It took me forever to find the right combination of training images. My first few iterations could only generate caricatures. Eventually training on 250+ pics was what worked.

Sample images generated with PopOffMix V9 Ultra Realism v2 and a .6 LoRa strength.