Skip to main content
Send one or more images in images and write the change you want in prompt. Use the same endpoint as for image generation. There is no edit mode, mask, or strength setting: the prompt says what to do with the images, whether that is changing a detail, restyling, or combining them.

How an edit request works

This edit changes one bird in a flock:
The instruction specifies a target, a position, and a color. Check the result when color accuracy or unchanged pixels matter. Pixel preservation shows measured examples.

Combine several references

With more than one image, give each a role and refer to it by its position in images. Here, image 1 is the subject and image 2 is the style: aspect_ratio: "auto" follows the first reference, so the result is landscape like the temple. Put the image whose aspect ratio you want first, or set aspect_ratio. Multi-reference editing covers roles, ordering, and larger compositions.

When to use boxes

Start with a written instruction. Add box rows when several similar objects match the description, or when you need to specify a position or size. Boxes can identify an element in a reference and its intended position in the output.

What an instruction can change

Each tab below sends the same input image with one instruction. The four edits are separate requests, not a chain.

Guides in this section

Single-reference editing

Write the instruction for one image, with a catalog of edit types.

Multi-reference editing

Combine up to 10 images: roles, ordering, and composites.

Edit with boxes

Target exact regions with box rows: recolor, replace, move, or remove.

FLUX 3 Image Editing

The API request for edits, with parameters and limits.