Created SDXL LoRA and faces with long shots Camera Distance

[Mosfett1975]

I can't understand the reason - I create SDXL LoRa, at close-up distances from the camera I get an acceptable result, the face being given looks like what was used in training. But if I apply the keyword "full body", the result is a completely different face. The situation is better with SD1.5 LoRa, but also far from what was expected.