The short and sweat: a furry-centric model merge built around e621 tags.
The long and hard: This is a checkpoint merge of the models HomoSimile Anima + Indigo Void Furry Anima and a custom LoRA consisting of 5473 images scraped from popular e621 posts. The dataset features top-rated posts from every month over the past six years, spanning a wide range of species with a strong emphasis on accurate furry anatomy. Hence the working title of this LoRA was called The Anatomy Problem. I've kept the e621 tags intact for this dataset and as such you'll want to prompt like it's Pony or Illustrious when using this checkpoint. Enjoy!
Recommended Generation Parameters
Sampler: ER SDE
Scheduler: Beta
CFG Scale: 4
Steps: 30–40 (32 recommended)
There are 921~ different artists within this datasetListed in Alphabetical Order of Artists With 5+ Images:
@anchee, @anhes, @arf-fra, @ashraely, @autumn sunrise, @backsash, @blotch, @bonnie bovine, @braeburned, @butterchalk, @catniplewds, @chunie, @clockhands, @clyde wolf, @cobaltsynapse, @codyblue-731, @cooliehigh, @cuteconfy, @daftpatriot, @dagasi, @delki, @dimwitdog, @edjit, @elicitie, @fenn.perrox, @feretta, @fluff-kevlar, @foxinuhhbox, @gewitter, @goonie-san, @hooves-art, @hyattlen, @imprettynormai, @inno-sjoa, @inu-sama, @iztli, @jarlium, @jishinu, @juiceps, @jumpy jackal, @kakhao, @keadanger, @kenket, @kilver, @kraaisha, @lazysnout, @letodoesart, @lichfang, @lockworkorange, @loimu, @lvlirror, @magnetus, @malaika4, @manmosu marimo, @marsccts, @nepentz, @notafurrytho, @oxfort2199, @patto, @personalami, @phenyanyanya, @r3drunner, @reallynxgirl, @ruaidri, @sagaris uwu, @scarf (artist), @scarlet-frost, @schmutzo, @scruffythedeer, @seibear, @sepulte, @seraziel, @sicmop, @sijimmy456, @singafurian, @snowskau, @spaal, @spuydjeks, @ssirrus, @ssssnowy, @swaybat, @the giant hamster, @thesecretcave, @tofu froth, @trmk2, @truegrave9, @twang, @unsavoryimpcat, @waspsalad, @watsup, @whyvern, @wizzikt, @wolfblade, @woolrool, @wugi, @zackary911, @zaush, @zeforge, @zica*(While these artists appear in the training data, their stylistic impact on this checkpoint may vary significantly)
Description
FAQ
Comments (7)
Absolute game changer. Thanks so much for this!
Has a lot of issues with andromorph/gynomorph body types. Certainly a step up from using base anima with the e621 lora.
Absolute belting model. Thanks so much for making!
Are you planning on making any more Animurr fine tunes? I was going to do one myself as the existing ones I found to be subpar, but no point as yours is already doing most of the stuff I wanted.
Absolutely, though I don't want to discourage you from trying. I think there's always room for improvement TBH.
The next version/s should be more style neutral; I'd like to rework the dataset some and add a few concepts from there too.
@Murf_ Sounds good! There's a lot of potential with Anima. I was going to do a complete finetune with 12k e621 images. Is this worth doing compared to a merge + lora like most seem to be?
@Annwyd Up until about a week ago doing a Fine Tune was a bad idea because of licensing issues. However, the Anima team has changed their TOS to allow for them now... It still won't stop Nvidia from coming after creators, but that would be highly unlikely IMO.
The biggest hurdle creators currently are facing has to do with the cost and vram constraints of training a fine tune. I personally don't have the vram to train a fine tune and renting a card for a few days could get expensive.
It is just a matter of time until somebody with some talent does. We'll see.