This model works correctly only with config file. Place this file in the same folder as this model.
Fluffy Universe (old name Furry DreamShape) is my personal staple mix that I update to suit my needs, and I do LORA training on it as well.
This model merged with such well-known models as YiffyMix, DreamShape, BB95, Koji, epic realism and the core of this model is the best model today in my opinion - FluffyRock
As for samplers, I most often use the most powerful "Restart". I experiment a little less often with various subtypes of DPM++ 3M SDE.
Tag Autocomplete File - courtesy of FluffyRock's tag database.
This model is universal and suitable for generating a variety of objects, including realism.
I express special gratitude to each author of the listed models present in my mix.
LORA training on models with the FluffyRock base includes the following instructions:
Must be activated "parameterization v". This setting is located in the tab when selecting the base model.
The program may display an error that this model does not support this function. Ignore the warningit is advisable to use the parameter "noise_offset" around 0.1-0.9.
Reminder
Model 1.0 is my old mix without FluffyRock. However, this mix is useful when working with common lore.
LoRAs trained on non-vpred FR models may perform significantly worse on newer versions my mix.
Description
FAQ
Comments (10)
I love more of these still come through and wonder I hope there will be a yiffymix for SDXL to be trained in order for copyrighted characters to be trained with images more accurate just like 1.5 does. heh
when pruned version?
Time to summon fp16 guy
Very great. :) I like the results I am getting so far, I just download the 2.0 version. Should I deleting the 1.0 after getting 2.0? 😏
Overall yes, 2.0 is more expansive and improved at the expense of FluffyRock.
@deygdrixe221 Thank you. 💕 That will saving some space.
@Mackya I am writing from the future, having gained experience. And yes, it’s too early to throw Mix 1.0 into the trash. Its distinguishing feature is that it works much better with most Civitai lora. While new mixes work well only with vpred lore.
@deygdrixe221 Very good, I will keep both on my HDD then. :3 Though I am excited to try the latest Fluffy Universe too. :3
Is this able to generate characters natively like in your examples or would a LoRA be needed to achieve the best results.
This model has a decent character base in general. You can familiarize yourself with this base on the FluffyRock page with their tag table. And in general, everything is very subjective and depends on the quality of LORA itself. Often the opposite, that with LORA the quality of certain characters will be even worse than in the basic model.