For drawing a girl with Ultraman outfit. See right panel for trigger words.
Additions of "ultraman bodysuit" may get better results.
If you want SDXL version LoRA, try Ultragirl XL.
v2.2-Blue
Focused on creating ultragirl with "Blue" color-timer and compensate v2.1-Blue's "Night Sky" background. So far no major problems with this version.
Happy holidays!
v2.2-Red
Focused on creating ultragirl with "Red" color-timer and compensate v2.1-Red's "blue bodysuit" tendency. I didn't think modifying certain tendency is this difficult, creating dozens of LoRAs and failed miserably. This one is still not satisfying but for now this is the best among them, so I publish this.
v2.1-Red
Focused on creating ultragirl with "Red" color-timer. As promised.
Still get "Blue" color-timer occasionally, but probability of getting "Red" color-timer is high enough IMO.
Note: This LoRA somehow tend to generate "Blue" bodysuit. If needed, add "red bodysuit" or so to compensate. The fix is planned alongside with v2.1-Blue's "Night Sky" background.
v2.1-Blue
Focused on creating ultragirl with "Blue" color-timer. "Red" version is planned, eta unknown though :P
Note: This LoRA somehow tend to generate "Night Sky" background.
v2.0
A bit improved version. Some may prefer v2beta... that's your choice.
Please note trigger word has changed to "Ultragirl", not "Ultgp". Use with "ultraman bodysuit" like before.
Also, from this version, I changed category to "Clothing" from "Character", since this LoRA intended to create "generic" ultraman outfit, not a certain character.
v2beta
Retrained one. I make many LoRAs for this theme but not get satisfying results so far. This one is still unstable but I'll put this to civital for now. Use with "ultraman bodysuit".
v1
This LoRA is not that strong; you can try weight 1.0 first.
Description
Compensate v2.1-Red's "Blue bodysuit" tendency.
FAQ
Comments (8)
Q1: where did you get the dataset for ultragirls?
Q2: how many versions are there, and do they have independent trigger words (like in a characterpack)?
A1:
Training images were gathered from various places before the rise of Stable-diffusion (I like Ultragirls :-) ). I didn't record where they come from, but I suppose mostly from old Twitter and Pixiv. Some are from danbooru too.
A2:
No, my LoRAs are not a characterpack (i.e. no independent trigger words). Versions of this series are just a result of repeated iterations for better ones.
@dolphin42 Q3: are there ways of using trigger words to refine unique or consistent designs?
Q4: Will there be Ultimate Girls LoRA?
@TomLucidor
Q3: If you mean patterns of bodysuits, I don't think there is any, as training dataset does not have distinct trigger words for each characters. ControlNet may be usable for that job.
Q4: Is this Ultimate Girls https://en.wikipedia.org/wiki/Ultimate_Girls ? Hm, maybe, if you know/have good source of datasets (images).
@dolphin42 Yeah for Q4, thinking getting dataset directly from anime is possible? Cus it is slightly niche for fanart
@TomLucidor
Creating a LoRA for Ultimate Girls now. Current (1st) batch sample: https://civitai.com/images/21617935
Quite blurry, but it seems it works somehow. I'll try some more parameters / datasets edits.
@dolphin42 Looks good as test LoRA. Thinking anime upscalers + psuedo-characterpack type data planning (individual character in image) might help?
@TomLucidor A bit modified version of the LoRA (Lycoris) is now released. https://civitai.com/models/606542
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.