FLUX 3 Generator

Edit a photo from one instruction, combine up to 10 references or lay out a poster with bounding boxes, with FLUX 3 Image Klein from Black Forest Labs. Pictures up to 4K, online in your browser, from 1 credit an image.

Make a Flux AI image online

Model

Prompt

0/5,000

Format

Aspect ratio

Size

2K · 1:1 · No watermark

A woman in a yellow raincoat cycling down a rain-soaked Tokyo alley at blue hour, lanterns reflected on the wet street, made with Black Forest Labs’ Klein modelExample

What is FLUX 3?

FLUX 3 is the newest model family from Black Forest Labs (BFL), the German lab behind the Flux image models. One model learns from images, video and audio together. BFL opened FLUX 3 in Early Access on July 23, 2026, and has since released FLUX 3 Video (August 4), FLUX 3 Image (October 1) and FLUX 3 Image Klein (October 15, 2026). On Melies AI you can use FLUX 3 Image Klein in your browser, with no GPU, install or API key.

FLUX 3 Image Klein is the open-weights image model of the family, with 5.6 billion parameters. One model does three jobs: text to image, editing a photo from an instruction, and building a new picture from up to 10 reference images. BFL says Klein shares the core abilities of the full FLUX 3 Image model: native 4K output, edits that leave the rest of the picture untouched, and bounding boxes that place each element where you want it.

In BFL’s own blind evaluations, Klein ranked first among open-weights models for text to image, multi-reference pictures and editing. It came ahead of the larger Qwen-Image 2.1 (7B) and Krea 2 (12.9B), and above the closed Ideogram V4.5 and FLUX.2 [pro]. Against FLUX.2 [klein] 9B it gained 91 Elo in text to image and 70 in editing, with 38% fewer parameters. These are BFL’s tests, not independent ones.

On this site you pick one of 9 shapes (1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, 9:21; an edit can also use Auto, which keeps the first image’s shape) and 1K, 2K or 4K, or type an exact width and height from 256 to 4,096 px. A 1K square is 1024 × 1024 pixels and a 4K 21:9 banner is 4096 × 1760. Prompts run up to 5,000 characters. Save as PNG, JPEG or WebP. No visible watermark is added.

Parameters
5.6B
Reference images
10
Pixels · longest side
4,096
Aspect ratios + Auto
9
Prompt characters
5,000
Credit · one 1K or 2K image
1

By Black Forest Labs · run through a third-party API · not affiliated with Black Forest Labs · Last updated

What Black Forest Labs’ Klein model does well

Five things the Klein model is built for, from one-line photo edits to layouts you place yourself.

Before and after: a sunny cobblestone street with a red bicycle, then the same street after a light snowfall, edited from one instruction with the Flux image editor
01

Edit a photo from one instruction

BFL built Klein for editing first. Upload a photo as @Image1 and say what to change: turn a summer street into winter, recolour a cup, add a bandana to a dog. The parts you do not mention stay as they were. Here one instruction turned a sunny street into a snowy one, and the houses, café tables, red bicycle and camera angle stayed the same.

Three product photos, a teal vase, a rattan armchair and a brass floor lamp, combined by the Flux AI image generator into one bright living room
02

Combine up to 10 references

Add up to 10 pictures (people, products, a room, a style) and describe the new scene. Name each one @Image1, @Image2 and so on, in the order you added them. Reference images anchor the content, while the prompt decides the composition, so say where each thing goes. Here three product shots became one furnished room.

Bounding boxes for a title, sun, pagoda, trees and date on a 0–1000 grid, next to the finished KYOTO AUTUMN 1962 travel poster with each element in its box
03

Place every element with bounding boxes

Add a JSON list of boxes to the end of the prompt, one per element, on a 0 to 1000 grid as [top, left, bottom, right], and tag each element in the description. The model draws each thing inside its box. In this travel poster the title, sun, pagoda, trees and date each landed in their own box. Boxes guide placement and scale; they are not a hard mask.

A kraft coffee bag on an oak shelf with a printed label reading NORTHBOUND ROASTERS, ETHIOPIA · YIRGACHEFFE and LIGHT ROAST · 340 g, made with Black Forest Labs’ image model
04

Type and graphic design

Klein renders text inside the picture, so posters, labels, stamps and covers can carry their words. Put each line in quotes and say where it sits, or give it its own box. This coffee label came out with every line spelled right, down to “340 g”. BFL notes that rendered text can still come out wrong, so check spelling before you use a design.

The same red fox in a snowy birch forest three ways: a wildlife photo, a Japanese woodblock print and a claymation figure, made with the Klein model
05

Photos and styles up to 4K

BFL shows Klein on detailed skin, fabric, weather and light, and on styles from woodblock and mosaic to clay, pixel art and anime. Here one fox comes as a wildlife photo, a woodblock print and a clay figure. Pick 4K for print, up to 4,096 pixels on the long side, or 1K for quick drafts at the same 1-credit price as 2K.

Klein vs the other Flux models

All three come from Black Forest Labs. Klein is the one on this site: the open, smaller member of the FLUX 3 family, compared here with its full-size sibling and the FLUX.2 model BFL measures it against.

FLUX 3 Image KleinFLUX 3 ImageFLUX.2 [klein] 9B
ReleasedOctober 15, 2026October 1, 2026January 15, 2026
Parameters5.6BNot published9B
WeightsOpen, FLUX Kommunity LicenseNot public; BFL APIOpen, FLUX Non-Commercial License
JobsText to image, edits, up to 10 referencesText to image, edits, up to 10 referencesText to image, edits, multi-reference
Largest size4K4K (about 16 MP)—
Bounding boxesYes, in the promptYes, in the prompt—
In BFL’s tests+91 Elo text to image, +70 editing over FLUX.2 [klein] 9B—The baseline
On this siteYesNoNo

How to use the Flux AI image generator

  1. 01

    Pick a tab

    Text to image to start from words, or Image edit to start from up to 10 of your own pictures.

  2. 02

    Write the prompt

    Describe the subject, light, camera and style, and put any text in quotes. In an edit, name your images @Image1, @Image2 and say what to change and what to keep.

  3. 03

    Choose shape and size

    9 shapes, 1K to 4K or an exact size, and PNG, JPEG or WebP. The price is on the button: 1 credit up to 2K, 2 at 4K.

  4. 04

    Generate and refine

    Press Generate. The picture stays under “My images” for 14 days. To change it again, download it and add it to Image edit.

Flux image pricing per picture

One price per picture, set by its size: 1 credit at 1K or 2K and 2 at 4K. Edits from up to 10 images cost the same. An exact width and height is billed at the smallest size with at least as many pixels. Packs from $9.90, no subscription, failed images refunded.

Model1K2K4K
Klein1 credit1 credit2 credits

Flux AI FAQ

Something else? Email support@meliesai.com.

You can try FLUX 3 Image Klein for free here. New accounts that sign up with Google get 4 free credits, enough for 4 pictures at 1K or 2K. After that a Klein picture costs 1 credit up to 2K and 2 at 4K, from one-time packs starting at $9.90. The Klein weights are also free to download and run on your own GPU.

A Klein picture costs 1 credit at 1K or 2K and 2 credits at 4K. Edits from 1 to 10 images cost the same as text to image. The $9.90 pack has 100 credits, enough for 100 pictures at 2K or 50 at 4K. The exact price shows on the Generate button first.

Black Forest Labs opened FLUX 3 in Early Access on July 23, 2026. FLUX 3 Video followed as a preview on August 4, FLUX 3 Image on October 1 and FLUX 3 Image Klein, the open-weights image model, on October 15, 2026. FLUX 3 Action, a 7B model for robot control, came out in September 2026.

Partly. FLUX 3 Image Klein has open weights on Hugging Face, a base model for fine-tuning and a distilled model for fast use, under the FLUX Kommunity License. That license is free for research, non-commercial use and commercial use by people and companies under $5M in yearly revenue. It is not an OSI open-source license. FLUX 3 Image and FLUX 3 Video have no public weights.

FLUX 3 Image Klein is the smaller, open-weights version of FLUX 3 Image. Both take the same kind of request: a prompt, up to 10 reference images, an aspect ratio, a size up to 4K and bounding boxes. Klein has 5.6 billion parameters and runs on a consumer GPU. FLUX 3 Image is BFL’s full model, offered through BFL’s API. Only Klein is on this site.

In Black Forest Labs’ own blind tests, yes. FLUX 3 Image Klein gained 91 Elo in text to image and 70 in image editing over FLUX.2 [klein] 9B, with 38% fewer parameters, and rated above FLUX.2 [pro]. The license also changed: FLUX.2 [klein] 9B is non-commercial, while the Klein model of FLUX 3 allows commercial use under $5M revenue.

Not this one. Klein makes still pictures. Black Forest Labs has separate video models, FLUX 3 Video and the faster FLUX 3 Max, which are not on this site. To animate a Klein picture, download it and upload it as the first frame in one of the video models here, such as Seedance 2.5 or Kling 4.0.

Open the Image edit tab and add 1 to 10 photos. Describe the change in one sentence, naming photos @Image1, @Image2 in the order you added them, and say what must stay the same. Keep the shape on Auto to match @Image1, pick a size, then press Generate. Klein aims to leave everything you did not mention unchanged, but check the result: an edit can still touch nearby areas.

Klein is trained on long, full scene descriptions. Black Forest Labs says to always run a prompt rewriter, because a short prompt sent as typed gives generic pictures. The Rewrite prompt switch is on by default and expands your words into a full description first. Switch it off only when your prompt already describes the whole scene.

Write the scene, tag each element with a short id in angle brackets, then add a JSON list at the end of the prompt. Each entry has the id, a bbox of [top, left, bottom, right] on a 0 to 1000 grid, and a short description. The grid is the same for every aspect ratio. In our travel-poster test, the title, sun, pagoda, trees and date each landed inside their boxes.

Yes. The weights are on Hugging Face and run in ComfyUI or Hugging Face Diffusers. They take 11.2 GB in BF16 and 5.6 GB in FP8, and the INT8 build runs on NVIDIA GeForce RTX cards with 8 GB of VRAM. NVIDIA measured about 4 seconds per 1-megapixel picture on an RTX 5090. The recommended prompt rewriter is a separate 9B model.

Three sizes: 1K, 2K and 4K. A 1K square is 1024 × 1024 pixels, a 2K 9:16 picture is 1536 × 2736 and a 4K 21:9 banner is 4096 × 1760. The shapes are 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 21:9, 9:21, plus Auto in an edit. A custom width and height can be 256 to 4,096 px, between 1:2.5 and 2.5:1.

Yes, here it is. Black Forest Labs filters harmful content during training, and on this site every Klein prompt and uploaded image goes through our content check and our API provider’s safety checker. Explicit sexual content, real people without consent and other material against our content standards are refused. A refused picture counts as failed, and its credits are refunded.

Pick GPT Image 2.6 for transparent backgrounds, GIF or SVG output, or a painted mask over the exact area to change. Pick Nano Banana 2.1 for 8:1 panoramas and infographics. Pick Seedream 5.5 for transparent PNGs and repeated edits. Pick Klein for 1-credit pictures up to 2K, quick photo edits and layouts placed with boxes. All of them are in the same image generator.

No visible watermark or logo is added to Klein pictures made on Melies AI. Pick PNG for the best quality, or JPEG or WebP for a smaller file. Each picture stays under “My images” for 14 days, so download the ones you want to keep.

Yes. Pictures you generate here are yours to use, including in ads, client work and products, as long as your prompt and uploaded images do not infringe anyone else’s rights and follow our content standards. Edits of real people need their consent. The FLUX Kommunity License revenue limit applies when you run the Klein weights yourself.

No. Melies AI is an independent platform that gives you access to Black Forest Labs’ FLUX 3 Image Klein model through a third-party API provider. It is not affiliated with, endorsed by or sponsored by Black Forest Labs. BFL’s own site is bfl.ai, with its Playground and API.

Make your first Flux picture

Write a prompt or upload your photos, pick a shape and a size up to 4K, and download the result.

Create an image