v1.0:

• Same dataset trained on linear rank 16 with 512+768+1024 resolution instead of rank 32 with 768+1024.

• Don´t ask me how but the image quality is a lot better compared to v0.5.

v0.5:

• No special trigger word required.

  • You could simply describe a woman withouth mentioning her figure and then try different descriptions from there.

  • I used "plus-size" in my example images which worked quite well.

• Tested with the regular Krea 2 Turbo checkpoint at a strength of 1. Additional NSFW LoRAs might be required for images involving sex.

• Fully original work.

• Check out my other BBW LoRA for slightly less curvy women: https://civarchive.com/models/2932508/krea2-voluptuous-bbw

• v0.5 for now. I`m trying aother training run with different settings. This version adds a bit too much/unwanted skin detail sometimes.