This lora is yet far from being perfect. To see which checkpoints and settings I used, you can load the pictures in ComfyUI.
# Training
Changed training tool
Changes to dataset (mostly clipping)
# Output
Generally better resemblence
# Samples
Changes to prompts and models