In Norse mythology, Ratatoskr (Old Norse, generally considered to mean "drill-tooth" or "bore-tooth") is a squirrel who runs up and down the world tree Yggdrasil to carry messages between the eagle perched atop Yggdrasil and to the roots of the tree to Nidhogg.

This model is Fine-tuned for Animals/ Creatures and Furries but can do way more :

This model can do:

• Photorealistic

• Digital Art

• Anime

• Comic

• Vector Graphics

• Line art

• Stickers

• can write short text

• etc

  The model needs no refiner! or any other enhanced lora

  SFW and NSFW

Changelog and credits:

V2.0

it includes some loras and other models i made for (Fenrisxl 15.9), Newdawnxl

and other models (see Credits of Gleipnir that was already used in Gleipnir but in other weights i also added following models:

SDXL Yamer's FurWorld

Furry Art Fantasy XL Mix

V3.7:

This Version is more than just an update its an complete overhaul and basically a DB Finetune of my model Asgard

• way more realistic

• more versatile

• overall way better

  Note you need set the CFG and steps a bit more down than in V2.0

  i recommend between 2-MAX !5 CFG and between 15-60 steps

Fp16 will be released in the next days after testing

V3.8

This Version has my Furry Enhancer baked in and some other adjustments

including to make it possible to train loras on it;-)

Please do not use this lora on this Version ;-) or any other detailer/ enhancer it will end in over backed pictures . And if you like to experiment with different strengths and combination of the Lora I can highly recommend to use Ratatoskr 3.7 instead. You may get even better results than with 3.8. .

V3.8 benefit is speed and it's easier to use "out of the box" with an pre adjustment Lora weight. It's recommend V3.8 for new user of SDXL with less experience with using Lora's. Also to save space and adjustment time.

V6.53:

New Fine-tuned Version of Ratatoskr basically a new model. the model has a huge bandwidth from drawing to realistic ;-) Note: this model is not a fast model i recommend at least 30 steps and at least cfg 2.5 and up for the best settings i recommend using the sampler DPM++ SDE or DPM++ 2S a

• more detailed

• more specim

• better nsfw part

V7.0

• faster in generation

• more details

• better nsfw

• more stable

• a bit more realistic

  This version contains parts of the new furry enhancer

  Note; If you are using Invoke please download the fp32 version!

  About this version

  V7.3

  FP32/BF16 Version

  • better contrast

  • add small details

  • even more stable

  V8

  more photorealistic

• more stable

• more species

• better with text

• better nsfw

• etc

I merged parts of Midgard and Midgard Pony in ,also some parts of Asgard

+ some unpublished Lora.

Then I did an Finetune with around 390 Pictures of Output Pictures of my new Flux Models on it.

New SDXL Model Type (SDXL HEL)

V10.94

This is a complete new alternate version of Ratatoskr so don't expect the same output finetuned on the Database of Helheim Flux with over 50k steps !

after that i added small parts of my Furry Enhancer and am small amount of Hayden's YiffMix Realism

i made this model to test new training methods for sdxl also to make a Christmas present for all the people that still love my sdxl models and cant run flux.

for settings its more or less the same as V8 What has changed on this model?

it has gained quality in most areas but the best is is understanding of different prompt languages, it understand my prompt speech, natural language, e621, to some extent pony gibberish, and also that it understand the sd1.5 prompt language but not perfect.

V11.16

I trained the dataset again as a lora and mixed parts of it inside the Model

• added a bit more details

• more stable

• added some new specim

V12.23

optimized version fine-tuned on better detail such for example fur , paws etc overall more photorealistic for comic art i still recommend using V11.16

for settings its the same than version before

V13.8

After 12.23 was focused more in realism this Version also is trained on comic, artwork

trained on partly new database 89K steps on V12.23

for photorealistic stick on V12.23 or use Ratatoskr Illustrious

• creative model

• also still realistic

  for Settings :

  Keep the cfg low max 4 cfg

  20-60 steps

  euler, dpmpp_2s_a, dpmpp_sde tested

  V14.1

  This version V14.1 is more focused again on general use .

  I made some small adjustments in the database by update older captions i made almost 1 year ago for example explain text elements in pictures better.

  added also around 400 new pictures into the database that overall get added for better fur and lights

  • better with text

  • more stable

  • add small details

    Note: overall the picture output won't change much only in small mostly better details.

Z- Image V1.0 (WIP)

This is the First Version (WIP) of Z-Image for an Ratatoskr Model

so why its in the old sdxl model section.... well because there will be no new SDXL Version in the future for this model instead you only will find the Z-Image Version or the Qwen Image Version

so this Version is not build on the Z-Image Turbo version instead its build on the De- Distilled Version De Turbo. by ostris

i choose it because the nature of turbo models distillation in short you can't train a Turbo model in best quality on 110K steps and be aware this is V1 so a lot of improvement its planned for the future

so this version is trained on the same Database of the SDXL Model for V1 its around 50% of the original base rest is a mix of the new Ratatoskr Qwen database overall i just trained 60% of the whole database. so there will sure some updates in the future.

i also build before an Z-Image Turbo Furry Enhancer but why using the model instead of the enhancer with a turbo model?:

• better consisted than Z-image turbo

• overall better quality

Settings:

Sampler: Euler Scheduler Beta57 / Beta

Steps 30

CFG: 3.5

i also provide a simple workflow with the settings for it too ( see training folder)

Note:

this model is its own type of SDXL i called it SDXL HeL

and not always compatible with other loras like standard sdxl or pony etc. So Loras work somtimes but its possible they need other weights or not working at all .

Ratatoskr Workflow and Settings

Made by Freek22

Please check out my other models too and give like ;-)