FLUX.2

FLUX.2 is the second generation of Black Forest Labs’ FLUX image generation models, released in November 2025. It is the leading image generation model family on Prodia for photorealistic output, creative control, and multi-image editing workflows.

Architecture

FLUX.2 is a 32B parameter model built on a latent flow matching architecture. It couples a Mistral-3 24B vision-language model with a rectified flow transformer, giving the model strong understanding of both visual and textual inputs. This architecture enables features that previous FLUX models couldn’t support:

32K-token prompts — describe complex scenes in extreme detail
Multi-image input — provide up to 10 reference images for editing, style transfer, or identity-preserving generation
HEX color understanding — specify exact colors in your prompt (e.g., “a bag in #FF5733”) and the model will match them
Native style presets — 17 built-in style presets (Dev variant) eliminate the need for style-specific prompt engineering

Choosing a variant

FLUX.2 offers four variants. Choosing the right one depends on your priorities:

Variant	Best for	Resolution	Generation time	Price
Dev	Prototyping, style presets, fine control	512–1920px	~3s	$0.01–0.015
Pro	Production workloads, high fidelity	64–4096px	~12s	$0.03
Flex	Multi-image editing, maximum control	64–4096px	~22s	$0.06
Max	Highest quality output	256–2048px	~15s	$0.04

FLUX.2 [dev] — The open-weight variant. Best for rapid iteration with adjustable steps (1–50), guidance scale (1–10), and 17 style presets. Supports up to 8 input images for img2img. The fastest variant at ~3s per generation.

FLUX.2 [pro] — The production workhorse. Generates at up to 4096x4096 pixels (multiples of 16) with strong prompt adherence. Supports up to 8 input images for editing with a combined 9MP limit.

FLUX.2 [flex] — Maximum creative control. Adjustable steps and guidance parameters plus support for up to 10 input images. Best for complex multi-reference editing workflows like product photography or identity-preserving edits.

FLUX.2 [max] — Highest quality output. Optimized for the best possible single-image generation at up to 2048px. When quality matters more than speed.

FLUX.1 vs FLUX.2

Feature	FLUX.1	FLUX.2
Architecture	12B flow matching	32B flow matching + Mistral-3 24B VLM
Max resolution	1920px	4096px
Multi-image input	Not supported	Up to 10 images
Prompt length	512–1024 tokens	32K tokens
HEX color control	No	Yes
Generation quality	Excellent	State-of-the-art

Job types

FLUX.2 [dev]:

Job type	Description
`inference.flux-2.dev.txt2img.v1`	Generate an image from text
`inference.flux-2.dev.img2img.v1`	Transform images with text guidance

FLUX.2 [pro]:

Job type	Description
`inference.flux-2.pro.txt2img.v1`	Generate an image from text
`inference.flux-2.pro.img2img.v1`	Edit images (up to 8 inputs, 9MP combined)

FLUX.2 [flex]:

Job type	Description
`inference.flux-2.flex.txt2img.v1`	Generate an image from text
`inference.flux-2.flex.img2img.v1`	Edit images (up to 10 inputs)

FLUX.2 [max]:

Job type	Description
`inference.flux-2.max.txt2img.v1`	Generate an image from text
`inference.flux-2.max.img2img.v1`	Edit images (up to 8 inputs)

Parameters

Common to all variants:

prompt (required) — text description, up to 32K tokens
width / height — output dimensions in pixels (ranges vary by variant, must be multiples of 16 for Pro/Flex/Max)
seed — integer for reproducible results
safety_tolerance — filter level from 0 (strict) to 5 (permissive), default 2

Dev variant extras:

steps — inference steps, 1–50 (default: 28). Lower steps = faster but less detailed
guidance_scale — classifier-free guidance, 1.0–10.0 (default: 4.0). Higher = more prompt-adherent
style_preset — one of 17 presets: 3d-model, analog-film, anime, cinematic, comic-book, digital-art, enhance, fantasy-art, isometric, line-art, low-poly, neon-punk, origami, photographic, pixel-art, texture, craft-clay

Flex variant extras:

steps — inference iterations, 1–50 (default: 50)
guidance — prompt adherence, 1.5–10.0 (default: 4.5)

Image-to-image (all variants):

images — array of input image filenames

Prompting tips

Be descriptive: FLUX.2’s 32K token context window rewards detailed prompts. Describe subject, action, style, lighting, composition, and mood
Use style presets (Dev): the built-in presets like anime, cinematic, photographic produce more consistent stylized output than adding style words to your prompt
HEX colors work: include specific colors like “#2563EB blue accent” and the model will match them
Multi-image editing: when using img2img with multiple reference images, your prompt should describe what to do with the inputs — “combine the product from the first image with the background from the second”
Safety tolerance: for artistic content, increasing safety_tolerance to 3–4 reduces false positives while keeping genuine safety filtering active

Examples

Text-to-image with style preset (Dev):

{
  "type": "inference.flux-2.dev.txt2img.v1",
  "config": {
    "prompt": "A serene mountain landscape at sunset, photorealistic, 4k",
    "width": 1024,
    "height": 1024,
    "steps": 28,
    "style_preset": "photographic"
  }
}

High-resolution generation (Pro):

{
  "type": "inference.flux-2.pro.txt2img.v1",
  "config": {
    "prompt": "Product photograph of a luxury watch on dark marble, studio lighting, sharp detail",
    "width": 2048,
    "height": 2048
  }
}

Multi-image editing (Flex):

{
  "type": "inference.flux-2.flex.img2img.v1",
  "config": {
    "prompt": "Place the product from the first image onto the marble surface in the second image, matching the lighting",
    "images": ["product.jpg", "background.jpg"],
    "width": 1024,
    "height": 1024,
    "steps": 50
  }
}

Guides

Generating Images Step-by-step guide for generating images with Prodia.

Transforming Images Guide for image-to-image transformation.