A generalized furry Anima diffusion model with more of the popular glossy ai-gen appearance. E621 concepts and natural language compatible prompting with text generation preserved.
WEAKLY recognized artist tags:
@accelo, @aycee, @bagelcollie, @baphypaws, @blitzpony, @braeburned, @budoti, @butterchalk, @chunie, @cobaltsynapse, @coffeesoda, @cometdragonite, @cytricacid, @dacad, @daftpatriot, @dagasi, @dingbat, @doodlemark, @edjit, @enginetrap, @fantine, @funkybun, @hinar miler, @hioshiru, @indexer, @jumpy jackal, @kashmere, @keffotin, @kenket, @keovi, @k1ngl30n, @letodoesart, @loimu, @mawmain, @meesh, @orlandofox, @reign-2004, @ruaidri, @sushiamazing, @tzarvolver, @waspsalad, @zackary911, @zaush
Additional artist tags recognized in version 2 and 3:
@cubi, @flipyart, @kabscorner, @latrans, @nopetrol, @panken, @zarckets
Additional artist tags recognized in version 4 and 4 10-step:
@allorexx, @jinxit, @leoncheetah, @mintyderg, @nicknak044, @sinz0ne, @tazara
Generation params: Sampler: euler Scheduler: sgm-uniform CFG: 4.5 Steps: 30-50 (38 recommended)
Description
TURBO IS ALREADY BAKED IN!!! It is NOT recommended to use the Anima turbo LoRA with this model!
The 10-step version of version 4. Turbo seems to really shift the style of Anima to a crisp flat-colored rendering style, and it was easily amplified by this merge. I've tried my best to make a 10-step version that doesn't have such a drastic style shift. This is a re-merge based mostly on Anima base 1.0. plus additional training on a 2600-image curated e6 dataset. Also included is the Furry Domain lora for additional species and e6 concepts. A sprinkling of WAI-ANIMA, a negative weight application of Nova Anime AM (removes the anime brain and the glossy slop-effect) and a tiny little touch of Photanima 2.0 (after removing the turbo step). After each step of LoRA application, I merge 20% of base Anima back into the mix to keep the balance and to prevent the model from exploding.
This version knows an additional 7 artist tags: @allorexx, @jinxit, @leoncheetah, @mintyderg, @nicknak044, @sinz0ne, @tazara
Recommended generation params for this version: Steps: 10, CFG: 1, Sampler: ER-SDE, Scheduler: Beta
Also recommend using the added workflow with the CLIPNegPip and Auraflow nodes.