDef, twintails, bra_over_clothes, midriff, detached_sleeves, crop_top, choker, chest_jewel, baggy_pants, boots, navel, suspenders_hanging, short_shorts, chaps, thighhighs, handbag, utility_belt, thigh_gap,2nd outfit
ViviAll, 3d, ViviFlow, earpiece, dress, microphone, side_ponytail, hair_bow, long_sleeves, pearl_bracelet, puffy_short_sleeves,
bare_shoulders, blue_eyes, spiked_collar, belt, boots, bra_strap, single_thighhigh, thigh_straps,DRAFT (Can't private so updating live)
I'm trying to re-add ss_tag_distribution to the model so it can be seen in ComfyUI with the Pysss custom nodes for ComfyUI. Alternatively:
3d, 71
stage, 59
bare_shoulders, 47
ViviDef, 46
smile, 41
ViviFlow, 41
midriff, 39
twintails, 36
hair_bow, 32
dress, 30
crop_top, 29
hair_intakes, 28
earpiece, 28
side_ponytail, 24
detached_sleeves, 22
boots, 21
belt, 21
navel, 20
spiked_collar, 19
ahoge, 18
single_thighhigh, 18
thigh_straps, 18
closed_eyes, 17
long_sleeves, 17
short_shorts, 16
choker, 16
chaps, 15
puffy_short_sleeves, 15
thighhighs, 14
close-up, 14
multicolored_hair, 14
pearl_bracelet, 14
suspenders_hanging, 13
bra_strap, 13
arms_up, 12
baggy_pants, 10
from_behind, 10
eyepatch, 10
thigh_gap, 8
utility_belt, 8
blush, 8
handbag, 7
;), 7
pink_hair, 7
purple_hair, 7
armpits, 7
outstretched_arms, 6
microphone, 6
on_the_street, 5
blue_eyes, 5
waving, 5
nightclub, 4
from_below, 4
multiple_girls, 4
simple_background, 3
v, 3
ass, 3
facial_tattoo, 3
heart-shaped_pupils, 3
black_dress, 3
sitting, 3
dancer, 3
hoodie, 3
eye_mask, 3
rock_on_(gesture), 3
sandals, 2
bathtub, 2
ViviYuk, 2
yukata, 2
breasts, 2
double_v, 2
full_body, 2
licking_lips, 2
hand_on_own_hip, 2
mouth_mask, 2
demon_wings, 2
demon_tail, 2
blood_splatter, 2
furrowed_brow, 2
dancers, 2
sweatpants, 2
thong, 1
open_mouth, 1
sleeping, 1
index_finger_raised, 1
soap_bubbles, 1
leaning_to_the_side, 1
armpit_crease, 1
flower_in_hair, 1
paw_pose, 1
lips, 1
fingers_to_mouth, 1
houshou_marine, 1
swept_bangs, 1
straddling, 1
finger_on_chin, 1
mask_pull, 1
devil_horns, 1
from_side, 1
gloves, 1
hands_together, 1
index_fingers_together, 1
exhausted, 1
sports_bra, 1
leg_up, 1
coat_on_shoulders, 1
sarong, 1
skirt, 1
leg_warmers, 1
black_hair, 1
squatting, 1
finger_frame, 1
bedroom, 1
grocery_store, 1
shopping_cart, 1
pointing_at_viewer, 1
:3, 1
I use my own script to help me tag videos. Code is available at: https://github.com/LiuYuer67/VideoTagger
I also provide 'save files' which list timings, bboxes and tags taken for every single video used to create the training data.
Description
More tags during training. This gives you more control over how your Vivi should look, while also making it require more tags to specify what a Vivi is.
