Place elements with bounding boxes
Every box is [top, left, bottom, right] on a 0-1000 grid, written at the end of the prompt. The example shows a wolf placed inside a bamboo grove alongside a lantern, a stream, and a fallen log.
FLUX 3 Image · Black Forest Labs
FLUX 3 Image online with bounding box control
Lay out every element on a 0-1000 grid, combine up to ten references, and edit a finished frame one box at a time while everything you did not touch stays put.
Every image in this gallery is official Black Forest Labs showcase material published on bfl.ai/models/flux-3-image, re-encoded for this page. They are provider examples, not output from this site.
01
Create with FLUX 3 Image
Write a prompt, pick an aspect ratio and resolution, and add up to ten reference images when the subject or style has to stay consistent.
AI Image
Results
Creation historyGenerated content will appear here
02 · Overview
FLUX 3 Image at a glance
Facts last checked:
- Provider
- Black Forest Labs
- API model
- flux-3-image
- Modes on this site
- Text to image, image to image, reference to image, image edit
- Resolutions
- 768sq, 1k, 1.5k, 2k, 4k (default 1k)
- Aspect ratios
- 16 options from 21:9 to 9:21, plus auto
- Reference images
- 1 to 10 per request, 256 × 256 px to 16 MP each
- Layout control
- Bounding boxes on a 0-1000 grid, [top, left, bottom, right]
- Prompt grounding
- Web and image search before generating, on by default
- Site price
- From 6 points per generation, by resolution
What is FLUX 3 Image?
FLUX 3 Image is the image model from Black Forest Labs, the research lab behind the FLUX family. Where most image models accept a prompt and return a composition, FLUX 3 Image adds a layout language: you describe each element and give it a bounding box on a 0 to 1000 grid, and the model renders the scene around those regions.
Black Forest Labs describes three things the model is built for. The first is generation from a prompt alone. The second is combining up to ten reference images, describing what each one contributes. The third is editing: the same boxes that place an element can recolor, move, replace, or remove it while the rest of the frame is left unchanged.
The API is a single endpoint, POST /v1/flux-3-image, where only the prompt is required. Responses return a polling URL; poll until the status is Ready, then download the sample within one hour. Every generation is a separate request, so the model does not keep a canvas between calls the way an interactive session does.
03 · Capabilities
What FLUX 3 Image can do
Four capabilities that separate it from a prompt-only image model, each shown with the provider’s own published example.
Edit one box, keep the rest
Recolor, move, replace, or remove a single element. The provider’s example replaces one drawing in a children’s room while the paper, card, and table stay exactly as they were.
Recolor with locked regions
Elements you do not name are treated as locked. In the provider’s example the wetsuit and board change color while the ocean wave, the cloudy sky, and the rest of the composition are held.
Ground the prompt with search
Grounding is on by default: the model runs a web and image search before generating, so real things come out right. A prompt asking for this year’s official Oktoberfest poster is the provider’s own example. Set grounding to false for faster results that use only your prompt.
What a FLUX 3 Image generation costs here
Black Forest Labs does not publish a price sheet for FLUX 3 Image, so the only prices that can be stated are this site’s own. Cost is charged per generation and scales with the resolution you pick.
See full pricing Starting at
points per generation
Cost by resolution
Charged when the generation runs
04 · Use cases
Where FLUX 3 Image fits
Cases where the bounding box layout or the one-box edit is the deciding factor, not just a nicer prompt.
Every image in this gallery is official Black Forest Labs showcase material published on bfl.ai/models/flux-3-image, re-encoded for this page. They are provider examples, not output from this site.
Official showcase gallery01
Architecture with a fixed layout
Housing and court layouts where the position of each element is decided in advance and the model fills in materials, light, and detail.
Official showcase gallery02
Nature scenes with depth cues
Forest and landscape work where foreground, midground, and background need to be separated before rendering.
Official showcase gallery03
Landscape with a wide canvas
Wide aspect ratios where the horizon and the foreground need explicit placement rather than whatever the prompt implies.
Official showcase gallery04
Illustrated interiors
Stylized interior scenes where several distinct objects each need their own region of the frame.
Official showcase gallery05
Object-focused illustration
A single subject centered in a simple frame, where composition matters more than scene complexity.
Official showcase gallery06
Multi-reference portraits
Street-style portraits assembled from several reference images at once, with each reference named in the prompt.
05 · Prompts
Prompt examples
Start from a real case, then rewrite the boxes to match your own layout. Each prompt is filled into the workspace with FLUX 3 Image selected, not sent for you.
Prompt 01
Add a subject to a finished frame
A pixel-level edit that adds two divers to an existing underwater scene, leaving the water and the existing figures where they were.
Add two divers to this image. Keep the water, the light shafts, and every existing figure exactly as they are, and place the new divers inside the open water on the left.
Prompt 02
Replace one object, keep the composition
A single object swap in a street scene. The crosswalk, the crowd, and the buildings stay fixed while the subject changes.
Replace the duck in this image. Keep the crosswalk, the crowd, the cars, and the buildings exactly unchanged, and match the new subject to the existing camera angle and lighting.
Prompt 03
Change one object in an interior
Swap the sofa in a living room. Walls, window light, and every other object stay as they are.
Change the sofa in this room. Keep the walls, the window light, the rug, and all other furniture exactly as they are, and match the new sofa to the existing perspective.
Prompt 04
Apply artwork to a product
Place a design onto a garment in a product shot while the fabric, the folds, and the background stay put.
Print this design onto the shirt in this product photo. Keep the fabric texture, the folds, the lighting, and the background exactly unchanged.
Prompt 05
Recolor without touching the rest
A hair color change on a portrait. Face, pose, clothing, and background are all held.
Recolor the hair in this portrait. Keep the face, the expression, the pose, the clothing, and the background exactly unchanged, and keep the same hair texture and highlights.
06 · Workflow
How to direct FLUX 3 Image
- 01
Write the scene, then the boxes
Describe the setting in plain language, then append one line per element as name [top, left, bottom, right] using the 0-1000 grid. The provider’s example places a concrete dome, a coastal town, swimmers, a crowd, and poster text this way.
- 02
Name each reference image
Uploaded references get tokens in upload order, starting at ref_image_0. Reference them by token in the prompt and say what each one contributes, as in the provider’s six-reference streetwear example.
- 03
Pick the ratio, then the resolution
auto follows the first reference image and otherwise falls back to 1:1. Choose 768sq for a square thumbnail, 1k as the default, or step up to 4k when the output has to hold up when zoomed.
- 04
Edit by naming what changes
For an edit, say what should change and then list what must stay identical. Anything you do not name is treated as unchanged, which is how the provider’s wetsuit recolor keeps the wave and sky.
- 05
Leave grounding on unless speed matters
Grounding searches before generating, which helps when the prompt names something real and current. Turn it off when you want only your prompt to drive the result, or when you need the fastest turnaround.
Responsible use
What to know before you generate
Reference images must each be at least 256 × 256 px and no larger than 16 MP. The site processes up to 4 MP per request, so very large references are downscaled before they reach the model.
Bounding boxes travel inside the prompt text. There is no separate layout field in the API, and this workspace does not draw boxes for you; coordinates are typed at the end of the prompt in the order [top, left, bottom, right].
A 5456 × 3072 px frame is 16.8 MP, which is the provider’s largest published output. Higher resolutions cost more points here, and the step up is priced per resolution rather than per pixel.
safety_tolerance runs from 0, the strictest setting, to 4. The default is 2. Raising it permits more, and rejecting more, than the default does.
This workspace prices every generation in points and shows the cost before you run it. Black Forest Labs bills its own API separately, and publishes no price sheet for FLUX 3 Image, so no official price is quoted on this page.
Sources
Sources
Model facts on this page come from Black Forest Labs’ own documentation and from this site’s live model catalog. Showcase images are the provider’s published examples.
07 · FAQ
FLUX 3 Image questions
What is FLUX 3 Image?
It is Black Forest Labs’ image model, built on the FLUX 3 family. Beyond generating from a prompt, it lets you place elements with bounding boxes on a 0-1000 grid, combine up to ten reference images, and edit a single region of a finished image while the rest stays unchanged.
How do bounding boxes work in FLUX 3 Image?
Each box is written at the end of the prompt as name [top, left, bottom, right], with all four values on the 0 to 1000 grid regardless of the output aspect ratio. Name the element first, describe it, then give the box. The provider’s festival poster example places poster text, a dome, a town, swimmers, and a crowd this way.
How many reference images can I use?
Between one and ten per request. Each must be at least 256 × 256 px and at most 16 MP. References receive tokens in upload order starting at ref_image_0, and you refer to them by those tokens in the prompt.
What resolutions and aspect ratios are supported?
Five resolutions: 768sq, 1k, 1.5k, 2k, and 4k, with 1k as the default. Sixteen aspect ratios from 21:9 down to 9:21, plus auto, which follows the first reference image and otherwise falls back to 1:1.
Can FLUX 3 Image edit an existing image precisely?
Yes, and that is one of its main capabilities. You describe what should change and then list what must stay identical. The provider’s own examples recolor a wetsuit and surfboard, replace one drawing in a children’s room, and swap a subject, each time with the rest of the frame held.
What does grounding do?
Grounding runs a web and image search before the model generates, which helps when your prompt refers to something real and current. It is on by default. Setting it to false skips the search and produces results faster from your prompt alone.
How much does a FLUX 3 Image generation cost?
On this site, a generation starts at 6 points and the price rises with resolution. The exact cost for your chosen settings is shown before you generate. Black Forest Labs publishes no price sheet for FLUX 3 Image, so the point price here is this site’s own.
Is this the official Black Forest Labs website?
No. This is an independent multi-model workspace that provides a FLUX 3 Image workflow through the model service it has integrated. It is not affiliated with or endorsed by Black Forest Labs.
Create next
Open the FLUX 3 Image workspace Start creating with FLUX 3 Image
Choose a ratio and resolution, add references if the subject has to stay consistent, and check the point cost before you generate.