So a bit more information this time:

First of all, trained on sd 1.5, around 1500 steps. You can use it on any checkpoint related to 1.5, works fine with the less realistic ones too.

Second of all, I recommend not using the restore faces option with this lora, the results are (imo) a bit better without it. I think 4 of the 5 images I added are without restore faces. Values between 0.4 and 0.7.

Third of all, this was a bit of a pain to get quite right, it is still not perfect (rigid poses, some skin inconsistencies) and I will return to this later, but part of it can be fixed with prompts/inpaint/img2img.

Fourth and last, if you like it, would appreciate a like :-)