TL;DR: A western style with some latent Niji aesthetic, no trigger words 💜
Following on an old experiment done with Un-niji-fy, i like to try adding a "Niji" LoRA with a negative strength to "Niji" checkpoints to discover new styles.
This is exactly what happened here with:
An example result here: https://civarchive.com/images/136164440 (this felt like it looked similar to Disco Elysium artworks a bit)
Now, with this, i made around 200 pictures (i'll share the dataset, some pictures are too pretty to stay in the shadow) and built a new style LoRA to try and do my own take on a style less "anime" i am less familiar with.
I'll probably incorporate it later in my own models. But as it is looking nice as a standalone, here it is 😉
NB: Beware, i did not tag "everything" when doing the captions and since i didn't add a trigger word, the LoRA may add unexpected elements to the result* (but that's the way i like it, and i know this kind of surprises is what made HoJ fun too)
*) When tagging, my current understanding is that the training will group everything not tagged in the picture under the provided tags. With a trigger word, it is easy, it "goes" in there. Without it, it creates a bias where non tagged elements may appear together with tagged elements regardless of what you actually prompted. It may be "a mistake", but i like surprises 😁
.
.
.
.
.
.
.
.
.
Wow, somebody read until here? 💜💜
Description
Posted on Civitai, upload found elsewhere are FORBIDDEN and done without the author permission.
FAQ
Comments (3)
It looks amazing, I can't wait to try it, thanks!
Hope you'll like it, for now i have tested it only with Base Anima and my own AIO Turbo, i don't know how it will react to other finetuned Anima model 😅
@n_Arno Don't worry, I'll clear things up for you. I like to run lots of different tests with various checkpoints. Anyway, if it doesn't work any other way, and only works this way with the Base version, that's fine too; it still has a fantastic design.