v1.3.4API credit, right in the headerChangelog →
MergeRef
Notify Me
Tools

Layout boxes (FLUX 3 Image)

Tell FLUX 3 Image where each element goes, or which parts of an image to keep, move, change or remove. Layout boxes appear in the prompt panel when FLUX 3 Image is selected.

Layout boxes — demo video

New in FLUX 3

Bounding boxes are a new feature of Black Forest Labs' FLUX 3 Image. They are written into the prompt: a caption that names each element as <id>, then a JSON table with each element's box as [top, left, bottom, right] on a 0–1000 grid and a description. To edit an image, each row also says where the element comes from and where it should end up. See Bounding boxes with FLUX 3 Image in the Black Forest Labs docs.

More accurate and intuitive with AI analysis

MergeRef adds AI analysis to FLUX 3 boxes, so you draw what you mean and the app turns it into an accurate table:

  • Draw instead of typing coordinates. Drag boxes on the frame or on the original; MergeRef writes the 0–1000 table.
  • AI finds the elements. Auto-extract sends the original to Gemini once and adds its main elements, up to ten, as Keep boxes with descriptions.
  • AI names what you drew. Draw a Keep or Remove box by hand and Gemini reads that region and fills in what the element is, so the model finds the right one.
  • AI checks for contradictions. A costume, pose or style reference fights a Keep box on the same element. MergeRef warns you, and auto-extract leaves people out while one is on.
  • Mistakes are caught before you pay: boxes too small to appear, a caption that never mentions an <id>, bad ids or coordinates, a ratio that would stretch the boxes, multi-grid output turned on.
  • Translation keeps the table intact: the caption and each description are translated; ids and boxes stay as they are.

Box types

  • Keep: the element stays exactly where it is. Its description only identifies it.
  • Move: drag a kept element to a new spot or size. The old spot shows dashed.
  • Edit: the box is generated from your description: replace, recolor or add.
  • Remove: the element disappears and the area is filled with what was behind it.
  • Composite: bring part of a reference image into this box. Pick the source area on the reference.
  • Without an original: new-element boxes lay out a fresh image.
Tip
Boxes guide placement and scale; they are not a hard mask, so an element can run slightly past its box. Keep the aspect ratio you drew the boxes for.

AI cost

Auto-extract and box naming each make one Gemini 3.8 Flash call with one image up to 1024 px: under $0.01 per call (official price checked 4 November 2026, Google pricing). They need a Gemini API key or a Gemini Enterprise Agent Platform (formerly Vertex AI) connection; without one, type the descriptions yourself. Boxes work with up to 10 reference images. For image prices see Black Forest Labs — FLUX.