V3.0 - Improved data set and captioning.

First slider LoRA (3rd LoRA overall) — still learning, but pretty happy with this one.

I made this mainly because most body weight LoRAs for Flux tend to mess with faces or clash with character LoRAs. This one is designed to avoid that.

What I did (dataset):

• Cropped all faces out of the training images → helps preserve character LoRAs

• All subjects on plain gray backgrounds → doesn’t seem to interfere with scenes

• Tested with multiple LoRAs → plays nicely, especially with character + breast size LoRAs

• All adult women in swimwear

How it behaves:

• Trained toward the low-noise side, so it’s strong even at low weights

• Higher weight = thinner body

• Sweet spot: -3 to +3

• Outside that range = gets extreme (very thin ↔ very large), around +/-7 it melts down

• Can still generate nudes (likely due to captioning used)

Prompting tips:

• Adds to the effect of the weight

• Use words like: slim, skinny, thick to steer results further

If you test it, I’d love to hear how it behaves with your setups 👍

All training images were AI generated, no images of real humans were used (living, or deceased).