live

MODEL IDbfl:flux@3-image-klein

FLUX 3 Image Klein

by October 15, 2026

FLUX 3 Image Klein is the distilled checkpoint of Black Forest Labs' open-weights FLUX 3 image model, covering text-to-image generation and reference-based editing in a single model. It is built for speed without giving up quality, which suits interactive applications, rapid iteration, and high-volume production.

Multi-reference composition

How to combine up to four references with FLUX 3 Image Klein: assigning roles, keeping a product across scenes, placing it in your own photo, and borrowing a style.

Introduction

FLUX 3 Image Klein takes up to four reference images in inputs.referenceImages and combines them into one new picture. A person, a garment, a bag and a location that were photographed on different days come back as a single frame, because the prompt tells the model which image supplies which part of the result.

A bearded man walking down a red brick cobblestone lane in an olive field jacket with a corduroy collar over a white t-shirt and gray chinos, a tan leather messenger bag on his shoulder

Built from 4 references

Image 1, the model
A full-length studio photo of a man in his late twenties with short dark hair and a trimmed beard, standing straight and facing the camera, wearing a plain white t-shirt, light gray chinos and white sneakers. Seamless light gray backdrop, soft even studio light. Photoreal e-commerce model photography, 50mm lens, neutral palette.
Image 2, the jacket
A flat packshot of an olive green waxed cotton field jacket with a brown corduroy collar, four flap pockets and brass snap buttons, laid flat on a seamless white background, shot from directly above. Even shadowless studio light. Photoreal apparel e-commerce packshot.
Image 3, the bag
A packshot of a tan leather messenger bag with a flap closure, two brass buckles and a wide canvas shoulder strap, standing upright on a seamless white background. Even soft studio light, a faint contact shadow. Photoreal accessories e-commerce packshot.
Image 4, the location
An empty cobblestone lane in an old European city on an overcast autumn morning, red brick townhouses with black doors on both sides, a few fallen yellow leaves on the stones. Soft flat daylight. Straight-on shot at standing height, 35mm lens, no people. Photoreal street photography, muted autumn palette.

This guide covers how to assign a role to each reference, how to keep one product consistent across several scenes, how to place a product into a photo you already have, and how to take the style of one image and apply it to new content.

Giving each reference a role

The references are numbered in the order you send them, starting at image 1. Refer to each one by that number and say what the model should take from it, such as the face from one image and the garment from another. An image the prompt never mentions has no job, and its influence on the result is left to chance.

Try in Playground

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'bfl:flux@3-image-klein',
  positivePrompt: 'The man from image 1, with the same face, hair and beard, walking toward the camera down the cobblestone lane from image 4, wearing the olive field jacket from image 2 open over his white t-shirt with his light gray chinos and white sneakers, and carrying the tan leather messenger bag from image 3 on his right shoulder. Full-length shot at standing height, 50mm lens, the lane receding behind him. Soft overcast daylight. Photoreal fashion lookbook photography, muted autumn palette.',
  inputs: {
    referenceImages: [
      'https://im.runware.ai/image/os/a14d18/ws/2/ii/cf75071b-232d-405e-8cad-0547652a8fd4.jpg',
      'https://im.runware.ai/image/os/a14d18/ws/2/ii/1298242a-8c16-47e9-a7ad-9627943cf030.jpg',
      'https://im.runware.ai/image/os/a14d18/ws/2/ii/8baad788-639c-446b-a6a1-4f586404bc31.jpg',
      'https://im.runware.ai/image/os/a14d18/ws/2/ii/94348d04-6be4-46a9-b379-017c58a6e2d7.jpg'
    ]
  },
  width: 832,
  height: 1248
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "bfl:flux@3-image-klein",
            "positivePrompt": "The man from image 1, with the same face, hair and beard, walking toward the camera down the cobblestone lane from image 4, wearing the olive field jacket from image 2 open over his white t-shirt with his light gray chinos and white sneakers, and carrying the tan leather messenger bag from image 3 on his right shoulder. Full-length shot at standing height, 50mm lens, the lane receding behind him. Soft overcast daylight. Photoreal fashion lookbook photography, muted autumn palette.",
            "inputs": {
                "referenceImages": [
                    "https://im.runware.ai/image/os/a14d18/ws/2/ii/cf75071b-232d-405e-8cad-0547652a8fd4.jpg",
                    "https://im.runware.ai/image/os/a14d18/ws/2/ii/1298242a-8c16-47e9-a7ad-9627943cf030.jpg",
                    "https://im.runware.ai/image/os/a14d18/ws/2/ii/8baad788-639c-446b-a6a1-4f586404bc31.jpg",
                    "https://im.runware.ai/image/os/a14d18/ws/2/ii/94348d04-6be4-46a9-b379-017c58a6e2d7.jpg"
                ]
            },
            "width": 832,
            "height": 1248
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "imageInference",
      "taskUUID": "2c3d4e5f-6a7b-4c8d-9e0f-1a2b3c4d5e6f",
      "model": "bfl:flux@3-image-klein",
      "positivePrompt": "The man from image 1, with the same face, hair and beard, walking toward the camera down the cobblestone lane from image 4, wearing the olive field jacket from image 2 open over his white t-shirt with his light gray chinos and white sneakers, and carrying the tan leather messenger bag from image 3 on his right shoulder. Full-length shot at standing height, 50mm lens, the lane receding behind him. Soft overcast daylight. Photoreal fashion lookbook photography, muted autumn palette.",
      "inputs": {
        "referenceImages": [
          "https://im.runware.ai/image/os/a14d18/ws/2/ii/cf75071b-232d-405e-8cad-0547652a8fd4.jpg",
          "https://im.runware.ai/image/os/a14d18/ws/2/ii/1298242a-8c16-47e9-a7ad-9627943cf030.jpg",
          "https://im.runware.ai/image/os/a14d18/ws/2/ii/8baad788-639c-446b-a6a1-4f586404bc31.jpg",
          "https://im.runware.ai/image/os/a14d18/ws/2/ii/94348d04-6be4-46a9-b379-017c58a6e2d7.jpg"
        ]
      },
      "width": 832,
      "height": 1248
    }
  ]'

runware run bfl:flux@3-image-klein \
  positivePrompt="The man from image 1, with the same face, hair and beard, walking toward the camera down the cobblestone lane from image 4, wearing the olive field jacket from image 2 open over his white t-shirt with his light gray chinos and white sneakers, and carrying the tan leather messenger bag from image 3 on his right shoulder. Full-length shot at standing height, 50mm lens, the lane receding behind him. Soft overcast daylight. Photoreal fashion lookbook photography, muted autumn palette." \
  inputs.referenceImages.0=https://im.runware.ai/image/os/a14d18/ws/2/ii/cf75071b-232d-405e-8cad-0547652a8fd4.jpg \
  inputs.referenceImages.1=https://im.runware.ai/image/os/a14d18/ws/2/ii/1298242a-8c16-47e9-a7ad-9627943cf030.jpg \
  inputs.referenceImages.2=https://im.runware.ai/image/os/a14d18/ws/2/ii/8baad788-639c-446b-a6a1-4f586404bc31.jpg \
  inputs.referenceImages.3=https://im.runware.ai/image/os/a14d18/ws/2/ii/94348d04-6be4-46a9-b379-017c58a6e2d7.jpg \
  width=832 \
  height=1248

{
  "taskType": "imageInference",
  "taskUUID": "2c3d4e5f-6a7b-4c8d-9e0f-1a2b3c4d5e6f",
  "model": "bfl:flux@3-image-klein",
  "positivePrompt": "The man from image 1, with the same face, hair and beard, walking toward the camera down the cobblestone lane from image 4, wearing the olive field jacket from image 2 open over his white t-shirt with his light gray chinos and white sneakers, and carrying the tan leather messenger bag from image 3 on his right shoulder. Full-length shot at standing height, 50mm lens, the lane receding behind him. Soft overcast daylight. Photoreal fashion lookbook photography, muted autumn palette.",
  "inputs": {
    "referenceImages": [
      "https://im.runware.ai/image/os/a14d18/ws/2/ii/cf75071b-232d-405e-8cad-0547652a8fd4.jpg",
      "https://im.runware.ai/image/os/a14d18/ws/2/ii/1298242a-8c16-47e9-a7ad-9627943cf030.jpg",
      "https://im.runware.ai/image/os/a14d18/ws/2/ii/8baad788-639c-446b-a6a1-4f586404bc31.jpg",
      "https://im.runware.ai/image/os/a14d18/ws/2/ii/94348d04-6be4-46a9-b379-017c58a6e2d7.jpg"
    ]
  },
  "width": 832,
  "height": 1248
}

Response

{
  "data": [
    {
      "taskType": "imageInference",
      "taskUUID": "2c3d4e5f-6a7b-4c8d-9e0f-1a2b3c4d5e6f",
      "imageUUID": "7d8e9f0a-1b2c-4d3e-8f4a-5b6c7d8e9f0a",
      "imageURL": "https://im.runware.ai/image/os/a14d18/ws/2/ii/7d8e9f0a-1b2c-4d3e-8f4a-5b6c7d8e9f0a.jpg"
    }
  ]
}

The prompt also restates the details that identify each item, such as the corduroy collar and the tan leather. The jacket keeps its collar and its flap pockets in the result, and the bag keeps its tan leather and brass hardware. The fourth image sets the light as well as the place, so the studio-lit model comes back under the flat overcast sky of the lane.

Keeping a product across scenes

One packshot can feed a whole set of lifestyle images. Send the product as the only reference and write a full caption for each scene, with the product described the same way every time.

A table lamp with a ribbed terracotta gourd-shaped base and a pleated off-white linen shade on a white background — Reference

The terracotta lamp switched on, standing on a walnut console beside a key bowl and art books under a round mirror — Entryway, evening

The terracotta lamp switched off on a white bedside table next to a bed with rumpled sage linen, a glass of water and a book beside it — Bedside, morning

The terracotta lamp switched on at the back of an oak desk with an open laptop, a mug of pens and a linen notebook — Home office, daylight

The double-gourd base and the pleated shade match the packshot in all three rooms, and the lamp sits where each framing clause put it. Repeating the same product description word for word in every prompt is what holds it, since each request is independent and the description is the only thing the three share besides the reference.

Placing a product in your own photo

When the scene already exists, send it as a reference too and tell the model which image is the scene. The product goes into the photo as it stands, instead of the model building a new background around the product.

The lime green running shoe standing beside a white lane line on a red running track, backlit by a low morning sun, with its shadow falling across the track

Built from 2 references

Image 1, the shoe
A side-profile packshot at shoe height, the shoe centered and pointing left: a running shoe with a lime green knit upper, white laces, a thick white foam midsole and a black rubber outsole, standing on a seamless white background. Even soft studio light, a faint contact shadow. Photoreal footwear e-commerce packshot, clean white palette.
Image 2, the scene
A low shot at ground level along an empty outdoor running track, the white lane lines running from the lower left toward the horizon. Red-orange rubber surface, a strip of green infield on the right and trees beyond it. Low early-morning sun from behind the camera, long soft shadows on the track. Photoreal sports photography, 35mm lens, warm morning palette, no people.

The track and the low sun are unchanged from the scene photo, and the shoe keeps its perforated knit and its midsole. The line about the shadow matters: asking for a shadow that matches the scene is what grounds the product, rather than leaving it floating with the soft studio shadow of its packshot.

Borrowing a style from a reference

A reference can supply a look instead of an object. Send one image in the style you want, then describe new content and say it should be drawn in the style of image 1, naming the traits that make up that style.

A flat vector illustration of a red tram crossing a stone bridge in front of townhouses and orange mountains, in teal, orange, cream and navy with paper grain

Built from 1 reference

Image 1, the style
A flat vector travel poster illustration of a mountain lake with a small wooden boathouse and pine trees, built from bold flat shapes with no outlines, a limited palette of teal, burnt orange, cream and deep navy, subtle paper grain texture, long soft geometric shadows. Mid-century modern travel poster style, no lettering.

The tram and the bridge are new, and the palette, the outline-free shapes, the long shadows and the paper grain all come from the lake poster. Naming the traits in the prompt keeps the model on the treatment rather than the subject of the reference. This is how an illustrated campaign or a series of app illustrations stays in one house style.

Tips for best results

Refer to every image by its number. Image 1 is the first in inputs.referenceImages. A reference the prompt never mentions has no defined job.
Say what to take from each image. "The face from image 1" and "the jacket from image 2" are instructions. "Use these images" is not.
Restate the identifying details. Name the collar or the base shape in the prompt as well as showing it, and those details hold more closely.
Name the scene image when you have one. "Use image 2 as the scene" keeps your photo and adds the product to it.
Ask for a matching shadow. A product dropped into a photo looks pasted on until its shadow follows the light of the scene.
Keep to four references. Requests with more than four are rejected, so combine items into one packshot when a look needs more pieces.