mossimageai

WORDS, IMAGES & COMPOSITION

The same vase, a different composition

An original 1024 × 1024 image generated with MOSS-Image SFT from a structured JSON layout. Inspect the actual input below, then reuse it in the studio.

The same vase, a different composition

The boxes show the input specification. The image is the actual generated output.

Original generated PNG · selected by our team · no retouching or upscaling.

Input and generation settings

ModelMOSS-Image SFT · moss-image-sft-1.0
Input formatLayout JSON
Output1024 × 1024 px · 1:1
StyleSpecified inside the JSON
Seed640231
Steps / guidance50 / 3
Created2026-10-23

The submitted layout JSON

The bounding boxes use centered pixel coordinates in [top, left, bottom, right] order. This case uses a 1024 × 1024 canvas. The boxes specify intended placement; inspect the actual image to see how the model followed them.

{
  "summary": {
    "meta": [
      "product poster"
    ],
    "canvas": "single",
    "caption": "A restrained product poster on a warm ivory background, featuring a small matte orange ceramic vase and a short dark headline. Soft light from the upper left, generous empty space."
  },
  "items": [
    {
      "label": "text",
      "caption": "Large dark uppercase sans-serif headline on a single line at the bottom.",
      "text": "A LITTLE COLOR",
      "bbox": [
        288,
        -416,
        448,
        416
      ]
    },
    {
      "label": "object",
      "caption": "One matte orange ceramic vase with a short neck, centered on a pale stone surface. A soft shadow falls to the right.",
      "bbox": [
        -384,
        -240,
        192,
        240
      ]
    }
  ],
  "bbox_coord_space": "abs_centralized",
  "bbox_format": "yxyx"
}

Words requested in this image

  • A LITTLE COLOR

The requested headline is below the vase. The model changes the vase proportions and lighting as well as the arrangement; the same seed does not lock every visual detail. Compare the requested boxes with the actual output.

Make this case your own

  1. Load the actual input and settings into the studio.
  2. Change the headline, element descriptions or bounding boxes.
  3. Press Generate image, then review your own full-size result.

Loading a case does not start generation. Results may differ, even with the same seed; model and service changes can affect reproduction.

Model and input format references: OpenMOSS project and official user guide.

More generation cases