Pam Beesly (Jenna Fischer): v2.0_portrait: pambeesly_lora_v2.safetensors

more training, turned out better for portraits but less versatile

Latest version is a complete overhaul. wd tagging, better image selection.

Trained on picXreal, but it works on many modern models.

Keep prompts simple. Start with positive prompt as "pmbsly, photograph, ..." and negative as ONLY: "bad proportions" it's a bit sensitive to too much negative prompting.