Single-Reference Editing

Send one image in images and describe the change in prompt. For the request format, see Image editing.

Write the instruction

Write the instruction in three parts:

Name the target precisely enough that only one thing matches. For a flock, specify the bird’s position and appearance. Pick it out by position, size, color, or what it’s next to.
Say what changes, and into what. “Replace the flower with a slice of lemon.” “Change the jacket from red to blue.” Name the element as it is now and as it should be.
Name any details that matter. When a specific detail must stay, name it: “keep her pose”, “keep the lace embroidery white”.

FLUX 3 example: recolor two elements, keep the rest

The prompt names both targets by what they are (“the large tiger”, “the small glowing butterfly”), states the change once, and lists the three things that must not change. The tiger’s stripes, the snow on its fur, and the butterfly’s glow carry over; only the color changes. This example comes from the FLUX 3 launch post, where it was sent with box rows: <ref_image_0> is the first image in images, and each <id> token matches a row at the end of the prompt. Edit with boxes shows the full rows. For a plain written edit, keep the same sentence structure and leave out the tokens.

Specify colors, text, and positions

Words like “pink” or “near the top” leave room for interpretation. When the exact value matters, write the value:

To get	Write	Example
A target color	A hex code	“change the color of only one bird in the middle to #e01075”
Exact words	The text in quotes, plus where it goes	“Change the text on the neon sign to ‘zum Schlappen’”
A specific place	A region of the frame	“On the lower left path, replace the tiny wolf with a cyberpunk dog”
A preserved detail	The detail by name	“keeping all lace embroidery details white and fully visible”

FLUX 3 renders text in images, including multi-line poster typography. For a text edit, quote the exact words and describe their placement and style (“in handwritten pink marker, diagonally across the top photo”).

Make several changes in one prompt

For a multi-part edit, write an edit record: name the image and the operation, give each change with its location, then list what stays. This is the pattern the FLUX 3 launch examples use:

Name the image and the operation. “Modify <ref_image_0> by making three replacements.”
Give each change: where, what, and into what. “On the lower left path, replace the tiny anthropomorphic wolf with a cyberpunk dog.”
List what stays. “Keep the rest of the scene exactly unchanged, preserving the stone lantern, the frozen stream…”

FLUX 3 example: three replacements at once

Each replacement names its location and subject to distinguish the targets. The preserve list names every major element of the scene, not only “the rest of the image”. This edit was also sent with box rows; Edit with boxes has the full request.

When to switch to box rows

A written instruction is enough when only one thing in the image matches your description. Add box rows to the prompt when:

Several similar things match and you want to change one of them, such as one car in a row of parked cars.
The change needs an exact position or size.
You want to move or resize an element and keep it otherwise the same.
You want to remove one specific object and let FLUX 3 fill the space it leaves.
You want to keep specific elements in place while you change others.

Each row names one element and gives its boxes on a 0–1000 grid: where it is in the source image and where it belongs in the output. Edit with boxes explains the row types and has a complete request.

Edit types

The catalog below covers the most common single-image edits, each with the prompt that produced it. Results captioned “FLUX 3 Image” were generated with FLUX 3 Image. The rest were generated with FLUX.2.

Background Replacement

Change or replace the background while keeping the subject intact. Describe the new setting and its lighting so the subject fits into it.

Can on a black surface

Bottle in strawberries

Output: the Spritz bottle standing in a pile of wet strawberries on a white background — FLUX 3 Image

Scene replacement

Output: subject in cozy home environment — FLUX.2 result

Style Transfer

Change the visual style or medium of an image, from illustration to photorealism or from photo to painting. Say what should keep its shape: “keep the proportions and layout exactly as in the illustration”.

Photo to oil painting

Output: oil painting style — FLUX.2 result

Illustration to photorealism

Output: photorealistic house — FLUX.2 result

Reskin to mountain vista

Output: mountain vista transformation — FLUX.2 result

Object Manipulation

Add, remove, or replace objects in a scene. Be specific about what should change and what should stay. With FLUX 3 Image, if the object you want to remove or replace is one of several similar ones, use a box row to point at it.

Remove object

Output: sprinkles removed — FLUX.2 result

Replace object

Output: flower replaced with lemon slice — FLUX.2 result

Add object

Output: goblins added to gorge wall — FLUX.2 result

Replace subject

Output: polar bear replaces DJ — FLUX.2 result

Selective replacement

Output: cherries replaced with sprinkles — FLUX.2 result

Remove vegetation

Output: clean stone statues — FLUX.2 result

Object swap: bike to horse

Output: person on rearing black horse — FLUX.2 result

Element replacement: feathers to petals

Output: portrait with rose petals — FLUX.2 result

Color & Material Changes

Recolor specific elements or transform materials. Use a hex code to specify a target color, then check the result.

Recolor with hex codes

Output: cow with custom colors — FLUX 3 Image

Material transformation: silver

Output: silver butterfly — FLUX.2 result

Material transformation: ice

Lighting, Weather & Season Changes

Shift the time of day, season, or weather with a short instruction. Name the new condition plainly: “Change it to Night”, “Change this to Winter”.

Time of day: night, mountain village

Season change: winter

Time of day: night, car on a salt flat

Lighting and color mood

Output: warm autumn lighting — FLUX.2 result

Text Editing

Add, replace, or modify text within images, from simple swaps to full ad layouts. Put the exact words in quotes and say where they go and how they should look.

Handwritten text on a photo

Simple text replacement

Output: text changed to Flux.2 — FLUX.2 result

Neon sign text

Output: neon sign with new text — FLUX 3 Image

Sign replacement + scene edit

Output: new neon sign and traffic light — FLUX.2 result

Ad creation with text overlay

Output: fashion ad with text and CTA — FLUX.2 result

Virtual Try-On & Clothing

Change outfits, add accessories, or adjust clothing colors. Useful for fashion and e-commerce.

Outfit change

Output: the same woman in a fuchsia dress against a green studio gradient — FLUX 3 Image

Add accessories with hex colors

Output: woman with fluffy jacket and hat — FLUX.2 result

Dress recoloring with detail preservation

Output: sky blue lace wedding dress — FLUX.2 result

Pose & Expression Changes

Adjust gaze direction, body pose, or facial expressions of subjects.

Eye/detail correction

Output: owl with open eyes — FLUX 3 Image

Gaze direction

Output: woman looking at camera — FLUX.2 result

Pose change

Output: woman in model pose — FLUX.2 result

Checklist: vague to specific

Instead of	Write	What it adds
“Make it better”	“Warm the light to late-afternoon sun and soften the shadows on her face. Keep the background unchanged.”	A concrete change and what stays
“Change the shirt”	“Change the man’s grey T-shirt to #c0392b, keeping the print on the front.”	The target, the exact new value, a preserved detail
“Remove the car” (with two cars in frame)	“Remove the white car parked on the right. Keep the red car and the street unchanged.”	Only one thing matches
“Add some text”	“Add the words ‘Open late’ in white sans-serif letters on the shop window.”	The exact words, placement, and style
“Fix the image”	“Open the owl’s eyes, making them look natural.”	The defect named

Multi-reference editing

Combine up to 10 images in one FLUX 3 Image request.

Edit with boxes

Target exact regions with box rows.

Video Generation with FLUX

Image Generation with FLUX