trained on 110 real life images of 2+ people in the same image.
Flux version improves focus a lot IMO with activated lora.