This is a lora that focuses on bottomheavy body shapes. It is trained on 100% anime/furry data, but can be made to produce realistic images just fine. You can adjust the body size with lora strength as needed (it should handle smaller sizes as well).
See Training Data for tags list
Changelog 2026-01:
Just decided to revive an old lora as a test for Flux2 Klein-4b-base. I'm starting to prefer Klein over Z-image, but ill wait and see what gets better adoption.
Description
This is a test of SDXL's ability to learn far fetched concepts. This is not up to par with most of my other models, but still wanted to show off what SDXL was capable of learning.
Training Details:
~400 images
14 epocs
batch 2
gradient accumulation 4
learning rate 2e-4
No text encoder
base model SDXL
clip skip 1
random flip
shuffle captions
tag drop chance 0.1
network dropout 0.4
bucketing at 1024
dim 16
alpha 8
conv_dim 8
conv_alpha 4
cosine with restarts
scale weight norm 1
FAQ
Comments (26)
Great stuff as always.
Thanks!, I guess its easy once you have the data all ready to go.
Edit: also thanks for the comparisons even though this one is a bit under trained. Good to see visually
cant wait for the hyperpreg sdxl.
lol, as I'm reading this, I'm on epoc 2 of hyperpreg SDXL. More of a test than a real run though. Who knows if it will be good enough yet.
@throwawayjm I don't have a good enough graphics card for sdxl anyway, but I plan to get one next month. I need to upgrade for starfield too.
@throwawayjm Is there any update on hyperpreg sdxl? I dream of the day it comes out.
@BBBenthusiast It didn't turn out well, and have tried multiple times now. Kinda gave up on SDXL for now. Focusing on improving hyperfusion on SD1.5 instead.
For when Hyper Pussy SDXL?
Not really my thing, but It only takes ~200 images to make a decent model, and apparently as little as 12 GB to train it
@throwawayjm I hope the 3060 12gb that I want to get is going to be good enough. But I actually would love either an unbirthing, or monster pussy/cloaca model for xl, like the existing ones for 1.5. that's just a suggestion, not a demand.
@Taco360 Yea, 12 should be enough according to some people. Just use similar config I put on this page under model details, and give it a try. You can always test it on SD1 first too. A 200 image model takes less than 2 hours to train.
Now we wait for hyperbreasts sdxl
imagine the size of those WATERMELONS....
I did try training hyperbreasts on SDXL, but it was disappointing. Not sure why since both bottomheavy and hyperbreasts were both trained with the same config.
Probably going to wait on a danbooru type finetune to release first to train on top of.
EDIT: It works in PonySDXL checkpoints (AutismMix PonyXL too). Generaly, styles/bodychanges/clothing work between SDXL and PonyXL.
Haven't tried yet.
I tested with PonyXL checkpoint, and AutismMix checkpoint based on Pony, and it worked so far (tested more with anime drawings).
I love happy and i am glad this is a thing to helping in making some arts
Really fun to use
I'm still surprised its been so popular for a quick test model when SDXL came out, lol
What happened to the pony version?
I don't think I trained a hyperass lora for pony yet (other than hyperfusion_sdxl). just bubble_butt so far. Maybe you were thinking of another person's lora?
@throwawayjm It was this one. I had always used it with Pony models, and suddenly it's not available anymore, but when I click on the model from prior made images, it takes me to this one. Not sure what happened.
Can we get this for QWEN?
Haven't touched the Qwen models. Hoping Z-image will be the new one people will focus on.
Amazing! <3
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.