FLUX 3 Image

FLUX 3 Image Editing

Edit an image with a written instruction, combine up to ten references, and add bounding boxes to place elements or target an edit.

Quickstart

Send an image in images and say what to change. This request recolors an owl and keeps the rest of the photo as it was. Drag the handle to compare:

Set BFL_API_KEY to a key from the BFL dashboard, then submit:

import os
import requests

response = requests.post(
    "https://api.bfl.ai/v1/flux-3-image",
    headers={"x-key": os.environ["BFL_API_KEY"]},
    json={
        "prompt": "In this image, change <owl_subject>, previously described as \"a short-eared owl in flight with brown and tan mottled feathers and yellow eyes\", to: a short-eared owl in flight with striking snowy white and charcoal grey mottled feathers and bright yellow eyes. Keep everything else in the image — composition, other elements, lighting, background — exactly unchanged.",
        "images": ["https://cdn.sanity.io/images/2gpum2i6/production/1731db585c799dbb4e2c48731ea0b5bfc92a47d7-1800x1193.webp"],
    },
    timeout=60,
)
response.raise_for_status()
print(response.json())

curl --fail-with-body https://api.bfl.ai/v1/flux-3-image \
  -H "x-key: $BFL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "In this image, change <owl_subject>, previously described as \"a short-eared owl in flight with brown and tan mottled feathers and yellow eyes\", to: a short-eared owl in flight with striking snowy white and charcoal grey mottled feathers and bright yellow eyes. Keep everything else in the image — composition, other elements, lighting, background — exactly unchanged.",
    "images": ["https://cdn.sanity.io/images/2gpum2i6/production/1731db585c799dbb4e2c48731ea0b5bfc92a47d7-1800x1193.webp"]
  }'

aspect_ratio defaults to auto, so the edit keeps the shape of the image you send. Poll the returned polling_url until status is Ready, then download the image from result.sample within 1 hour. FLUX 3 Text to Image has complete clients that submit, poll, and save.

Edit with an instruction

These examples call generate() from the complete clients, which submits a request, polls until it is ready, and saves the image.Send the image in images and write the change as an instruction. Name the element precisely enough that only one thing in the image matches. This request changes the color of one bird in a flock:

generate({
    "prompt": "change the color of only one bird in the middle to #e01075",
    "images": ["https://cdn.sanity.io/images/2gpum2i6/production/4792198dfeba9223bf3ecf020fed2942de7f0bd7-1800x1200.webp"],
}, name="pink_bird")

With aspect_ratio left at auto, the output keeps the aspect ratio of the image you send.You don’t have to draw a box for a single edit. Before it generates, FLUX 3 Image expands your instruction into a detailed prompt, and for many edits it also finds the element you named and adds a box for it. The expanded prompt, including any boxes it added, comes back in result.prompt. To control exactly which elements change, use bounding boxes.

Multi reference

Refer to each image by its position or by what it shows. “Image 1” is the first entry in images, “image 2” the second, and so on up to ten. Here, image 1 is the scene and image 2 supplies the object to place in it:

generate({
    "prompt": "replace the cartridge from image 1 with the van in image 2, adjust the lighting on the van to integrate well within image 1",
    "images": [
        "https://cdn.sanity.io/images/2gpum2i6/production/ec365238b6efa1b29f573e83b774c6d539408ea8-1800x1200.webp",
        "https://cdn.sanity.io/images/2gpum2i6/production/ff3bf81b5a8b38f0d870561217b68a72964259ab-1131x1414.webp",
    ],
}, name="turntable_van")

With aspect_ratio at auto, the output uses the aspect ratio of the first image in images. The turntable is first, so the result is landscape. For a different shape, put a different image first or set aspect_ratio.For product shots, try-ons, and group compositions, say how the images relate: “the person from image 1 wears the jacket from image 2 and stands in the kitchen from image 3”.

Examples

Each example shows the inputs, the result, and the prompt that produced it.

Single Image Reference

Send one image and write the change as an instruction. Drag the handle to compare:

Multi Image Reference

Refer to each image by its position in images. The output keeps the aspect ratio of image 1.Replace an object with a new designAdd elements from several imagesFurnish a room

Edit with Bboxes

Each edit changes one boxed element and anchors the rest. Select an example, then switch between before and after.

Send a request

Copy a request body from the bounding box examples and paste it where indicated below, including its outer braces. To adapt an edit to your image, update the image URL, element descriptions, and source and target boxes.

import json

body = r'''
Paste the complete copied JSON body here.
'''
generate(json.loads(body))

curl --fail-with-body https://api.bfl.ai/v1/flux-3-image \
  -H "x-key: $BFL_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<'EOF'
Paste the copied request body here.
EOF

generate() is defined in the complete clients. It submits the request, polls until the image is ready, and saves it. With cURL, poll the returned polling_url until status is Ready, then download result.sample.

Coordinate conversion and editing tips

Use the aspect ratio you designed the layout for, because the grid stretches with the frame. For edits, auto follows the first reference; without a reference, it produces a square.

def to_bbox(left, top, right, bottom, width, height):
    return [round(top / height * 1000), round(left / width * 1000),
            round(bottom / height * 1000), round(right / width * 1000)]

to_bbox(384, 108, 1536, 972, 1920, 1080)  # [100, 200, 900, 800]

To move or resize an element, use different source and target boxes. Use matching boxes to ask for it to stay in place. State additions and removals in the written instruction as well as the rows.Very small boxes can fail. In our tests, a new element in a box about 40 × 25 pixels often did not appear, so give new elements a larger box.Give text its own box and quote the words in desc. See Layout prompts for more examples.

Try drawing a box

Draw over a marked region, or choose one below the image, then select an object. This demo uses saved edits and makes no API requests.

Pixel-perfect edits

Pixels outside the boxes usually stay the same. Shadows, reflections, or nearby lighting may still change.Compare each input with its edit. Switch to Changed pixels to see which areas the edit touched.

Keep exploring

FLUX 3 Text to Image

Complete clients that submit, poll, and save, plus edits and multi-reference requests.

API reference

The full request schema, limits, and response codes.

Layout prompts

How to write the caption and element table for a layout.

Get Started

Account Management

FLUX 3

FLUX.2

FLUX Tools

API Integration

Training

Legacy Models

FLUX 3 Image Editing

FLUX 3 Image Editing

Quickstart

Edit with an instruction

Multi reference

Examples

Single Image Reference

Multi Image Reference

Edit with Bboxes

Send a request

Keep exploring

FLUX 3 Text to Image

API reference

Layout prompts

​FLUX 3 Image Editing

​Quickstart

​Edit with an instruction

​Multi reference

​Examples

​Single Image Reference

​Multi Image Reference

​Edit with Bboxes

​Send a request

​Keep exploring

FLUX 3 Text to Image

API reference

Layout prompts

FLUX 3 Image Editing

Quickstart

Edit with an instruction

Multi reference

Examples

Single Image Reference

Multi Image Reference

Edit with Bboxes

Send a request

Keep exploring