A model for large breasted waifus or semi-realistic characters.
V6 Changelog 2023/06/03:
Considering this was my first and most popular LoRA, I figured it was time to improve on it. Probably the last version too. Hyperfusion is more or less the successor to this.
Less harmful to the overall style, but slightly smaller sizes overall, feels like a good balance
Wider variety of breast shapes in the dataset
Some breast shape tags were added, but hard to say how helpful they are
Uses LoRA LoCon, but does not require any additional extensions to run
Changed "large breasts" to "big breasts" to match my hyperfusion model tags
The model prefers resolutions around 640, but can do up to 768 to a degree
Training tags attached under model download section
V5 Changelog:
v5 now uses LoRA!
What changed since v4? I trained a breast size classifier to auto label the dataset which resulted in better control over breast size, and made it easier to generate larger sizes.
Note that the trigger words in v5 are ordered in the following from smallest size to largest:
breasts, large breasts, huge breasts, gigantic breasts
V4 Info:
the keyword for v4 remains hyperbreasts, if you want better control over the size use v5
Notes:
I used this to train my image tagging classifiers for breast sizes
https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification
Description
Dreambooth model. Cleaner dataset, more images, longer training time at lower learning rate than before. Better quality overall.
Training details:
1.6k images
learning rate 7e-7
steps 30k
batch size 2
transforms: left right flip
No class or class images used
FAQ
Comments (18)
Nice! Do you have an email are discord?
Yes, that would be very helpful. I get only black images with this model and can't use it :/ But V4 works like always.
Adding "--no-half-vae" to command line args usually helps with black images. Otherwise your best resource would be searching here for similar issues https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues?q=is%3Aissue+black+image
I probably won't be able to help much on my own, but feel free to ask questions here
Hey @throwawayjm! I already tried different command line args but nothing helped. Just reinstalled SD, added lora and did all the needed steps but now i get no image. Only this line: "NansException: A tensor with all NaNs was produced in Unet." I'm also not sure if i'm using your model with lora right. I never used that. I really really like your v4 model but v5 with lora doesn't work. Maybe you have a solution.
Unfortunately no, it seems to be working for everyone else, so it's either the LoRA extension or some other local difference. You could try to roll back the LoRA extension version to a previous commit. But which, I'm not sure.
Great model hope to see more
Great model. There is a safetensor version?
Also, just curiosity, how did you get a dataset to train it?
Hey, thanks for trying it out. I'm far too lazy for safetensor right now. I usually rely on python's "pickle scan" package if I want to scan pickled files before using them. Ill consider it for future versions at least.
The dataset was just cobbled together images from multiple sources, r-34, e621, danbooru, etc
Thanks for the answer!
Expecting future versions.
Added benefit of SafeTensors is they can be loaded much faster, so definitely would look forward to that.
Amazing model, just wondering how many images did you have in your dataset?
Also are you considering training a LoRA?
This was around 1.6k images. And funny you mention it I just trained a LoRA version of it today. Trying to decide if its good enough to upload yet.
Glad your working on it, i'm really interested to see what a LoRA does when paired with other models.
Really appricate the reply, ur a legend.
question how do i pair lora model ?
Pairing, or mixing, is just how LoRA works. You apply LoRA to any stable diffusion model you want at runtime with the LoRA extension for Automatic1111. Otherwise you have to merge LoRA with another model via command line the manual way.
Just wondering, if I want to merge with my current model using add difference, which ratio should I use?
Hey, try with alpha value of 1 to start. If it looks weird or blown out, then reduce it. Usually between 0.5 and 1 works. Occasionally even larger values too. Just depends on the model you are adding it to.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.