JSON Prompting with MOSS Image 5B

Looks like MOSS Image lets you prompt in JSON. Here's the JSON prompt:

{
  "summary": {
    "meta": ["architectural cutaway", "fantasy diorama"],
    "canvas": "single",
    "caption": "A breathtaking highly detailed architectural cross-section of a magical six-room apartment building, with exactly three columns and two rows of equally sized rooms. Every room has a completely different inhabitant and activity. Perfectly aligned rectangular rooms separated by clearly visible walls and floors. Beautiful miniature diorama aesthetics, warm cinematic lighting, sophisticated 3D rendering, intricate furnishings, no text or signage."
  },
  "items": [
    {
      "label": "object",
      "caption": "TOP LEFT ROOM: A cozy observatory where an elderly astronomer in a white sweater looks through a brass telescope pointing out of a large window toward a sky filled with stars. Books, globes, and astronomical instruments.",
      "bbox": [-440, -720, -40, -280]
    },
    {
      "label": "object",
      "caption": "TOP CENTER ROOM: A graceful ballerina in a white dress practicing alone in an elegant miniature dance studio with wooden floors, a grand mirror, and a glowing chandelier.",
      "bbox": [-440, -230, -40, 220]
    },
    {
      "label": "object",
      "caption": "TOP RIGHT ROOM: A cheerful Italian chef tossing a pizza into the air inside a warm rustic kitchen with a flaming brick oven and hanging copper pans.",
      "bbox": [-440, 280, -40, 720]
    },
    {
      "label": "object",
      "caption": "BOTTOM LEFT ROOM: An underwater laboratory completely filled with turquoise water, where a scuba diver examines colorful coral and small tropical fish. The water remains confined within this room.",
      "bbox": [40, -720, 460, -280]
    },
    {
      "label": "object",
      "caption": "BOTTOM CENTER ROOM: A small friendly robot carefully repairing a vintage computer at a cluttered workbench, surrounded by circuit boards, tools, tiny mechanical parts, and warm desk lamps.",
      "bbox": [40, -230, 460, 220]
    },
    {
      "label": "object",
      "caption": "BOTTOM RIGHT ROOM: A botanist watering enormous glowing blue mushrooms in an indoor greenhouse filled with exotic plants, moist soil, glass terrariums, and bioluminescent vegetation.",
      "bbox": [40, 280, 460, 720]
    }
  ],
  "bbox_coord_space": "abs_centralized",
  "bbox_format": "yxyx"
}

The MOSS Image generated image:

Just to compare, I tried on ChatGPT:

What the prompting looks like in Maestro (You literally paste the full JSON in):

Post

JSON Prompting with MOSS Image 5B

Replies