Trained on 99 images, use flat color and between 0.7 - 1.0 for best results. Most images were genned with a-detailer activated, but I found during tests that a-detailer sometimes makes the face worst. May lose coherence / produce artifacts at higher strengths.