About Illustrious/SDXL Version:
Main Trigger Word: J@zz@OC
Helpful Trigger Words: furry anthro, leopard girl, blue eyes, glitter eyeshadow, sideswept white hair, purple tips, tail, digitigrade, masterpiece
Currently pretty good, hands and feet still need some work and the tail is often too long, but the hair and body are very good. Randomly has cheetah-style stripe on muzzle or not, I do like that better, so future versions will try to stabilize that more. I'm releasing it now "mostly" finished, so others can enjoy it while I work on different things before coming back for an update. Not perfect yet, but I hope you have fun with it!
For Anima Version:
This is a LoRA for one of my OC characters, Jasmine "Jazza" Rausch. @CitronLegacy fell in love with this character, and even made a LoRA of her before I made a lore! Normally I prefer to do things myself, but since I rarely use Anima I didn't mind leaning on someone else's expertise for this one! The showcase images based on my works they shared with the model file which are uploaded here are gorgeous, and I tested it some and was really impressed, but full disclosure... I'm not good with Anima myself, so all credit for the LoRA itself and the 17 images in the showcase post here goes to @CitronLegacy!
Main Trigger Word: J4zz4
Helpful Words: J4zz4, blue eyes, white hair, short hair, multicolored hair, two-tone hair, furry female, body fur, animal nose, leopard ears, leopard girl, snout, purple hair, leopard tail, animal feet, digitigrade
Jazza is an anthro furry female snow leopard — not a human/elf character, so prompting should lean on furry/anthro descriptors (body fur, snout, animal nose, leopard ears, leopard tail, animal feet, pawpads) rather than human anatomy tags. Her signature look combines blue eyes with white, two-tone hair carrying purple highlights.
-Works best with Danbooru-style tag prompting, matching Anima's native prompting style.
-Combine J4zz4 with the furry/appearance tags above for the most consistent likeness; drop or swap tags to vary outfit/pose while keeping the character trigger.
-If you redistribute the Anima version LoRA file itself, please credit both the character creator and rights holder, @FreijaFoxy, and the LoRA trainer, @CitronLegacy.
Credits:
-Character (Jazza / Jasmine): Original character by @FreijaFoxy
@FreijaFoxy holds the rights to own, sell, and redistribute this LoRA
-LoRA training: @CitronLegacy
Description
Main Trigger Word: J@zz@OC
Helpful Trigger Words: furry antho, leopard girl, blue eyes, glitter eyeshadow, sideswept white hair, purple tips, tail, digitigrade, masterpiece
Currently pretty good, hands and feet still need some work and the tail is often too long, but the hair and body are very good. Randomly has cheetah-style stripe on muzzle or not, I do like that better, so future versions will try to stabilize that more. I'm releasing it now "mostly" finished, so others can enjoy it while I work on different things before coming back for an update. Not perfect yet, but I hope you have fun with it!
FAQ
Comments (10)
So I am tired and unfortunately I couldn't make it work with my default Illustrious checkpoint (hair looked wrong the whole time), so I had to dig out a checkpoint I haven't used for some time. That worked nice, but the images are a bit unispired since I am borked^^
Sorry...
Congrats on making a nice LoRA.
They look wonderful! And unfortunately, yeah, checkpoints without strong anthro awareness just don't do well for these characters, even with a LoRA. I have seen several Foxy images made with realism or some anime checkpoints that just didn't get it right. Although my one friend has told me to try making loras using SDXL base and work harder to make them universal, I tend to make and use them with my preferred Plant Milk- Walnut, it gives the look I like best, and sometimes others can be hit-and-miss...
@FreijaFoxy Thanks! Well, it's not like the checkpoint I was using can't do furry characters, but the hairstyle did not look correct. If you did use Plant Milk-Walnut as training checkpoint that would maybe explain it. There is no merge recipe available for it, but it seems like it's a mix of illustrious and NoobAI checkpoints. But still it is strange that the hair style is the thing that is most wrong. I would not recommend to use SDXL 1.0 for training anime/furry characters either, but I would recommend to use illustriousXL_v0.1 for best compatability with illustrious and NoobAI checkpoints (NoobAI is a finetune from illustrious). That's what I use for every illustrios LoRA.
@FreijaFoxy I just trained a LoRA on NoobAI_eps_v1.1 and it works much better with the checkpoint I used for the Jazza images I published than the checkpoint I usually use for illustrious. So I guess the incompatibility problem really is that Plant Milk contains NoobAI weights...
Gotta check that further^^
@RisingV fascinating! Thanks for sharing! My knowledge of the internal secrets of checkpoint weights comes down to "well, it's all SDXL in the end, so it should work..." but I tend to only use my favorites and those are what "worked well" with the loras of others and mine that I already have, so there might be something to that...
@FreijaFoxy check message on discord I wrote you ;)
That was worth the wait :) Very clean and crisp, and i loved the unprompted glitter for certain clothing.
In lieu of any lore, i just dressed her up a bit, tried a bit of cosplay, then because i can't ever stop myself from doing it, i experimented with some tags for way too long.
I added 'whiskers' for a couple of images to see how they'd look, then i had another go at the tail (It's always the tail with furry OC's!)
Nothing i tried in the positive seemed to help but then i tried 'long tail' in the negative, It didn't override the baked in length but seemed to stop it from getting longer randomly (or it was a placebo effect :P) I'm curious if anyone else would have the same result using it.
Thanks once again for another cute OC, i love her! (｡♥‿♥｡)
You're right on this one the tail randomly likes to be unusually long, I have no idea why! But for the eventual update I will make sure to use only training images with a smaller tail and hopefully it learns that. The glitter is a little baked in, but yeah, I don't think unwelcome! I will try to do a short lore soon, sorry! Thank you for trying her LoRA out!! 💖
@FreijaFoxy Don't mention it, i love trying out cool OC's, makes a nice change of pace! I did try a lot of generations with 'long tail' in the negative and it did seem to stabilize it somewhat, only posted the one using it though (the beach shot) but it worked like a charm for me at least so it might be worth a try. I'll definitely be back for another round when you post some lore though :) Thanks for the tips btw, you're too kind! ≽^•⩊•^≼ ❤