Update Oct 14:
Happy to release a brand new version!
This new version was rebuilt from the ground up using a diverse dataset, that was used in my Qwen breast fix LoRA as well.
I find it to work very well in videos and images, trigger word usage can be a bit hit or miss but it gets the job done.
Dataset was trained with the following key words:
tiny/small/medium/large sized areoles
tiny/small/medium/large sized breasts
pale/brown/dark/ghost areoles
hard/erect nipples
Join my Discord server for updates on new LoRAs, tips & tricks, Workflows, RunPod templates and the holy right to be close to an amazing charismatic person like myself:
https://discord.gg/fyha5Pzm
Version 0.1:
A quick LoRA I trained on 28 images of boobs generated with SDXL.
I tried to keep it diverse as possible with this little dataset but I found this to be quite flexible so decided to post and I might create an expanded V2 in the future.
Trained on an H100 using my RunPod Diffusion Pipe Template
Description
FAQ
Comments (10)
YOOOO my most anticipated release of 2.2, the 2.1 version still works amazingly well in 2.2 but I can't wait to try this
Best thing about V1 was how it didn't change the face at all, which some other nude body Loras do. Looking forward to trying this.
an areoles is a type of cactus
If I use this for I2V and leave out the trigger words, will it help the overall quality/accuracy? Or is this only useful for deliberately dictating what kind of breasts to make?
Am I right that the newest version is very much T2V only? I've only tried it in a couple for I2V, but it fried the image, a lot of distortion, ghosting, etc (Which I've had happen with other T2V LoRAs in I2V.)
I get blurry nipples a lot. Other than that, it's working great. Any idea why I get blurry nipples? Does resolution of the generated video matter? I usually generate videos with 496px as the longest side.
Makes consistently attractive and well proportioned breasts, which is awesome for making idealized women.
If you are trying to make different looking breasts, I've found cranking the weight on the LoRA helps, as well as putting additional weight on what you're looking for. To create itty bitty titties with large areolas I went with 2.0 on the LoRA weight and 'large areolas, (tiny breasts:1.5)' to make it happen.
The new version does not work I'm sorry to have to say. It has a few major issues.
1. It constantly forces the view downward or simply starts with the mouth and below visible. It is very difficult even at lower strengths to get it to show above the mouth, and again if you do it always drifts down without exception until head is out of view. I'm guessing this is due to it being trained without fully visible heads.
2. It has a really extreme identity bias trying to shift women to look like a specific person.
Unfortunately, even when lowering the strengths I found this genuinely unusable. Sucks to say since training takes time/resources, but it is what it is.
Please add more breasts of different breast types, such as expanded chest, spindle chest, narrow spacing chest, wide spacing chest, small chest, etc., as well as breasts in different states, such as soft and inelastic breasts, whether there are chest moles, etc
Please Add Natural Indian Nipples
Dark Indian Nipples
Details
Files
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
Mirrors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors
BreastHelper_HIGH.safetensors
BreastsLoRA_ByHearmemanAI_HighNoise-000070.safetensors