(I forgot to say that the krea2 model support the full 14 character even if i did not stated them all in the trigger list i just did not have time to prepare a image for all of them ! )
(new edit: Crazy how that simple experience on a random anime that I liked the female organisation lead to so many character pack x) so of course If one was getting a trial run for Krea2 it was going that one, thank for the day 1 14 character pack x) supporting this pack is what made the other happen ! Krea2 was a bit more expensive to do but liking the model so i'll give it shoot i don't think it's that much more expensive for single to 3 or even 6 person lora x) but man 14 i had to do so much optimisation in my training setting lol any feedback exept like the size of the lora which i know is big and hope to work on that in the future by testing if downscalling the size after training would reduce it because i wanted to keep the big versatility of Krea2 on every character as a 800 mb version mixed up character x) ) (I also updated the guide with my setting and stuff for krea2 just a small update since the same general concept)
(Edit:made an article on what setting and stuff I used in case ! https://civarchive.com/articles/29782/how-i-do-my-group-character-lora-for-anima-and-illustrious-from-frankainstein )
There is a lighter version of both Lora but all previews image on both model were done on the full version, not the downscalled thought I made some test on them and they work just in general for consistency and adherence the bigger are better for all those character !
I got a break today and decided to finish this ! it's another test at a big character pack ! This time not just 3, but 14 female character of eminence of shadow ! The list are in the tags but basically all Shadow garden member exept maybe Chi...I forgot her and Already had thousand of image and prompt to deal with...and then I added like some less shown character like the martial art lady, witch of calamity and the purple sword girl be kind on them i had lot less to work with for them !
prompt were trained on natural sentence and danborus but recommend using danboru for both illustrious and Anima thought for Anima you can use the danboru tags for the character then switch to natural sentence...as it was too many char to do just Delta do this, but i did made prompt like Delta do this, Alpha is there, side views of Beta stuff like that in natural sentence you can use their name but i would encourage having their danboru tags at the start.
Description
Anima version !
FAQ
Comments (6)
Thanks for the people already out there dowloading it and using it ! As I said it's my first lora that is this big and character pack ! Really hope it'll work great for all of you be it anima or Illu x) if it's provide good result for most of you maybe i'll do a league of legend one once but super busy ! so no promise as to when x)
this is awesome, i wanna try my hand at a Naruto version of this, could you share with me the settings you used for this.
@Ceylon_Ai NETWORK DIM (LINEAR RANK)
64
NETWORK ALPHA (LINEAR ALPHA)
32
CONV DIM (CONV RANK)
32
CONV ALPHA
16
LEARNING RATE
0.00005 (more or less it was my first attemps and like I said a thousand image well like 960 or something and took me a hour to train on a cloud gpu with 80gb ofvram so it might take a lot of time if u wanna do the same number of people ! ) I might do a small post on my group character methodology x) it's just still tinkering with it to say it's the definitive way you can probably optimise it you might not need to use rank 64 if you are not doing that many character for a trio lora I used 32 rather than 64 and you probably can co lover than 32 if the character are similar in style and lot of way.
@Ceylon_Ai I pushed out an article if u wanna check https://civitai.red/articles/29782/how-i-do-my-group-character-lora-for-anima-and-illustrious-from-frankainstein
太棒了，我一直在烦恼怎么把不同的女主角们放在一起3P
Thank for the word, glad you enjoying it and taking the time to comment I did manage to make a duo pic but did not try 3P.
This lora was for goal to make more of a character pack in one ! So i did not focus on group compared to my trio lora of superpuff as Already was a 71 with trying to cover every angle and everythign for each one of them x) You know by exemple for a show with 3 sister it's easy for me to know how to compose them and make training data but something like shadow garden it would take much more to figure out which character i train wit witch specially since it not just an anime there is also a gatcha game etc x) so i focused on just having them all work first !