Sorry for the delay! Versions 1 and 2 were trained on 512x512. Do you have some example prompts that are resulting in multiple subjects? There was generally only one woman in most of the training images.

This has no impact on the output, but 2.8=2.50, it's just a display thing. I realized that this is related to the file metadata. You can configure Automatic1111 to display either the actual file name or the name of the output file that was specified at the time of training. When I was training it, I called the output file 2.50 (version 2, running for 50 epochs), but then I selected the best models from that training run and compared them with extensive testing, ultimately settling on the 8th one. So the file was named 2.08 and that's what I called the version in CivitAI, but there's still some metadata in the file calling it 2.50 because that's what it was called during training.