FLUX 3 Image Klein

About

FLUX 3 Image Klein is the fast, open-weights version of FLUX 3 Image from Black Forest Labs, the team behind FLUX, released on October 15, 2026. One 5.6-billion-parameter model creates images from text, edits photos, combines up to ten reference images and sets the words you quote as text in the image.

It is built for speed. On an RTX 5090, a 1024 × 1024 image took 1.9 s, against 3.7 s for Krea 2 Turbo, and a one-photo edit took 3.6 s, against 8.4 s for Qwen Image Edit 2511 Lightning and 9.7 s for Krea 2 Identity Edit. Larger models such as Krea 2 and Qwen Image still win on some images.

On Sogni it runs on the Supernet, a network of community GPUs: use it in the browser or the app, or call it through the API, with nothing to download. Text-to-image renders at any size from 256 to 4096 px per side, including 4K, from 1.80 Spark per image.

What FLUX 3 Image Klein can do

One render uses either a guide image or reference images, not both.

How to use FLUX 3 Image Klein on Sogni

• Create from text — Open the image creator with FLUX 3 Image Klein selected, or pick it under Next-Gen Diffusion in the model picker. 16 steps is the default; 4 works for quick drafts.
• Edit a photo — Choose FLUX 3 Image Klein from Image Editing on Home, add the photo and describe the change, such as “tie a red bandana around her neck”. Say what should stay the same.
• Edit by box — With one photo attached, draw a box where something should be added, moved, resized or removed. See how it works.
• Combine several images — Add up to ten reference images and refer to them by order: “print the goldfish pattern from the second image all over the green dress in the first image”.
• Build with it — Call flux3_image_klein_sd_int8_convrot through the Sogni API and SDKs, or flux-3-klein on the creative tools generate_image and edit_image. Code is in the API section and the FLUX 3 Image Klein guide.

The prompts for the edits at the top of the page:

Put a glowing camping lantern on the ground next to the bicycle's back wheel. Keep everything else unchanged.

In <ref_image_0>, add a potted olive tree <tree_1> on the wet floor between the sofas. Keep the rest of the image completely unchanged.

[{"id":"tree_1","from":null,"src_bbox":null,"tgt_bbox":[400,300,880,450],"desc":"a potted olive tree in a terracotta pot"}]

Print the goldfish pattern from the second image all over the green dress in the first image, as a fabric print. Keep the dress shape, the dress form and the background unchanged.

Edit by box

Edit by box lets you point at the part of a photo you want changed. In the Sogni web app, choose FLUX 3 Image Klein, attach one photo and turn on Edit by box. Draw a box, pick Add object, Move or resize or Remove object, and name the object. Klein changes what you marked and keeps the rest of the photo.

A real edit: an olive tree in a drawn box

• One edit per render is the most reliable — You can draw up to three; with several, one can be missed.
• Boxes set place and size — They are not exact cutouts.
• Give new objects room — A very small box often stays empty.

Ten reference images in one scene

Klein takes up to ten reference images in one render. We made ten product photos with Krea 2 Turbo and asked Klein to build one reading nook from all of them. Every object made it into the room, in 36.6 s on an RTX 5090.

A cozy reading nook on a sunny afternoon, warm light from a tall window. The red ceramic teapot from the first image sits on a small wooden side table beside a worn leather armchair. The brass desk lamp from the second image stands on the table behind the teapot. The potted monstera from the third image fills the corner by the window. The vintage film camera from the fourth image rests on the armrest of the chair. The stack of hardcover books from the fifth image lies on the floor beside the chair. The ginger tabby cat from the sixth image naps curled up on the armchair cushion. The wicker basket of yarn from the seventh image sits on the floor by the window. The brass alarm clock from the eighth image stands on the windowsill. The vase of sunflowers from the ninth image stands on the side table beside the lamp. The red plaid blanket from the tenth image is draped over the back of the armchair. Natural colors, photograph.

The same scene with one to nine references, and the side-by-side tests against Krea 2 Identity Edit and Qwen Image Edit 2511 Lightning, are in the full comparison on the Sogni blog.

Prompting tips

• Describe the whole scene — Subject, framing, setting, light and style, in full sentences. In the web app, the Prompt Enhancer turns a short prompt into a full description first; turn off Auto under the Enhance button to send your words as typed.
• Put exact text in quotes — Say where the words go and how they look: ‘THE LOW TIDES’ very large at the top, in black wood type. End with “nothing else written” to keep stray lettering out.
• Name references by order — The first image, the second image and so on, and say what to take from each.
• Say what stays the same — In an edit, name what must not change, such as the background, the light and the framing.
• Ask for what you want — A negative prompt has no effect, so ask for “a clean, empty desk” rather than “no clutter”.

Layout boxes in a prompt

Edit by box writes boxes into the prompt for you. You can also write them yourself, in a text-to-image prompt or through the API: tag each element with an id in angle brackets, such as <obj_1>, and end the prompt with a list giving each id a box as [top, left, bottom, right] on a 0–1000 grid and a short description. The balloons in the gallery were placed this way:

Full-color outdoor photograph at sunrise, with a broad dawn sky <sky_1> above a green valley <valley_1>. On the left, a large hot-air balloon <obj_1> floats over the valley, while a smaller blue hot-air balloon <obj_2> hangs farther back on the right. Along the horizon, the rising sun <sun_1> appears between distant mountain ridges <mountain_1>, casting low orange light across the slopes and leaving the foreground meadow in cooler shadow. [{"id": "obj_1", "bbox": [80, 60, 480, 380], "desc": "A large red and yellow striped hot-air balloon with alternating vertical fabric panels, a rounded envelope, dark suspension cables, and a small rectangular wicker basket hanging below; the lower edge of the envelope is partly shaded by the sunrise angle."}, {"id": "obj_2", "bbox": [200, 620, 420, 800], "desc": "A small blue hot-air balloon with a rounded cobalt-blue envelope, narrow darker seams, fine suspension lines, and a tiny brown basket."}, {"id": "sky_1", "bbox": [0, 0, 650, 1000], "desc": "A clear sunrise sky grading from pale orange near the horizon through peach to muted blue overhead, with several thin horizontal clouds catching warm light."}, {"id": "sun_1", "bbox": [470, 455, 535, 525], "desc": "A low circular sunrise disk in bright yellow-orange, partially veiled by the distant ridge line."}, {"id": "mountain_1", "bbox": [430, 0, 690, 1000], "desc": "Layered distant mountain ridges forming an uneven horizon, with blue-gray silhouettes in front and warmer brown-green slopes along their upper edges."}, {"id": "valley_1", "bbox": [560, 0, 1000, 1000], "desc": "A broad green valley with rolling grassy slopes, scattered darker tree masses, and a winding river reflecting a narrow strip of orange morning light; the nearest meadow contains patches of deep green shadow and uneven vegetation."}]

Speed and price against the other fast open image models on Sogni

Time for one 1024 × 1024 image on an RTX 5090 with the model loaded, and its price, at each model’s default settings.

Spark per 1024 × 1024 image at each model’s default steps (1 Spark = $0.005); edit models with one reference image. Sizes are parameter counts as each maker publishes them. Sogni’s Unlimited plans cover all of these under fair use.

Photo edit speed

1-megapixel edits on one RTX 5090 with each model loaded. In the same test, Klein was the only one of the three to change a line of printed text correctly. Every image is in the full comparison on the Sogni blog.

Which one to pick

• Speed — Klein was the fastest model in this test: about 1.9 s per image and 3.6 s per photo edit.
• One model for creating and editing — Draft from text and revise with edits without switching models.
• More references — Klein takes up to ten. Qwen Image Edit 2511 Lightning takes three and Krea 2 Identity Edit two.
• LoRAs and personas — Krea 2 Turbo has Sogni’s LoRA catalog and persona LoRAs.
• The most detail — Krea 2 (12 billion parameters) and Qwen Image (20 billion) are larger than Klein (5.6 billion) and were ahead on some images in our tests.

From Black Forest Labs

Black Forest Labs calls FLUX 3 Image Klein “the most capable open weights image model” and describes it as “frontier quality with the speed and customization of open weights”. In its launch post, Klein ranks first among open-weights models in Black Forest Labs’ own evaluations for text-to-image, multi-reference generation and image editing, and runs on consumer GPUs down to an RTX 3090.

• What it makes — photography with “detailed skin, fabric, weather, and light”; illustration from woodblock and mosaic to clay, pixel art and anime; and type and design.
• Two versions — the distilled Klein (16 steps), which Sogni runs, and the undistilled Klein Base (50 steps) for fine-tuning, LoRA training and research.
• Your own LoRAs — Sogni Unlimited members can import Klein LoRAs from Civitai or Hugging Face in My LoRAs.

Black Forest Labs’ launch post, FLUX 3 Image Klein page, documentation, open weights on Hugging Face and FLUX Kommunity License.

FLUX 3 Image Klein vs FLUX.2 [klein]

FLUX.2 [klein] is Black Forest Labs’ earlier compact family, released in January 2026 in 4B and 9B sizes. FLUX 3 Image Klein is a newer, different model. In Black Forest Labs’ evaluations it scores 91 Elo higher than FLUX.2 [klein] 9B in text-to-image and 70 higher in image editing, with 38% fewer parameters. The licenses differ too: FLUX.2 [klein] 4B is Apache 2.0 and 9B is non-commercial; FLUX 3 Image Klein uses the FLUX Kommunity License.

Pricing

Use pay-as-you-go Spark packs for each render (1 Spark = $0.005), or choose a flat-rate Sogni plan for credit-free fair-use generation in the app.

Text-to-image costs 1.80 Spark ($0.009) per image at the default 16 steps, up to about 1.5 megapixels; larger images scale with pixels, and fewer steps cost less. A one-photo edit runs at the photo’s own size, up to 4 megapixels: 2.19 Spark at 1 MP, 5.35 at 2 MP and 14.25 at 4 MP.

On an RTX 5090, a 3840 × 2160 image took 27 s and a 4096 × 4096 image 94 s. A one-photo edit took 3.6 s at 1 MP, 8.3 s at 2 MP and 22.1 s at 4 MP.

API

One Sogni API key reaches every model on the Supernet — call FLUX 3 Image Klein with the exact model id.

import { readFile } from 'node:fs/promises';
import { SogniClient } from '@sogni-ai/sogni-client';

const client = await SogniClient.createInstance({
  appId: crypto.randomUUID(),
  apiKey: process.env.SOGNI_API_KEY,
  network: 'fast',
});

// Text-to-image: leave contextImages out. Edits: add one to ten references.
const project = await client.projects.create({
  type: 'image',
  modelId: 'flux3_image_klein_sd_int8_convrot',
  positivePrompt:
    'Print the goldfish pattern from the second image all over the green dress in the first image, ' +
    'as a fabric print. Keep the dress shape, the dress form and the background unchanged.',
  contextImages: [await readFile('gown.jpg'), await readFile('goldfish.jpg')],
  numberOfMedia: 1,
});

const [url] = await project.waitForCompletion();
console.log(url); // result link — download within 24h

curl https://api.sogni.ai/v1/creative-agent/workflows \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $SOGNI_API_KEY" \
  -d '{
    "input": {
      "title": "First render",
      "steps": [{
        "id": "step1",
        "toolName": "generate_image",
        "arguments": { "prompt": "Photograph of a white ceramic mug of coffee on a worn wooden table by a window, morning sun from the right casting a long shadow, shallow depth of field", "model": "flux-3-klein" }
      }]
    },
    "confirm_cost": true
  }'

The JavaScript example edits with two references; leave contextImages out for text-to-image. The curl example uses the creative tools workflow alias flux-3-klein, which edit_image also accepts. Settings and limits are in the FLUX 3 Image Klein guide. Full reference at docs.sogni.ai.