Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Check out all the Art styles I’ve made: https://civarchive.com/user/CitronLegacy/models?tag=style
This version has tons of great data in it and is better than ever at making 2 character images as well as cinematic art!
Its fair to say I've put more effort into each of my Prisma Illya Style Loras than I have any other Lora.
For some reason I always end up generating tons of images and then using those to refine the style. I'm pretty sure Fate UBW would have been a better style but I have a very detailed dataset for this so I keep making it.
I barely remember the Prisma Illya anime (maybe I didn't actually watch all of it?) but I love the music.
When creating epic art please listen to Emiya Illya's theme:
Description
<lora:PrismaIllyaIXL_v2:1.0>,
FAQ
Comments (11)
I can't help but notice that the samples are all Illya and Chloe
Yeah I was wondering. Does include Miyu and the other girls? Because I want the whole cast to be added.
Yes it makes Miyu. I made a lot of custom art and scenes so there are extra things hidden in this Lora. Please experiment with this Lora to see if you can unlock all its secrets. ;3
I was focused on Chloe and Illya because I really like their dynamic.
I'm posting a lot of Miyu art now.
@Netryn Yes Miyu is here. https://civitai.com/posts/9846235 This is mostly a style Lora so it focuses on the main 3 and images containing 2girls. It might know about other characters but I think you'll need character Loras to get uncommon characters like their classmates.
A style Lora and character Lora for 3girls all in one is amazing, like damn man good job. Also remember that IL knows more characters so if you want their outfits look for them on Danbooru to get the prompts/tags and it should work.
Thanks! Back in SD1.5 someone made this style and it was one of my favorites so I put a ton of effort into making this as good as possible.
Plus I think Multi-Girl Loras are a really fun concept because it makes the image composition so much more interesting/dynamic.
Yes! One of my favorite ways to use Illustrious is to copy the tags from a Danbooru image and see how closely IL can replicate it.
@CitronLegacy When you train Multi character Lora do you put each character in a separate folder or just tag them and put them in the same folder? I might train one myself at some point.
@Lancelot69lol I've tested multiple ways of creating multi-character loras. The first few ones I made are just the two individual datasets combined or two individual Loras merged. They would not go into separate folders.
I used a better technique for this Primsa Illya one is because I had a lot of images of Illya and Kuro together. In addition to the official art of them together I made a lot of AI generated art of them together.
I'll let you decide which technique you think is better.
Pony multi character comparison:
This one was individual loras merged together: https://civitai.com/models/620449?modelVersionId=693587
Both of these were trained on images of the characters together: https://civitai.com/models/990326 and https://civitai.com/models/1011422
Illustrious multi-character comparison:
Again this one was individual loras merged together: https://civitai.com/models/620449?modelVersionId=1069069
The Prisma Illya page we are on was trained with the characters together.
(Don't compare Illustrious with Pony: Illustrious is way better at multi-character than Pony)
@CitronLegacy oh right I forgot you could merge Lora together. I have also been thinking on wheather or not using images that have more than 1 character for a multi Lora was a good idea or not but from what you've tested I guess it works.
Thank you for the explanation, I may have to try this for a multi Lora for Momo/Nana from To Love-Ru, they are often together so they would be good to test this with.
@Lancelot69lol Mono and Nana together sounds great! I need to check if the Illustrious ToLoveRu lora can do that! I'm actually looking for more multi-girl projects so I'd be curious to hear your ideas.
.
Yes multi-character images is good for multi-character loras.
You dont have to do what I'm about to describe but its an extra step that works well:
I make sure to have images for some of the scenes I wanted my lora to be able to make. For example, this image is a specific 2girl scene I had several images of https://civitai.com/images/44060838 when I trained the Lora. I have training data saying "this is what it should look like when they hug" or other scenarios.
I did a lot of "specific scene training" for this lora: https://civitai.com/models/1051362
This scene was training with other characters but because it had enough repeats I was able to replicate this scene and have it work for other characters: https://civitai.com/images/46335979
@CitronLegacy What a good idea, it makes sense to include scenes you'd want to replicate. Thank you for the advice 🙏
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.