A model for hyper pregnant anime or semi realistic characters. This model was very difficult to train compared to my others, so expect plenty of weirdness.
Hyperfusion is the successor to this model, I suggest giving it a try. This hyperpreg model is pretty old now anyway.
V4 Info:
v4 now uses LoRA! A1111 now supports LoRA models with the latest version Jan/2023.
What changed since v3? Aside from LoRA, You get better control over belly sizes with the tags provided in the model details below. I also included a wider variety of belly sizes in the dataset.
Most examples use AbyssOrange as the base model.
Tips:
"hyperpreg, (big belly) is a good keyphrase to start with
If the results are too large for you, lower the LoRA strength, or remove "big" or "huge" from the prompt
Description
A1111 now has support for LoRA models!
Unlike my previous models that used the trigger word hyperpreg this one allows you to control the size with one of the tags listed above in addition to the main keyword hyperpreg. You can also use outie navel to add a bigger belly button most of the time
Training Details:
2.3k images
LR 1e-4
batch 2
GA 8
dim 64
base model Av3
Av3 VAE
flip aug
clip skip 2
Only Unet training
bucketing at 768
trained with captions
FAQ
Comments (6)
Hello, do you know how I can use the Lora file in google colab?
Never tried with colab personally, but Im pretty sure some other people have been running LoRA in colab. model just goes in the ./models/lora folder for A1111
@throwawayjm thank you 🍻
Thank you for the great model! One question: Perhaps I am doing something wrong, but how do I put a girl on the belly? It doesn't work no matter what I try. How many images in the original image set are tagged as "laying on belly" or "laying on top" or something like this? Also it's pretty hard to get an image from the back.
StableDiffusion has difficulties with laying positions to begin with, but I don't think any of this dataset was "lying on belly". Hyperfusion has a handfull of that image type, but I think it is still going to be hard to get since the total image count is low.
As for from the back shots, hyperfusion will be better at it for sure.
I'm sorry If you are wondering how I made that picture in the gallery I generated the belly and body separately. I then put them together and used inpaint to get the belly look. I used skin texture loras to fix the skin quality and bam... belly bed photo. This lora is great for keeping hyper belly shapes even if it cant generate them in that specific way. It doesn't work with everything ofc, I tried making a belly shot from behind with the belly touching the ground the same way I made this one. That didn't work at all.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.