Multi-Reference Editing

Multi-reference editing combines several input images into one output. Use it to restyle a photo after an artwork, put a product into a scene, dress a person in garments from other photos, or keep a character consistent across variations. In the prompt, clarify the purpose of each image and how the pieces fit together.

FLUX 3 example: put an object from one image into another

Image 1 is the scene and image 2 supplies the object, and the prompt names both roles. It also says how the new object should fit: the van takes the lighting of image 1. The result is landscape like the turntable because aspect_ratio: "auto" follows the first reference.

How FLUX 3 reads multiple references

Send up to 10 images. FLUX 3 Image takes 1 to 10 entries in images, as URLs or base64 strings. Each must be at least 256 × 256 pixels and at most 16 megapixels. Refer to each image by position. “Image 1” is the first entry in images, “image 2” the second, and so on. Use the same words every time you mention an image. Specify what each image supplies. A reference can supply more than one feature, such as both the material and the background. Name those roles explicitly:

Role	What it supplies	Example phrasing
Subject	A person, animal, or character to keep recognizable	“the woman in image 2”
Product	An object whose shape, color, and label you want to keep	“the bottle from image 3, label facing the camera”
Setting	The scene, room, or background everything goes into	“the kitchen in image 1”
Style	Palette, medium, texture, or lighting to apply	“in the style of image 4”

Say how the pieces relate. Name positions and interactions, not only the parts: “the woman in image 2 sits on the swing in image 1, with the cat from image 3 on her lap.” Put the base image first. With aspect_ratio: "auto" (the default), the output takes its aspect ratio from image 1. Put the setting or base photo you are editing first, or set aspect_ratio explicitly when no reference has the shape you want.

Specify an object’s position

A written instruction such as “put the lamp from image 2 on the sideboard” lets FLUX 3 choose the position. To specify it, add box rows to the end of the prompt. A row with from: "ref_image_1" takes an element from the second image: src_bbox marks the object in image 2, and tgt_bbox marks where it belongs in the output. Anchor rows (from: "ref_image_0" with the same source and target box) ask the model to keep those elements in place.

import json

room = "https://example.com/living-room.jpg"  # image 1: the scene to edit
lamp = "https://example.com/ceramic-lamp.jpg"  # image 2: the product

record = (
    "Modify <ref_image_0> by placing the ceramic table lamp <lamp_1> on the left end "
    "of the oak sideboard <sideboard_1>. Keep the rest of the room exactly unchanged, "
    "preserving the framed print <print_1> above the sideboard."
)
rows = [
    {"id": "lamp_1", "from": "ref_image_1", "src_bbox": [80, 250, 950, 750], "tgt_bbox": [300, 120, 560, 280], "desc": "a ceramic table lamp with a linen shade"},
    {"id": "sideboard_1", "from": "ref_image_0", "src_bbox": [540, 60, 880, 940], "tgt_bbox": [540, 60, 880, 940], "desc": "an oak sideboard"},
    {"id": "print_1", "from": "ref_image_0", "src_bbox": [120, 330, 420, 670], "tgt_bbox": [120, 330, 420, 670], "desc": "a framed botanical print"},
]

generate({
    "prompt": f"{record} {json.dumps(rows)}",
    "images": [room, lamp],
}, name="lamp_on_sideboard")

The coordinates are illustrative: measure the boxes on your own images, as [top, left, bottom, right] on a 0–1000 grid. generate() is the helper from FLUX 3 Text to Image. Placement is a strong hint, not a hard constraint, and multi-reference compositions can place an element partly outside its box. Edit with boxes covers every row type.

More examples

These examples were generated with FLUX 3 Image. Hover an input to highlight where the prompt refers to it.

Paintings on coins, from four references

One image sets the scene and three supply artwork. Each painting appears on its own coin, and the hand and page stay the same.

A tower sinking into the sea

The tower is tilted and cut off at the waterline, with waves where it meets the sea.

A knife made of candy

The shape comes from image 2 and the material and background come from image 1, including the loose candies around the edge.

A heart in paper

Only the material changes. The shape, the veins, and the black background stay the same.

Use cases

The categories below cover the most common multi-reference tasks, each with the prompt that produced the result. The “Animal placed in scene”, “Pattern onto plate”, “Fill bottles with liquid”, and “Logo engraved in tree” results were generated with FLUX 3 Image. The rest were generated with FLUX.2.

Scene Compositing

Combine elements from multiple source images into a single coherent scene.

Vintage skating scene from three references

FLUX.2 result:

Animal placed in scene

Underwater room

FLUX.2 result:

Style & Material Transfer

Apply the visual style, texture, or material of one image onto the content of another.

Impasto style transfer

FLUX.2 result:

Animal pattern transfer

FLUX.2 result:

Pattern onto plate

Object Replacement

Replace or fill objects with elements from another reference image.

Fill bottles with liquid

Logo & Branding

Place logos from one image onto objects or scenes in another.

Logo engraved in tree

Smoke shaped as logo

FLUX.2 result:

Checklist: vague to specific

Instead of	Write	What it adds
“Combine these images”	“Place the couple from image 2 at the table in image 4, with the view from image 1 through the window.”	A role and a position for each image
“Use the style of the other one”	“Turn image 1 into the style of image 2, keeping the temple’s shape and layout.”	Which image is which, and what to keep
“Add the product”	“Put the bottle from image 3 on the counter in image 1, label facing the camera, lit by the window on the left.”	The product’s placement, orientation, and light
“Make them wear it”	“The person from image 1 wears the jacket from image 2 and stands in the kitchen from image 3.”	How the pieces relate

Single-reference editing

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

Edit with boxes

Place and anchor elements with box rows.

Video Generation with FLUX

Image Generation with FLUX

Image Editing with FLUX

Use Cases with FLUX

FLUX 3 example: put an object from one image into another

How FLUX 3 reads multiple references

Specify an object’s position

More examples

Paintings on coins, from four references

A tower sinking into the sea

A knife made of candy

A heart in paper

Use cases

Scene Compositing

Style & Material Transfer

Object Replacement

Logo & Branding

Checklist: vague to specific

Single-reference editing

Edit with boxes

​FLUX 3 example: put an object from one image into another

​How FLUX 3 reads multiple references

​Specify an object’s position

​More examples

​Paintings on coins, from four references

​A tower sinking into the sea

​A knife made of candy

​A heart in paper

​Use cases

​Scene Compositing

​Style & Material Transfer

​Object Replacement

​Logo & Branding

​Checklist: vague to specific

Single-reference editing

Edit with boxes

FLUX 3 example: put an object from one image into another

How FLUX 3 reads multiple references

Specify an object’s position

More examples

Paintings on coins, from four references

A tower sinking into the sea

A knife made of candy

A heart in paper

Use cases

Scene Compositing

Style & Material Transfer

Object Replacement

Logo & Branding

Checklist: vague to specific