@NeuralVandal im just doing some more tests concerning facial consistancy..... therefore I would love to know: the Bea trainset picture dimensions - 1:1 aspect ratio? Or mixed?

@NeuralVandal I suspected this cause my 3:4 renders do not have a stretch related lack of facial consitancy with the bea lora. If I talk with ki about krea2, it always points out the 1:1 matrix, and that is will bring doom on me if i train loras not following that. I mostly render in 3:4 so I would also prefer training loras on that aspect ratio. If you get this rediclious good results without 1:1 trainsets, I will give that a try. Thx for the info.