(IF YOU NOTICE YOUR PROMPTS IN MY COVERS DM ME FOR COMPENSATION)
NAME HAD TO BE CHANGED DUE TO RUDE/NEGATIVE COMMENTS FOR ATTEMPTING TO MAKE MY OWN FINETUNE OF EASY FLUFF.
Banana Kiwi XL
A combo of the cartoonish looks from Kiwi Mix, Oops! All Furries + Oops All Toons!
These models were a huge help in guiding this models aesthetic!
Small prompts produce easy masterpieces
NOTE: If you want muscular males make sure to consider using a few of these tags
masculine, muscular, manly, bara, athletic, Description
Models in Mix for those who are interested
Oops! All Toons.
Oops! All Furries
Kiwi Mix
Caps Merge
WAI-Illustrious-NSFW
Banana Splitz V5
Fabulosity V2 (Lora)
ToonCity (Lora)
Rule 34 Dataset V2 (Lora)
FAQ
Comments (16)
Praying that prompt adherence is amazingggg
It seems to be pretty solid. small, vague prompts seem to get me what I want easily.
If I had to give you an honest estimate, I would say this had a 8/10 for prompt adherence.
maybe a 7.... lol
@LatteLeopard Thanks!
Also, I personally think you'd be able to make a really good transformation LoRA or something. It's probably not your thing but being a TF guy myself, the prompt adherence in your models have been pretty good.... man... I wish. But thanks for responding!
@Feem I'm a bit confused what do you mean, a transformation Lora?
I could convert models to Loras but it wouldnt have the full weights.
@LatteLeopard Yeah, a Transformation/TF LoRA. Like this one for example: https://civitai.com/models/174742/transformation-furry
If you don't feel like doing one I understand, I feel like I'm pressuring you lol
@Feem oh wow I cant believe I didnt realize you meant transforming like that.
tbh I dont know if I could do better than that lora you provided. try messing around with the weights of the lora. For more percise control, try a full block weight,
you wiegh lora beyond just like this "0.4"
you can do this. "0.1,0.5,1,0,0,0,0.2,0,0.4,0.2......." up to 20 blocks
each one of those corrosponds to this
"BASE, IN00, IN01, IN02, IN03, IN04, IN05, IN06, IN07, IN08, MID00, OUT00, OUT01, OUT02, OUT03, OUT04, OUT05, OUT06, OUT07, OUT08"
Overall analysis:
BASE : text encoder
IN00 - IN05 : handles anatomy and details (very basic outline of anatomy)
IN06 - IN08 : handles tone and poses
MID00 : how strongly the lora and model will mix together.
OUT00 - OUT03 : handles shape of face and body (this is were you can fix eyes and hands)
OUT04 - OUT08 : handles textures
Overall analysis:
BASE : text encoder
IN00 - IN05 : handles anatomy and details
IN06 - IN08 : handles tone and poses
MID00 : over all similarity
OUT00 - OUT03 : handles shape of face and body
OUT04 - OUT08 : handles textures
this is known as the UNET blocks, the shape is litearlly in a U
so the middle is the the most potent. the end points of the U are the least potent.
It is not the kind of EasyFluff :/
The real EasyFluff was somehow special, both SD1 and the original Illustious old mixes
You dare call it EasyFluff? You already have (y)iffyMix
Yeah I did my own indigo furry mix too. Don't forget that.
Sorry for trying to finetune models.
Instead of just making complaints about my audacity to call it easy fluff. What exactly is wrong with it?
How can it be adjusted to feel like what you are attempting to describe.
I LOVE IT!!! ❤️😍
Love this model, but a ton of anthro-centric prompts are getting blocked as 'bestiality', any idea why that is?
Ugh, that has to do with civitai's generator, auto-detect feature they implement a month or two ago.
I believe it gives you an option to mark your blocked content as mistakenly blocked. Doing this actually does help. This is because the intensity of the auto-detect feature fluctuates day to day as it learns.
There is also a method that was created by civitai that helps prove you aren't prompting anything illegal and ALSO removes the chances of anything illegal seeping in.
Use these two embeddings
put safe_neg in the negative
and safe_pos in the positive
People discovered these secret embeddings, becuase when doing img2img after a successful txt2img generation they noticed those embeddings were mysteriously added.
Heres some sources that help explain the auto-detector and the embeddings
Introduction of the auto-detect system - quick explination, better than I can explain
https://civitai.com/articles/11072
@LatteLeopard Got it, thanks for the heads up! :)
Had to change the name of the model, because a few, but loud mouthed users, created trouble for me.
Do you know why we cant use this checkpoint? I tried once but bow is not possible
@Styloo So, I have place a bid for my models to be used online. Theres currently a drop in the number of spots. So that I dont hog all the spots. I am hosting my top 3 models + 1 random model each week.
Things are getting betters. It had went from 200 spots down to 100. but now its back up to 150.
You should see this available maybe next week.
This is so amazing, I'm glad it's back for online generation.