MossImage

MOSS-IMAGE · LAYOUT · CHECKED OCTOBER 21, 2026

MOSS Image Layout Prompt Tool

MOSS-Image can take a JSON layout that places each headline, product or person in its own box. Writing those coordinates by hand is slow, so this MOSS Image layout tool lets you draw the boxes, fill in what goes in each one, and copy a JSON prompt in the format the official user guide describes.

Format from the official MOSS-Image user guide, where it is marked experimental. MossImage is an independent site.

Build a MOSS Image layout

Drag on an empty part of the canvas to draw a box. Boxes snap to 16 px.

Items (2)

  • 1
  • 2

Layout JSON

{
  "summary": {
    "meta": [
      "poster"
    ],
    "canvas": "single",
    "caption": "A vintage screen-printed travel poster of Kyoto at dusk, warm orange sky, soft paper grain."
  },
  "items": [
    {
      "label": "text",
      "caption": "Bold cream serif headline",
      "text": "KYOTO",
      "bbox": [
        -448,
        -400,
        -272,
        400
      ]
    },
    {
      "label": "building",
      "caption": "A red five-storey pagoda in silhouette against a dusk sky",
      "bbox": [
        -192,
        -224,
        384,
        224
      ]
    }
  ],
  "bbox_coord_space": "abs_centralized",
  "bbox_format": "yxyx"
}

The MOSS Image JSON layout format

KeyTypeWhat it holds
summary.metaList of stringsShort type tags for the whole image, for example poster.
summary.canvasStringA canvas note; the guide’s example uses “single”.
summary.captionStringOne description of the whole scene and its style.
items[].labelStringThe element type: text, product, person, building…
items[].captionStringWhat that element looks like.
items[].textString, optionalExact words to render inside the box.
items[].bbox4 numbersThe box, in the format and space named below.
bbox_format"yxyx"Order: y_min, x_min, y_max, x_max.
bbox_coord_space"abs_centralized"Pixels measured from the canvas centre.

How the coordinates work

Most tools measure from the top-left corner. MOSS-Image measures from the centre: subtract half the width from x and half the height from y. On a 1024×1024 canvas, a box whose top-left corner is at (112, 64) and which is 800 wide and 176 tall becomes [-448, -400, -272, 400].

Keep the canvas the same size as the image you generate, with both sides a multiple of 16; otherwise the boxes no longer line up with the picture.

A worked MOSS Image layout example

The tool opens with a small travel poster so you can see how the pieces fit. The overall caption sets the scene: a vintage screen-printed poster of Kyoto at dusk with an orange sky and paper grain. It does not mention the headline or the pagoda, because each of those has its own box.

Box 1 is the headline. Its label is text, its caption describes the lettering (bold cream serif) and its text field holds the exact word, KYOTO. It spans most of the width near the top, so the model has room for large letters. Box 2 is the subject, a red pagoda in silhouette, placed in the lower middle and kept clear of the headline.

To adapt it, change the canvas to the shape you will publish, rewrite the overall caption for your scene, and replace the two boxes with your own elements. Add a third box for a date or a logo area, keep a gap between text boxes, and copy the JSON when the warning line under the items is gone.

When a layout beats a plain prompt

A plain sentence is enough for most single-subject images, and MOSS-Image already places headlines sensibly. Reach for a layout when position matters more than wording: a product that must sit in the left third, a price badge in one corner, two languages in separate blocks, or a series of posters that need the same grid.

Layouts are also easier to review. Each element has its own caption and text, so when something comes out wrong you can change one box instead of rewriting the whole prompt. Because the format is experimental, compare the result with a plain prompt of the same scene and keep whichever works better. Nothing you type here is uploaded or stored: the JSON is built in your browser, so you can try as many versions as you like.

Tips for layouts that work

  • Start with three to five boxes: a headline, the main subject and one supporting element.
  • Put the exact words in text and describe the type style in caption, such as bold cream serif.
  • Use the overall caption for the background, palette and mood, not for things already placed in boxes.
  • Leave gaps between boxes. Overlapping text boxes tend to merge.
  • For simple posters, plain sentences may be enough; see the MOSS Image prompt guide.

Running MOSS-Image in ComfyUI? The native nodes include a MOSS-Image Layout Prompt node; see the official ComfyUI guide and our MOSS Image ComfyUI setup guide.

MOSS Image layout questions

What is a MOSS Image JSON layout?

An experimental prompt format from the official user guide. A summary describes the whole image, and a list of items gives each element a label, a caption, optional text to render and a bounding box, so you can say exactly where a headline or a product should go.

How are the bounding box coordinates measured?

In pixels from the centre of the canvas, with x growing to the right and y growing downward, in the order y_min, x_min, y_max, x_max. On a 1024×1024 canvas both axes run from -512 to 512. This tool converts the boxes you draw for you.

Where do I use the JSON?

The user guide documents the format and notes that the model can mix natural language with structured input. ComfyUI’s native nodes include a MOSS-Image Layout Prompt node. How you pass the JSON depends on your setup, so check the official guides for your runtime.

Does the layout guarantee exact positions?

No. OpenMOSS marks the format as experimental. Boxes guide placement, but the model can still move, resize or merge elements. Keep the number of boxes small and check the result.

Do I need an account to use the layout tool?

No. The tool runs in your browser and nothing is uploaded. Copy the JSON when you are done.