Image model by Black Forest Labs

FLUX 3 Image

Put every element of an ad exactly where you want it by giving each one a box in the prompt. FLUX 3 Image also changes one thing in a picture while leaving the rest alone, and can look real places up on the web before it draws.

Make an image with FLUX 3

A VELO NORD e-bike ad: a graphite e-bike on a misty pine-forest road at dawn with the headline Ride further than the coffee and a NEW badge — The prompt, boxes included
A finished print ad for an invented Scandinavian e-bike brand, photographed at dawn on an empty pine-forest road with mist between the trees and low golden sun from the right. Lay it out with these regions, given as [top, left, bottom, right] on a 0 to 1000 grid: [80, 60, 230, 560]: the headline "Ride further than the coffee." in a bold, tight white sans-serif. [235, 60, 290, 520]: a small white line of copy, "VELO NORD · Model T2 · 120 km range". [300, 420, 940, 960]: the e-bike itself, side view, graphite-gray matte frame with a slim integrated battery, tan leather saddle and tan grips, black tires, a small round front light glowing. [70, 790, 230, 950]: a circular cream badge that reads "NEW" in small dark capitals. Keep the left lower area as open road and soft mist. Photorealistic, crisp product detail, editorial outdoor advertising look.
Made with FLUX 3 Image on Predis.ai on October 26, 2026, 1K, 16:9, grounding off.

Exact edits

Change the frame. Leave the forest.

We asked for one thing, a sand-colored frame, and compared the two images pixel by pixel. About 2.5% of the picture changed: the frame and battery, plus faint traces around the headline's letters.

The VELO NORD ad with the graphite frame

The same ad with the frame recolored to desert sand

A difference map: only the bike frame and the edges of the headline are highlighted

The edit prompt

Edit image 1. Change only the color of the e-bike's frame and battery from graphite gray to a warm desert sand color with a soft matte finish. Keep everything else exactly as it is: the saddle, grips, tires and light, the headline, the line of copy, the NEW badge, the road, the trees, the mist, the light and the framing.

Made with FLUX 3 Image on Predis.ai on October 26, 2026. The difference map is our own comparison of the two files, with every pixel that moved by more than 10% highlighted.

Grounding

Look it up first, or draw from memory

With grounding on, FLUX 3 searches the web and images before it draws, which Black Forest Labs says helps with real landmarks, products and people. Off, it uses your prompt alone and returns faster. Same prompt, both ways.

The Sagrada Família at golden hour reflected in a pond, made with grounding on — **Grounding on** Searched before drawing

The Sagrada Família at golden hour reflected in a pond, made with grounding off — **Grounding off** Prompt only

The prompt

Travel photograph of the Sagrada Família basilica in Barcelona at golden hour, seen from the pond in Plaça de Gaudí, with its tall spires, the Nativity facade detail and the reflection in the water. Warm low sun, a few palm trees, clear sky, 35mm lens, realistic architectural detail.

Both made with FLUX 3 Image on Predis.ai on October 26, 2026, 1K, 4:5. Grounding is on by default; switch it off for invented scenes.

Reference images

Three pictures, called by number

Name each reference by its number in the prompt and say what it's for. Here a room, a coat and a helmet became one still life. FLUX 3 takes up to 10.

image 1the setting
image 2the coat
image 3the helmet

An oilskin coat on a brass hook and a brass diving helmet on a chart table inside a teal submarine bridge

The prompt

Image 1 is the setting: the vintage submarine bridge with the brass periscope and teal riveted walls. Image 2 is the dark olive oilskin coat. Image 3 is the antique brass diving helmet. Compose one moody still-life photograph inside the bridge from image 1: the coat from image 2 hangs from a brass hook on the teal wall on the left, and the helmet from image 3 sits on a worn wooden chart table in the foreground on the right, beside a rolled sea chart and a small brass compass. Warm amber light from the caged lamps, teal shadows, shallow depth of field, no people, no text.

The three references were made with Nano Banana Pro for our Grok Imagine Video page. The still life was made with FLUX 3 Image on Predis.ai on October 26, 2026.

Spec sheet

What FLUX 3 Image offers in the Predis.ai Model Studio. For video from the same family, see FLUX 3 Video.

Made by: Black Forest Labs
Released: October 1, 2026
Starts from: Text, or up to 10 reference images
Sizes: 768px, 1K, 1.5K, 2K or 4K
Shapes: 15 aspect ratios from 21:9 to 9:21, or auto
Layout: Boxes in the prompt, [top, left, bottom, right] on a 0 to 1000 grid
Grounding: Web and image search, on by default
Per run: Up to 4 images
Open weights: Only for the smaller Klein version

Fine print

Boxes are strong hints, not walls

In our second layout we moved everything around. The bike and the badge landed in their new boxes, but the headline spilled left past its box and the line of copy sat higher than asked. Toggle the layout to see where we asked for each one.

The second VELO NORD layout: the bike on the left and the badge top left, with the headline running past its box — What changed in the prompt
Same brief, with the boxes moved: [300, 40, 940, 580]: the e-bike, on the left. [90, 600, 300, 960]: the headline, right-aligned. [310, 640, 360, 960]: the line of copy, right-aligned. [70, 50, 230, 210]: the NEW badge, top left.

Elements can drift

Text in particular can run past its box, as the headline above shows. Check the layout, and give long text a wider box or run it again.

Untouched usually means untouched

Edits leave most pixels alone, but not all. Our recolor also touched the edges of the headline.

4K takes longer

The largest sizes can take minutes. Work out the image at 1K, then make the final at 4K.

Big references get scaled down

Reference images above 4 megapixels are reduced before use, so very fine detail in them can soften.

FLUX 3 Image questions

What is FLUX 3 Image?

It's the image model in Black Forest Labs' FLUX 3 family, released on October 1, 2026. It makes images from text or from up to 10 reference images, lets you place elements with boxes in the prompt, and edits one element while leaving the rest of the picture alone.

How do layout boxes work in FLUX 3?

Add a line to your prompt for each element with its position as [top, left, bottom, right] on a grid from 0 to 1000, then say what goes there. The VELO NORD ad on this page used four boxes: the headline, a line of copy, the bike and a badge. Treat the boxes as strong hints, since elements can drift outside them.

What does grounding do in FLUX 3 Image?

With grounding on, FLUX 3 runs a web and image search before it draws, which Black Forest Labs says helps it show real places, products and people accurately. With it off, it works from your prompt alone and returns faster. On Predis.ai grounding is on by default and you can switch it off.

How many reference images can FLUX 3 Image use?

Up to 10 on Predis.ai. Refer to them by number in the prompt, such as "image 1 is the setting, image 2 is the coat", the way the still life on this page was made from 3.

Is FLUX 3 Image open source?

The full model isn't. Black Forest Labs released open weights for a smaller version, FLUX 3 Image Klein, on October 15, 2026, but the model on Predis.ai is the full FLUX 3 Image.

What sizes can it make on Predis.ai?

From 768 pixels up to 4K, in 15 aspect ratios from 21:9 to 9:21, or auto to let the model choose. You can make up to 4 images from one prompt.

Draw the layout in words

Give each element a box and FLUX 3 builds the ad around it.