There’s a few variations in prompt weighting, but mostly like so (this is the one with flaming eyes, but reproduction is not guaranteed with SDXL beta):

  "prompt": "Shape of a skull, Love hate mind beauty ugliness heart pain mystery magic fantasy color detail desire vanity cruelty dark theme hot cold scorching frozen earth air fire water passionate apathy bright (neon:0.6) darkness, intricate plotting threads of despair, broken holes, heaven puddles arterial drops drips ### (nsfw, topless, bottomless, words letters name title caption watermark credit website, Asian, nsfw:0.6), (2d flat anime cartoon drawing, black and white, b&w, greyscale, sepia:0.4)",
  "params": {
    "cfg_scale": 9,
    "seed": "14655602",
    "sampler_name": "k_euler_a",
    "height": 1280,
    "width": 1280,
    "steps": 20,
    "karras": true,
    "hires_fix": false,
    "clip_skip": 1,
  },
  "models": [
    "SDXL_beta::stability.ai#6901"