kontext lora to make breasts and butts slightly bigger you can run the output in again if it is not big enough or play around with the lora weights
utilize the prompt:
make the woman's breasts larger and her hips wider
trained on a dataset of 50 pairs of images from VAM in which the same pose + clothing + camera angle is used on 2 different morphs one with small breasts and ass one with big breasts and ass, so far testing on realistic images works well. Males were included in the training data, in which they are not altered, however it seems that the model is sometimes altering male chests and butts. Works for anime as well. Works for both nude and and clothed bodies. In general it handles complex poses pretty well as the dataset used a good variety of poses.
link to full write up on how this was trained and learnings from this experiment:
https://www.reddit.com/r/StableDiffusion/comments/1m7qd8y/flux_kontext_lora_sliders/
edit:
after further testing it seems that the adjustment is a bit small, thus requiring too many loops of refeeding the output into the input to get truly massive breasts. Feeding the output into the input starts to generate noticeable degradation in quality which starts becoming very noticeable 3 iterations in. The dataset breast and butt enlargement is about twice as significant as the lora results so for future trainings I will keep in mind to exaggerate the before and after in training data
Description
FAQ
Comments (16)
can someone tell me how to share a dataset on civitai
You can put your dataset in a zip file and upload it along with the Lora safetensors you uploaded.
Where your Lora safetensors file show, click 'Manage Files', upload zip.
can i use this lora with negative value to get opposite results ? , you trained this on ai generated photos .. i think better to use real photos and the pair of each is to use hand edit one by one.
no it does not work i tried it i would have to retrain with start and end switched in order to make boobs and butts smaller. Hand editing is much slower than using vam I was able to get 50 pairs in less than 1 hour, if I were cracked and good at doing it i think i can build a smaller dataset in 15 minutes. Also VaM is computer generated graphics not ai. It doesn't matter the source style for flux when training a concept like this, believe it or not the "anime clothes removal" lora that was posted earlier works perfectly fine with real characters
ultraautism Yes, so long as you hit the correct conceptual plane within the model through proper data labelling and counter-examples (i.e. data that has Nothing To Do with what's being trained, the 'don't change any of this' data), then strong models tend to generalize very well.
ultraautism I read this:
"The dataset I created included 21 pairs of images where both a woman and a man are both featured. While the woman received differences in body shape in the start and end the man did not. The prompt is also trained as “make the woman's breasts larger and her hips wider” which means the flux kontext transformation should only affect the woman but in many generations it affected the man as well. Maybe the flux kontext text encoder is not very smart."
And it makes me think that you probably need duplicate images (i.e. "no change") of just dudes in various scenarios, some with the phrase, and some without the trigger phrase.
If you're "cooking an image", then krita-ai-diffusion tends to do a fantastic job as an image editor + diffuser implementation.
However, I can point out - that there's nothing stopping you from using Kontext 'modifiers' like this on an image, possibly with a few loops, and then returning the image to be mildly re-cooked in the model that originally generated the image.
As this accomplishes the same thing as doing manual edits in something like Krita.
So the more FUNNY SLIDERS, the merrier!
not work on nunchaku
Try this version: https://civitai.com/models/1807009/kontext-voluptuous
thanx, working good (in Wan2GP by deepbeepmeep)
@ultraautism Can you please make a version that makes only breasts larger but not hips/butts?
u can use this prompt, works well for me.
make the woman's breasts larger and her body slim
can this do boobs only and not butts? i have a few images that are keeping the female shape right but when changing clothing kontext is making them practically flat chested, i just need a way to restore the bustline alone.
use mask
@dabaiguzi392 masks and kontext dont really seem to get along all that well, is all.
You can create one that focuses on the legs, thighs, glutes
Details
Files
kontext_big_breasts_and_butts.safetensors
Mirrors
ultraautism_kontext_big_breasts_and_butts.safetensors
kontext_big_breasts_and_butts.safetensors
kontext_big_breasts_and_butts.safetensors
bbb_kontext.safetensors
kontext_big_breasts_and_butts.safetensors
kontext_big_breasts_and_butts.safetensors
kbbab.safetensors
kontext_big_breasts_and_butts.safetensors
kontext_big_breasts_and_butts.safetensors
kontext_big_breasts_and_butts.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.