FLUX.1 - AI Image Models Tool

Overview

FLUX.1 is an open-source, state-of-the-art text-to-image model family from Black Forest Labs focused on prompt adherence, high visual detail, and diverse outputs. The suite ships in three variants: FLUX.1 [pro] (highest-quality commercial tier), FLUX.1 [dev] (guidance-distilled, open-weight variant for non-commercial use), and FLUX.1 [schnell] (very fast, Apache‑2.0 licensed variant for local/personal use). FLUX models are available to run on Replicate and via Black Forest Labs’ own API; the project also publishes inference code and weights for several variants on GitHub and Hugging Face. (Sources: Replicate blog, Black Forest Labs, GitHub.) Technically, public FLUX.1 releases use a hybrid architecture of multimodal + parallel diffusion transformer blocks scaled to roughly 12 billion parameters, and training builds on flow‑matching methods; the design also uses rotary positional embeddings and parallel attention to improve quality and hardware efficiency. Black Forest Labs has continued development (notably releasing FLUX1.1 [pro] with major speed and quality improvements), and the model family is being benchmarked on third‑party leaderboards. (Sources: Replicate model readme, Black Forest Labs announcement, GitHub.)

Key Features

• Three variants: pro (commercial), dev (open-weight, non-commercial), schnell (fast, Apache‑2.0).
• Strong prompt adherence—handles complex compositional instructions and repeated/precise text well.
• High visual detail and diverse outputs across styles and aspect ratios.
• Conditioning tools: Canny/Depth structural conditioning and in/out‑painting variants available.
• Hardware/efficiency features: guidance distillation, rotary embeddings, and parallel attention layers.
• Local inference support with provided repo, TensorRT instructions, and ComfyUI integration.
• Available via Replicate, Black Forest Labs API, and weights on Hugging Face (where licensed).

Example Usage

Example (python):

import replicate

# Example: generate an image with FLUX.1 [dev] on Replicate
# Requires: pip install replicate, and set REPLICATE_API_TOKEN env var

client = replicate.Client()
model = "black-forest-labs/flux-dev"

output = client.run(model, input={
    "prompt": "A detailed cinematic portrait of an astronaut in a baroque suit, dramatic lighting, ultra-detailed",
    "width": 1024,
    "height": 1024,
    "steps": 28
})

# 'output' is usually a list of image URLs returned by Replicate
print("Generated:", output)

Pricing

Platform pricing varies. Replicate lists FLUX.1 [pro] at $0.055/image, FLUX.1 [dev] at $0.030/image, and FLUX.1 [schnell] at $0.003/image. Black Forest Labs’ own BFL API publicity lists slightly different per-image rates for BFL-hosted endpoints (example published rates: FLUX.1 [dev] ≈ $0.025/img, FLUX.1 [pro] ≈ $0.05/img, FLUX1.1 [pro] ≈ $0.04/img). For commercial licensing, Black Forest Labs requires a paid license or API agreement—see the vendor pages for current terms and enterprise pricing.

Benchmarks

Model size: ~12 billion parameters (Source: https://internal.replicate.com/black-forest-labs/flux-1.1-pro)

Artificial Analysis (reported) Elo: 1153 (FLUX1.1 [pro] reported top Elo on Artificial Analysis) (Source: https://venturebeat.com/ai/black-forest-labs-releases-flux-1-1-pro-and-an-api/)

Replicate per-image pricing (platform listing): FLUX.1 [pro] $0.055 / img; FLUX.1 [dev] $0.030 / img; FLUX.1 [schnell] $0.003 / img (Source: https://replicate.com/blog/flux-state-of-the-art-image-generation)

Key Information

• Category: Image Models
• Type: AI Image Models Tool