FLUX.3 Image

byBlack Forest Labs

Posters, packaging and hero art with type that reads like real print, plus edits that change one element and leave the rest alone

Set the type
A man in a cream suit on a red deckchair at the bottom of a drained blue swimming pool, reading a newspaper headlined NOTHING HAPPENED TODAY

How FLUX.3 Image works

Write in plain directorial sentences, choose the shape you will deliver in, and attach references only when you want an edit or a composite. Scale up to 2K or 4K once the layout works.

FLUX.3 Image node on the Fuser canvas with a text prompt wired in and its output: a sheet cake iced SORRY ABOUT THE GARAGE

Describe the frame plainly

Write direct sentences covering subject, setting, framing and light, and put any words you want rendered in quotes.

FLUX.3 Image node settings showing 4K resolution and 1:1 aspect ratio above a sheet cake iced SORRY ABOUT THE GARAGE

Pick shape and add references

Set the aspect ratio for your delivery format, or connect up to 10 images and name each one by order, like image 1 and image 2.

FLUX.3 Image node showing its 4K output: a 1979 kitchen snapshot of a sheet cake iced SORRY ABOUT THE GARAGE

Draft small, finish large

Test composition at 512px to 1K, then regenerate at 2K or 4K in PNG for the final.

What FLUX.3 Image is good at

Layout discipline, readable words and edits that respect what you didn't ask to change, across tiers from 512px drafts to 4K finals.

Gold-leaf lettering reading MAISON VERNET, Chapeaux sur mesure, on a Paris milliner's window, a black felt hat on a wooden head form behind the glass

Typography that sits in the scene

Headlines, signage and date lines render as part of the picture, like print or painted lettering rather than a pasted layer. Quote the exact words and say where they sit. Dense paragraphs can still drift, so keep copy short.

Reference photos of a white lifeguard chair, a terrazzo living room and a swimmer, composited into the swimmer on the chair watching over the empty room

Multi-reference compositing

Switch to the edit-image variant and blend a product, a surface and a backdrop from up to 10 references into one balanced frame. Give every image a role by order, because unclear roles cause blending.

A dachshund in a velvet armchair shown twice, first in a mustard knitted sweater and then in a blue trench coat, everything else unchanged

Local edits that keep the rest steady

Replace a garment, restyle one object or remove an element while color, stitching and geometry elsewhere hold still. Name the element and add "keep everything else identical".

Macro of a couturier's thimbled hand stitching seed pearls and glass bugle beads onto ivory silk organza in an embroidery hoop

Print-ready 4K detail

Native 2K and 4K tiers keep faces, texture and small type crisp when cropped. Use the text-to-image variant for prompt-only hero frames, and save 4K for finals since it costs and takes the most.

Made with FLUX.3 Image

A spread of poster, campaign, film, architecture and product frames, each built around one subject, a tight palette and quoted text where the scene calls for it.

A giant fingertip pressing the doorbell of a white card architectural model house with a tiny yellow front door, lit by a desk lamp

Architectural visualization with a single yellow door

A man in a grey coat leaps after loose pages caught mid-air along a concrete arcade, papers frozen in flight

Film still with arrested motion

Street poster headlined WARM BY THE WATER above a folded plum wool blanket, a woman in a camel coat with a yellow umbrella walking past

Campaign frame with integrated headline

A young man leaps from a harbour wall at dusk, a yellow towel flying overhead against a violet sky

Vertical short-form moment

A polished chrome kettle with a curl of steam sitting on a field of ripe purple plums

Product still life on a surreal set

What people build with FLUX.3 Image

Designers, brand teams and stylists reach for it when a frame has to carry readable type, survive a crop, or change one element without a full reshoot.

Poster and album art designers

01

Build gig posters, record sleeves and typographic covers where the headline has to read cleanly and the layout is composed for 2:3, 1:2 or square from the start.

Packaging and signage mockups

02

Test label wording, shop signs and shelf-ready packaging at 2K or 4K, with quoted text rendering as part of the scene instead of an overlay.

E-commerce teams swapping products

03

Combine a product, a plinth and a backdrop from separate references in the edit-image variant, and restyle one object while the rest of the frame stays identical.

Fashion and lookbook stylists

04

Swap a single garment on a base photo and keep the face, pose, background and lighting where they are, then run the same base through other pieces.

Brand campaign art directors

05

Produce wide banners, tall stories and print-ready hero images in the delivery ratio you need, with clean composition that holds up when cropped.

FLUX.3 Image specs

Plans and credits
Inputs
Prompt (text), Images (up to 10 images)
Output
image
Resolution
512px, 768px, 1K, 2K, 4K
Aspect ratios
21:9, 2:1, 16:9, 3:2, 7:5, 4:3, 5:4, 1:1, 4:5, 3:4, 5:7, 2:3, 9:16, 1:2
Cost
57–836 credits per run

FLUX.3 Image questions

Text-to-image generates from a prompt alone, while edit-image works from connected reference images. Text-to-image suits posters, typographic layouts, concept frames and hero images, but has no reference control, so consistency depends on your wording. Edit-image replaces a garment, restyles an object, removes an element or blends references into a composite, but needs clear references and prompts that assign roles by order.

Use the edit-image variant. Put the base photo first, name the garment image by order ("replace the jacket with the one from image 2"), and add "keep everything else identical". Unclear roles cause blending, and high-resolution edits are slower and costlier.

Up to 10. Order matters because the prompt addresses them as image 1, image 2 and so on. When aspect ratio is Auto, the first image sets the framing, so put the base image first.

Yes, short headlines and signage render well when you put the wording in quotes and say where it sits and what it is on. Long or dense passages can still drift or misspell, and text at 512px or 768px loses legibility, so use 2K or 4K for text-heavy layouts.

Skip it for rapid throwaway ideation at 4K, since higher tiers cost far more and can take minutes. It is also a poor fit for pure vector or icon output, for guaranteed face-lock across a long series, and for keyword-stacked prompts like "masterpiece, 8K", which flatten the look.

It is made by Black Forest Labs, announced in July 2026 and reaching image generation in early October 2026. An open-weight Dev image variant is reported as announced but not yet released, and the version here is hosted only.

Try FLUX.3 Image on Fuser

Posters, packaging and hero art with type that reads like real print, plus edits that change one element and leave the rest alone

Set the type

Related models

FLUX.3 Image docsAll models

FLUX.2

Black Forest Labs

FLUX.1 Kontext

Black Forest Labs

FLUX.1 [Dev]

Black Forest Labs

FLUX1.1 [Pro] Ultra

Black Forest Labs

Gemini Image

Google DeepMind

GPT Image

OpenAI

FLUX.3 Image in our guides

All articles