A lora that aims to improve overall people representation, including face / skin / hands / color correction and face diversity.

Trained on 120 high-quality images, with diverse male and female subjects in different photographic styles (magazine images, portraits, selfies, editorial photos etc.)

Any strength works, with less effect on less strength

In addition to these improvements , strength of 0.6+ seems to address the prevalent hi-dream JPEG artifacts that sometimes can come up with this model.

Please consider tipping if you like the Lora, training for hi-dream costs quite a lot in GPU rent time :)