Lora trained on 62975 images of Blue Archive Characters
This is missing a few characters and only has 86 characters in total.
This took me over 24 hours to make since I only have a 3070ti 8gb.
I release all my models for free, if you feel like supporting me and my work you can support me here and buy me a drink.
Please note that some characters may not come out perfectly due to not individually being able to select the epochs since this is a combined training and not a merge. If you find this to be the case use the individual character model instead.
Default Prompts:
yuzudef, yuukadef, yoshimidef, wakamodef, uidef, tsurugidef, tsukuyodef, tsubakidef, tokidef, saoridef, miyudef, mikadef, maridef, kokonadef, koharudef, shizukodef, shirokodef, shiguredef, serinadef, serikadef, senadef, sayadef, sakurakodef, sakidef, nonomidef, nerudef, natsudef, nagisadef, mutsukidef, momokadef, momoidef, moedef, miyakodef, misakidef, minedef, mimoridef, midoridef, michirudef, megudef, mashirodef, marinadef, makidef, koyukidef, kotoridef, kazusadef, kayokodef, karindef, kannadef, kahodef, kaededef, junkodef, izunadef, izumidef, irohadef, ioridef, hoshinodef, hiyoridef, hinatadef, hinadef, himaridef, hifumidef, hibikidef, hasumidef, harunadef, harukadef, haredef, hanakodef, hanaedef, fuukadef, fubukidef, eimidef, chisedef, chinatsudef, chihirodef, cherinodef, azusadef, ayanedef, atsukodef, asunadef, arudef, akaridef, airidef, shundef, arisdef, akodef, akanedef
Alternative Outfit Prompts:
akanebun, asunabunny, asunaschool, hasumigym, hibikicheer, hifumiswim, hinaswim, hoshinoswim, ioriswim, izunaswim, karinbunny, marigym, nerubunny, nonomiswim, shirokocycling, shirokoswim, shirokoterror, tokibunny, wakamoswim, yuukagym, azusabeach, chiseswim, hinawhite, karincasual, kayokonewyear, miyuswim, mutsukinewyear, sakurakojustina, serikaswim, tokisuit, hinapajama, shizukoswim, ayaneswim, fuukakim, hanaexmas, hanakoswim, harukakim, harunakim, kotoricheer, mashiroswim, serinaxmas, tsurugiswim, arisbunny, arismaid, arunewyear, chinatsuspring, junkonewyear, koharutrack, serikanewyear, yuzumaid, cherinoswim, izumiswim, serinanurse, shigurespring, marijustina, sayacasual, arissailor, atsukounder, harunagym, mariswim, yuukaflight, aruclip, asunacheer, azusasport, harunaswim, koharuswim, yuukaidol, koyukibunny, miyubunny, azusadress, miyugym, arischeer,
Random Prompts:
yuzurnd, yuukarnd, yoshimirnd, wakamornd, uirnd, tsurugirnd, tsukuyornd, tsubakirnd, tokirnd, saorirnd, miyurnd, mikarnd, marirnd, kokonarnd, koharurnd, shizukornd, shirokornd, shigurernd, serinarnd, serikarnd, senarnd, sayarnd, sakurakornd, sakirnd, nonomirnd, nerurnd, natsurnd, nagisarnd, mutsukirnd, momokarnd, momoirnd, moernd, miyakornd, misakirnd, minernd, mimorirnd, midorirnd, michirurnd, megurnd, mashirornd, marinarnd, makirnd, koyukirnd, kotorirnd, kazusarnd, kayokornd, karinrnd, kannarnd, kahornd, kaedernd, junkornd, izunarnd, izumirnd, iroharnd, iorirnd, hoshinornd, hiyorirnd, hinatarnd, hinarnd, himarirnd, hifumirnd, hibikirnd, hasumirnd, harunarnd, harukarnd, harernd, hanakornd, hanaernd, fuukarnd, fubukirnd, eimirnd, chisernd, chinatsurnd, chihirornd, cherinornd, azusarnd, ayanernd, atsukornd, asunarnd, arurnd, akarirnd, airirnd, shunrnd, arisrnd, akornd, akanernd
Trained on NAI, gen'd on AOM2 hard
Description
FAQ
Comments (39)
AHHH IT'S FINALLY HERE
praise the loooooord
I pray for your poor 3070ti and Powerbill 🙏
yeah power bill in summer is big lol
How did you do this?
magic, but in all seriousness several days of work goes into making large packs like these, at least for me...
Collecting datasets, sorting datasets, pruning datasets, tagging datasets, training the individual models, then finally I just take all the datasets from the individual models and train it all at once. The whole training models individually is just what I do because it gives better results at lower epochs per 1 character vs the higher epochs required for all characters in a combined training (ie just gives choices).
@Numeratic wow, that's a lot of work. Great job!
haven't tried it yet but bruh... that's some serious commitment. kudos to you
can you do Rachel Gardner from angels of death pls Ps: I love the lora you make
I dont even play BA, but I believe this Lora is so crazy that everyone should take it a try.😳
I dont play blue archive either lol
THE ABSOLUTE MADMAN WENT AND DID IT!
Will try later
dude, this is so incredible! I am kneeling on my gaming chair right now while using this masterpiece.
Any interests on Arknights? I play that too, just wondering if u play the same games as mine. LOL
Never played blue archive before or genshin impact for that matter lol. I've only played azur lane arknights and nikke but even then ive uninstalled those from gacha boredom...
@Numeratic Well, i also stopped playing all those games, nonetheless, i also would love Arknights and, i know thats quite a lot to ask, but an azur lane lora would be what i want the most.
holy shit, I'm surprised it can all fit in 40mb..
..ill keep using the individual ones just because its easier to work with and probably preserves more detail..
Yeah individual ones would give more detail since they are more to the epoch comparatively. However it is nice to just be able to have 1 lora toggled and prompt multiple chars at once with regional prompter.
Could you please explain what's the difference between def and rnd when we write tags? Thanks for your work, I will recommend your work to my followers
Hi there, def just means their default outfit. Where as random or rnd just holds all their random outfits and should be more flexible than their default outfit prompt for putting em in different clothing. Also there are some outfit tags as well that just do some of their outfits that they had enough pictures for if you read the description i list all of them. Ive also found that just putting the characters name without rnd or def can also cause the character to appear.
335 640 steps in 24 hours mean an average of 4 it/s with a 3070 ti. How do you achieve this magic, great wizard?
I only get max like 3.86it/s when training batch size 2 resolution 512x512
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
But also I had everything shutdown on my pc and my monitors off, who knows maybe that helps too.
I downloaded your genshin pack and now i find another one made by you, even bigger this time. I love you, whoever you are, I hope your life has nothing but pleasure for you
The text on the chart is blocked by the pictures, making it difficult for me to get the key names, which is a pity
Trigger words are in the words list and in the description.
@Numeratic Thanks for your reply, but what I meant was that tag combined with pictures and made into grid pictures would identify characters better, after all I've never played Blue Archive😂
@B80_8888 ive never played it either lol
Amazing!! Added to the big list https://civitai.com/articles/262 Thanks!!
Is there yurizono seia 百合園セイア?
Interesting lol
Can anyone help? I'm new to this, and I don't know how to open the file after I downloaded them, plsss help! இ௰இ
you cant open this, you need to put in the stable diffusion lora folder, search in youtube, how to install and use loras
Hello, there are many new characters added to "Blue Archive", can you update this model? Thanks.
Also hope that you'll come back and do an update for SD and not an xl lora.
Will you update lora for new characters as seia, ichika or nozomi?
chinese?
Hi, how do you make a LoRa for so many characters can you explain? It seem LoRa get confuse with more than one face, even after I name it properly. with character name
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.
