FLUX quickstart

FLUX.2

Black Forest Labs has released FLUX.2 with support on Together AI. FLUX.2 is the next generation of image models, featuring enhanced control through JSON structured prompts, HEX color code support, reference image editing, and exceptional text rendering capabilities. Four model variants are available:

Model	Best for	Key features
FLUX.2 [max]	Ultimate quality	Highest fidelity output, best for premium use cases
FLUX.2 [pro]	Maximum quality	Up to 9 MP output, fastest generation
FLUX.2 [dev]	Development & iteration	Great balance of quality and flexibility
FLUX.2 [flex]	Maximum customization	Adjustable steps & guidance, better typography

Which model should I use?

Use [max] for the ultimate quality and fidelity in premium production workloads.
Use [pro] for production workloads requiring high quality and speed.
Use [dev] for development, experimentation, and when you need a balance of quality and control.
Use [flex] when you need maximum control over generation parameters or require exceptional typography.

Generating an image

Here’s how to generate images with FLUX.2:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="A mountain landscape at sunset with golden light reflecting on a calm lake",
    width=1024,
    height=768,
)

print(response.data[0].url)

import Together from "together-ai";

const together = new Together();

async function main() {
  const response = await together.images.generate({
    model: "black-forest-labs/FLUX.2-pro",
    prompt: "A mountain landscape at sunset with golden light reflecting on a calm lake",
    width: 1024,
    height: 768,
  });

  console.log(response.data[0].url);
}

main();

curl -X POST "https://api.together.ai/v1/images/generations" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "black-forest-labs/FLUX.2-pro",
    "prompt": "A mountain landscape at sunset with golden light reflecting on a calm lake",
    "width": 1024,
    "height": 768
  }'

Using FLUX.2 [dev] The dev variant offers a great balance for development and iteration:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-dev",
    prompt="A modern workspace with a laptop, coffee cup, and plants, natural lighting",
    width=1024,
    height=768,
    steps=20,
)

print(response.data[0].url)

import Together from "together-ai";

const together = new Together();

async function main() {
  const response = await together.images.generate({
    model: "black-forest-labs/FLUX.2-dev",
    prompt: "A modern workspace with a laptop, coffee cup, and plants, natural lighting",
    width: 1024,
    height: 768,
    steps: 20,
  });

  console.log(response.data[0].url);
}

main();

Using FLUX.2 [flex] The flex variant provides maximum customization with the guidance_scale and steps parameters. It also excels at typography and text rendering.

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-flex",
    prompt="A vintage coffee shop sign with elegant typography reading 'The Daily Grind' in art deco style",
    width=1024,
    height=768,
    steps=4,
    guidance_scale=3.5,
)

print(response.data[0].url)

import Together from "together-ai";

const together = new Together();

async function main() {
  const response = await together.images.generate({
    model: "black-forest-labs/FLUX.2-flex",
    prompt: "A vintage coffee shop sign with elegant typography reading 'The Daily Grind' in art deco style",
    width: 1024,
    height: 768,
    steps: 4,
    guidance_scale: 3.5,
  });

  console.log(response.data[0].url);
}

main();

Parameters

Common parameters (all models)

Parameter	Type	Description	Default
`prompt`	string	Text description of the image to generate	Required
`width`	integer	Image width in pixels (256-1920)	1024
`height`	integer	Image height in pixels (256-1920)	768
`seed`	integer	Seed for reproducibility	Random
`prompt_upsampling`	boolean	Automatically enhance prompt for better generation	true
`output_format`	string	Output format: `jpeg` or `png`	jpeg
`reference_images`	array	Reference image URL(s) for image-to-image editing	-

Additional parameters for [dev] and [flex] FLUX.2 [dev] and FLUX.2 [flex] support additional parameters:

Parameter	Type	Description	Default
`steps`	integer	Number of inference steps (higher = better quality, slower)	Model default
`guidance`	float	Guidance scale (higher values follow prompt more closely)	Model default

Image-to-image with reference images

FLUX.2 supports powerful image-to-image editing using the reference_images parameter. Pass one or more image URLs to guide generation. Core capabilities:

Capability	Description
Multi-reference editing	Use multiple images in a single edit
Sequential edits	Edit images iteratively
Color control	Specify exact colors using hex values or reference images
Image indexing	Reference specific images by number: “the jacket from image 2”
Natural language	Describe elements naturally: “the woman in the blue dress”

Single reference image Edit or transform a single input image:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="Replace the color of the car to blue",
    width=1024,
    height=768,
    reference_images=[
        "https://images.pexels.com/photos/3729464/pexels-photo-3729464.jpeg"
    ],
)

print(response.data[0].url)

import Together from "together-ai";

const together = new Together();

async function main() {
  const response = await together.images.generate({
    model: "black-forest-labs/FLUX.2-pro",
    prompt: "Replace the color of the car to blue",
    width: 1024,
    height: 768,
    reference_images: [
      "https://images.pexels.com/photos/3729464/pexels-photo-3729464.jpeg",
    ],
  });

  console.log(response.data[0].url);
}

main();

curl -X POST "https://api.together.ai/v1/images/generations" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "black-forest-labs/FLUX.2-pro",
    "prompt": "Replace the color of the car to blue",
    "width": 1024,
    "height": 768,
    "reference_images": ["https://images.pexels.com/photos/3729464/pexels-photo-3729464.jpeg"]
  }'

→

Multiple reference images Combine elements from multiple images. Reference them by index (image 1, image 2, etc.):

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="The person from image 1 is petting the cat from image 2, the bird from image 3 is next to them",
    width=1024,
    height=768,
    reference_images=[
        "https://t4.ftcdn.net/jpg/03/83/25/83/360_F_383258331_D8imaEMl8Q3lf7EKU2Pi78Cn0R7KkW9o.jpg",
        "https://cdn.pixabay.com/photo/2020/05/20/08/27/cat-5195431_1280.jpg",
        "https://images.unsplash.com/photo-1486365227551-f3f90034a57c",
    ],
)

print(response.data[0].url)

→

Using image indexing Reference specific images by their position in the array:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="Replace the top of the person from image 2 with the one from image 1",
    width=1024,
    height=768,
    reference_images=[
        "https://img.freepik.com/free-photo/designer-working-3d-model_23-2149371896.jpg",
        "https://img.freepik.com/free-photo/handsome-young-cheerful-man-with-arms-crossed_171337-1073.jpg",
    ],
)

print(response.data[0].url)

→

Using natural language FLUX.2 understands the content in your images, so you can describe elements naturally:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="""The man is leaning against the wall reading a newspaper with the title "FLUX.2"
The woman is walking past him, carrying one of the tote bags and wearing the black boots.
The focus is on their contrasting styles — her relaxed, creative vibe versus his formal look.""",
    width=1024,
    height=768,
    reference_images=[
        "https://img.freepik.com/free-photo/handsome-young-cheerful-man-with-arms-crossed_171337-1073.jpg",
        "https://plus.unsplash.com/premium_photo-1690407617542-2f210cf20d7e",
        "https://www.ariat.com/dw/image/v2/AAML_PRD/on/demandware.static/-/Sites-ARIAT/default/dw00f9b649/images/zoom/10016291_3-4_front.jpg",
        "https://i.pinimg.com/736x/dc/71/1c/dc711cc4c3ebafcd21f2a61efe8fd6cd.jpg",
    ],
)

print(response.data[0].url)

→

Color editing with reference images To change colors precisely, provide a color swatch image as a reference:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="Change the color of the gloves to the color of image 2",
    width=1024,
    height=768,
    reference_images=[
        "https://cdn.intellemo.ai/int-stock/62c6cc300a6a222fb36a2c8e/62c6cc320a6a222fb36a2c8f-v376/premium_boxing_gloves_from_top_brand_l.jpg",
        "https://shop.reformcph.com/cdn/shop/files/Blue_9da983c6-f823-4205-bca1-b3b8470657cf_grande.png",
    ],
)

print(response.data[0].url)

→

Best practices for reference images

Use image indexing: Reference images by number (“image 1”, “image 2”) for precise control.
Be descriptive: Clearly describe what you want to change or combine.
Use high-quality inputs: Better input images lead to better results.
Combine with HEX colors: Use specific color codes or color swatch images for precise color changes.

FLUX.1 Kontext editing

FLUX.1 Kontext is the previous generation’s editing family. It transforms a single reference image based on a text prompt, passed through the image_url parameter rather than reference_images. FLUX.2 handles reference editing natively, so prefer it for new work. Kontext remains available on serverless if you already build on it.

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.1-kontext-pro",
    width=1536,
    height=1024,
    steps=28,
    prompt="make his shirt yellow",
    image_url="https://github.com/nutlope.png",
)

print(response.data[0].url)

import Together from "together-ai";

const together = new Together();

async function main() {
  const response = await together.images.generate({
    model: "black-forest-labs/FLUX.1-kontext-pro",
    width: 1536,
    height: 1024,
    steps: 28,
    prompt: "make his shirt yellow",
    image_url: "https://github.com/nutlope.png",
  });

  console.log(response.data[0].url);
}

main();

curl -X POST "https://api.together.ai/v1/images/generations" \
     -H "Authorization: Bearer $TOGETHER_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "model": "black-forest-labs/FLUX.1-kontext-pro",
       "width": 1536,
       "height": 1024,
       "steps": 28,
       "prompt": "make his shirt yellow",
       "image_url": "https://github.com/nutlope.png"
     }'

A portrait edited by FLUX.1 Kontext to turn the subject's shirt yellow

Two Kontext models are available:

FLUX.1-kontext-pro: Best balance of speed and quality.
FLUX.1-kontext-max: Maximum image quality.

Kontext supports the following key parameters:

model: black-forest-labs/FLUX.1-kontext-pro or black-forest-labs/FLUX.1-kontext-max.
prompt: Text description of the transformation you want to apply.
image_url: URL of the reference image to transform.
aspect_ratio: Output aspect ratio (e.g., “1:1”, “16:9”, “9:16”, “4:3”, “3:2”). Alternatively, you can use width and height for precise pixel dimensions.
steps: Number of diffusion steps (default: 28, higher values may improve quality).
seed: Random seed for reproducible results.

Common uses include style transfer, object modification (changing colors or adding elements), scene transformation (daytime to nighttime, season changes), and character restyling. For complete parameter documentation, see the images overview.

JSON structured prompts

FLUX.2 is trained to understand structured JSON prompts, giving you precise control over subjects, composition, lighting, and camera settings. Basic JSON prompt structure

from together import Together

client = Together()

json_prompt = """{
  "scene": "Professional studio product photography setup",
  "subjects": [
    {
      "type": "coffee mug",
      "description": "Minimalist ceramic mug with steam rising from hot coffee",
      "pose": "Stationary on surface",
      "position": "foreground",
      "color_palette": ["matte black ceramic"]
    }
  ],
  "style": "Ultra-realistic product photography",
  "color_palette": ["matte black", "concrete gray", "soft white highlights"],
  "lighting": "Three-point softbox setup with soft, diffused highlights",
  "mood": "Clean, professional, minimalist",
  "background": "Polished concrete surface with studio backdrop",
  "composition": "rule of thirds",
  "camera": {
    "angle": "high angle",
    "distance": "medium shot",
    "focus": "sharp on subject",
    "lens": "85mm",
    "f-number": "f/5.6",
    "ISO": 200
  }
}"""

response = client.images.generate(
    model="black-forest-labs/FLUX.2-dev",  # Can also use FLUX.2-pro or FLUX.2-flex
    prompt=json_prompt,
    width=1024,
    height=768,
    steps=20,
)

print(response.data[0].url)

JSON schema reference Here’s the recommended schema for structured prompts:

{
  "scene": "Overall scene setting or location",
  "subjects": [
    {
      "type": "Type of subject (e.g., person, object)",
      "description": "Physical attributes, clothing, accessories",
      "pose": "Action or stance",
      "position": "foreground | midground | background"
    }
  ],
  "style": "Artistic rendering style",
  "color_palette": ["color 1", "color 2", "color 3"],
  "lighting": "Lighting condition and direction",
  "mood": "Emotional atmosphere",
  "background": "Background environment details",
  "composition": "rule of thirds | golden spiral | minimalist negative space | ...",
  "camera": {
    "angle": "eye level | low angle | bird's-eye | ...",
    "distance": "close-up | medium shot | wide shot | ...",
    "focus": "deep focus | selective focus | sharp on subject",
    "lens": "35mm | 50mm | 85mm | ...",
    "f-number": "f/2.8 | f/5.6 | ...",
    "ISO": 200
  },
  "effects": ["lens flare", "film grain", "soft bloom"]
}

Composition options

Option	Description
`rule of thirds`	Classic balanced composition
`golden spiral`	Fibonacci-based natural flow
`minimalist negative space`	Clean, spacious design
`diagonal energy`	Dynamic, action-oriented
`vanishing point center`	Depth and perspective focus
`triangular arrangement`	Stable, hierarchical layout

Camera angle options

Angle	Use case
`eye level`	Natural, relatable perspective
`low angle`	Heroic, powerful subjects
`bird's-eye`	Overview, patterns
`worm's-eye`	Dramatic, imposing
`over-the-shoulder`	Intimate, narrative

Hex color code prompting

FLUX.2 supports precise color control using HEX codes. Include the keyword “color” or “hex” followed by the code:

from together import Together

client = Together()

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="A modern living room with a velvet sofa in color #2E4057 and accent pillows in hex #E8AA14, minimalist design with warm lighting",
    width=1024,
    height=768,
)

print(response.data[0].url)

Gradient example

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="A ceramic vase on a table, the color is a gradient starting with #02eb3c and finishing with #edfa3c, modern minimalist interior",
    width=1024,
    height=768,
)

Advanced use cases

Infographics FLUX.2 can create complex, visually appealing infographics. Specify all data and content explicitly:

prompt = """Educational weather infographic titled 'WHY FREIBURG IS SO SUNNY' in bold navy letters at top on cream background, illustrated geographic cross-section showing sunny valley between two mountain ranges, left side blue-grey mountains labeled 'VOSGES', right side dark green mountains labeled 'BLACK FOREST', central golden sunshine rays creating 'SUNSHINE POCKET' text over valley, orange sun icon with '1,800 HOURS' text in top right corner, bottom beige panel with three facts in clean sans-serif text: First fact: 'Protected by two mountain ranges', Second fact: 'Creates Germany's sunniest microclimate', Third fact: 'Perfect for wine and solar energy', flat illustration style with soft gradients"""

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt=prompt,
    width=1024,
    height=1344,
)

Website & app design mocks Generate full web design mockups for prototyping:

prompt = """Full-page modern meal-kit delivery homepage, professional web design layout. Top navigation bar with text links 'Plans', 'Recipes', 'How it works', 'Login' in clean sans-serif. Large hero headline 'Dinner, simplified.' in bold readable font, below it subheadline 'Fresh ingredients. Easy recipes. Delivered weekly.' Two CTA buttons: primary green rounded button with 'Get started' text, secondary outlined button with 'See plans' text. Right side features large professional food photography showing colorful fresh vegetables. Three value prop cards with icons and text 'Save time', 'Reduce waste', 'Cook better'. Bold green (#2ECC71) accent color, rounded buttons, crisp sans-serif typography, warm natural lighting, modern DTC aesthetic"""

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt=prompt,
    width=1024,
    height=1344,
)

Comic strips Create consistent comic-style illustrations:

prompt = """Style: Classic superhero comic with dynamic action lines
Character: Diffusion Man (athletic 30-year-old with brown skin tone and short natural fade haircut, wearing sleek gradient bodysuit from deep purple to electric blue, glowing neural network emblem on chest, confident expression) extends both hands forward shooting beams of energy
Setting: Digital cyberspace environment with floating data cubes
Text: "Time to DENOISE this chaos!"
Mood: Intense, action-packed with bright energy flashes"""

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt=prompt,
    width=1024,
    height=768,
)

Stickers Generate die-cut sticker designs:

prompt = """A kawaii die-cut sticker of a chubby orange cat, featuring big sparkly eyes and a happy smile with paws raised in greeting and a heart-shaped pink nose. The design should have smooth rounded lines with black outlines and soft gradient shading with pink cheeks."""

response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt=prompt,
    width=768,
    height=768,
)

Photography styles

FLUX.2 excels at various photography aesthetics. Add style keywords to your prompts:

Style	Prompt suffix
Modern photorealism	`close up photo, photorealistic`
2000s digicam	`2000s digicam style`
80s vintage	`80s vintage photo`
Analog film	`shot on 35mm film, f/2.8, film grain`
Vintage cellphone	`picture taken from a vintage cellphone, selfie style`

# Example: 80s vintage style
response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="A group of friends at an arcade, neon lights, having fun playing games, 80s vintage photo",
    width=1024,
    height=768,
)

Multi-language support

FLUX.2 supports prompting in many languages without translation:

# French
response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="Un marché alimentaire dans la campagne normande, des marchands vendent divers légumes, fruits. Lever de soleil, temps un peu brumeux",
    width=1024,
    height=768,
)

# Korean
response = client.images.generate(
    model="black-forest-labs/FLUX.2-pro",
    prompt="서울 도심의 옥상 정원, 저녁 노을이 지는 하늘 아래에서 사람들이 작은 등불을 켜고 있다",
    width=1024,
    height=768,
)

Prompting best practices

Golden rules

Order by importance: List the most important elements first in your prompt.
Be specific: The more detailed, the more controlled the output.

Prompt framework Follow this structure: Subject + Action + Style + Context

Subject: The main focus (person, object, character).
Action: What the subject is doing or their pose.
Style: Artistic approach, medium, or aesthetic.
Context: Setting, lighting, time, mood.

Avoid negative prompting FLUX.2 does not support negative prompts. Instead of saying what you don’t want, describe what you do want:

❌ Don’t	✅ Do
`portrait, --no text, --no extra fingers`	`tight head-and-shoulders portrait, clean background, natural hands at rest out of frame`
`landscape, --no people`	`serene mountain landscape, untouched wilderness, pristine nature`

Troubleshooting

Text not rendering correctly

Use FLUX.2 [flex] for better typography.
Put exact text in quotes within the prompt.
Keep text short and clear.

Colors not matching

Use HEX codes with “color” or “hex” keyword.
Be explicit about which element should have which color.

Composition not as expected

Use JSON structured prompts for precise control.
Specify camera angle, distance, and composition type.
Use position descriptors (foreground, midground, background).

See all available FLUX models in the image models catalog.

GUIDES

MODEL QUICKSTARTS

BUILD APPS

BUILD AGENTS

WORK WITH CODING AGENTS

SDK INTEGRATIONS

FLUX.2

Generating an image

Parameters

Image-to-image with reference images

FLUX.1 Kontext editing

JSON structured prompts

Hex color code prompting

Advanced use cases

Photography styles

Multi-language support

Prompting best practices

Troubleshooting

​FLUX.2

​Generating an image

​Parameters

​Image-to-image with reference images

​FLUX.1 Kontext editing

​JSON structured prompts

​Hex color code prompting

​Advanced use cases

​Photography styles

​Multi-language support

​Prompting best practices

​Troubleshooting

FLUX.2

Generating an image

Parameters

Image-to-image with reference images

FLUX.1 Kontext editing

JSON structured prompts

Hex color code prompting

Advanced use cases

Photography styles

Multi-language support

Prompting best practices

Troubleshooting