It is trained on RealisticVision5 so prefer that
Steps: 40
Weight: 0.6-0.9
Generate 512x512 and upscale using highres fix
For real faces use img2img for upscaling instead of highres fix
Check images for prompt and avoid making your prompt too complex
