Fill the rest of the JSON schema (high_level_description, style_description.aesthetics/lighting/color_palette, compositional_deconstruction) to describe the scene.

Example JSON prompt

json

{
  "high_level_description": "A young girl with dark hair in a ponytail sits at a wooden desk, writing in a notebook while wearing headphones labeled 'LOFI'. A sleeping orange tabby cat rests on the windowsill beside her. A red desk lamp illuminates the workspace, with a laptop, books, and stationery nearby.",
  "style_description": {
    "aesthetics": "warm, cozy, detailed, anime-inspired",
    "lighting": "soft, directional light from a red desk lamp and natural light from a large window",
    "art_style": "Studio Ghibli style",
    "medium": "anime_screencap",
    "color_palette": ["#272A33", "#BCBCB5", "#6A707A", "#A0695C", "#2A7A74", "#96223E"]
  },
  "compositional_deconstruction": {
    "background": "A large window with dark frames shows a view of multi-story buildings with red-tiled roofs and white walls. A potted plant with trailing green leaves hangs from the window frame.",
    "elements": [
      {"type": "text", "bbox": [393, 347, 865, 808], "text": "LOFI", "desc": "A young girl with dark brown hair tied in a high ponytail, wearing a green hoodie with a pink hood, writing in a notebook."},
      {"type": "obj", "bbox": [518, 155, 630, 443], "desc": "An orange tabby cat curled up and sleeping peacefully on the windowsill."}
    ]
  }
}

Specifications


Base model	`ideogram-ai/ideogram-4-fp8`
Architecture	ideogram4
Network	lokr (linear 32, alpha 32, factor 8, full rank)
Training steps	5000
Resolution	512 / 768 / 1024
Optimizer	automagic3
Learning rate	1e-5
dtype	bf16