1.1v
-All suit tags have been changed to avoid confusion.
-PD suit should be more solid
TV: hoshi-ruri, hoshi-bridgetv, 1girl, alt-twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, school uniform, sleeveless vest, orange vest, black tie, collared shirt, white long sleeves, emblem patch, black skirt, fitted outfit, formal attire, academy uniform, professional design
PD: hoshi-ruri, hoshi-bridgepd, 1girl, twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, military uniform, sci-fi outfit, high collar, white cape, gold trim, black bodysuit, armored sleeves, gold cuffs, asymmetrical design, shoulder guard, decorative shoulder pad, red accents, short gloves, fitted outfit, functional design, black thigh-high stockings, integrated leggings, segmented armor, tactical design
(The tv version is still very unstable.)
1.0v
ko-fi commission!
I honestly don't know what to recommend here
All images were generated in: WAI-NSFW-illustrious v9
Basic Tv: hoshi-ruri, hoshi-bridgetv, 1girl, twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, large and rounded shape, sharp bold upper outline
Basic Prince of Darkness: hoshi-ruri, hoshi-bridgepd, 1girl, twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, large and rounded shape, sharp bold upper outline
Bridge outfit TV: hoshi-ruri, hoshi-bridgetv, 1girl, twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, large and rounded shape, sharp bold upper outline, necktie, uniform, hair bobbles, hair ornament, vest, orange vest, uniform, skirt, black skirt
Bridge PoD: hoshi-ruri, hoshi-bridgepd, 1girl, twintails, yellow eyes, long hair, retro artstyle, grey hair, soft golden-brown eyes, large and rounded shape, sharp bold upper outline, outline, long sleeves, white gloves, capelet, white capelet, uniform, dress, white dress, skirt, black skirt, pencil skirt, pantyhose
You can use “hoshi-bridgetv” or "hoshi-brigepd" to switch the version
(although tv is a bit unstable)
Description
FAQ
Comments (3)
It looks like I found the reasons of unstability.
For training you use a many long tags which make learning slower and harder for neural network.
For example,
soft golden-brown eyes ---> should be "yellow eyes"
large and rounded shape --- what the heck is that? 80 instances of it in dataset.
sharp bold upper outline --- I guess it should be untaggable because it's the style, so it needs to be removed.
delicately curved lower edge --- why did you tag this one? I think it should be removed.
high-contrast highlights for a glossy effect --- the same
"smooth tonal gradients" - the same, should be removed
"subtle shading for depth": the same, should be removed
"luminous and innocent expression" -- the same, should be removed
All these tags are pointed out for 80 pictures out of 101. So that means these tags are becoming style tags and of course your lora is quite unstable.
BTW tag "alternate hairstyle" should be used for unofficial hair styles.
Dude, you're too demanding of people like the author of this LoRa. They don't care about quality at all, as long as the LoRa is minimally functional.
@magenta Well, because of these kind of thoughts we can't have nice things.
I understand the issue with this lora. Author used Joycaption tool for tagging, not wd. Joycaption likes to create booru tags which do not exist.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.