FLUX Family of Models (Black Forest Labs) - AI Image Models Tool

Overview

FLUX (Black Forest Labs) is a commercial family of API-accessible image-generation and editing models that span high-resolution text-to-image, inpainting/outpainting, edge/depth-guided editing, and rapid image variation. The suite includes production-grade variants (FLUX1.1 [pro] and FLUX1.1 [pro] Ultra) for high-fidelity commercial outputs, developer/open-weight versions (FLUX.1 [dev], FLUX.1 [schnell]) for local prototyping and research, and specialized editing models (FLUX.1 Fill, Depth, Canny, Redux). The collection is distributed through partners such as Replicate and BFL’s API, and supports fine-tuning, multi-step chains (generate → edit → restyle), and control inputs like Canny/Depth maps for structure-preserving edits. ([replicate.com](https://replicate.com/collections/flux)) Black Forest Labs positions FLUX1.1 [pro] as a performance-first model: Ultra mode generates up to 4 megapixel images with reported per-sample times in the range of ~10 seconds and a “raw” realism toggle for candid-style outputs. Public documentation and model READMEs describe a ~12 billion parameter architecture built with flow-matching techniques, rotary positional embeddings and parallel attention to improve quality and hardware efficiency. Pricing is credit-based (1 credit = $0.01) and BFL publishes per-image credit costs (e.g., FLUX1.1 [pro] Ultra ≈ $0.06/image). ([bfl.ai](https://bfl.ai/flux-1-1-ultra?utm_source=openai))

Key Features

• High-resolution text-to-image up to 4 megapixels (Ultra mode).
• Raw mode for more candid, less synthetic photographic outputs.
• Professional inpainting/outpainting with FLUX.1 Fill for seamless edits.
• Edge (Canny) and depth-guided control for structure-preserving edits.
• Open-weight Dev and Schnell builds for local prototyping and research.

Example Usage

Example (python):

# Example: generate one image with FLUX1.1 Pro Ultra via Replicate
# Requires: pip install replicate and REPLICATE_API_TOKEN environment variable
import replicate

output = replicate.run(
  "black-forest-labs/flux-1.1-pro-ultra",
  input={
    "prompt": "Photo‑realistic close-up portrait of a woman in soft studio light, shallow depth of field",
    "raw": True,              # enable 'raw' realism mode (model-specific option)
    "num_outputs": 1
  }
)

# Save the returned FileOutput to disk
with open("flux_output.png", "wb") as f:
    f.write(output[0].read())

# See Replicate docs for async/stream and other input options. (Replicate run pattern shown in their docs.)
# Source: Replicate model pages and Replicate Python docs. ([replicate.com](https://replicate.com/black-forest-labs/flux-1.1-pro-ultra))

Pricing

Credit-based pricing: 1 credit = $0.01 USD. FLUX.1 family uses per-image credit costs (examples from BFL docs: FLUX1.1 [pro] ≈ $0.04/image; FLUX1.1 [pro] Ultra ≈ $0.06/image; FLUX.1 Fill and other edit endpoints have per-image credit costs). See Black Forest Labs pricing docs for the full table and megapixel-based tiers. https://docs.bfl.ml/quick_start/pricing

Benchmarks

Artificial Analysis (image arena) Elo: Highest overall Elo reported on launch (public articles report ~1153 in Oct 2024) (Source: https://artificialanalysis.ai/image/arena)

FLUX1.1 [pro] Ultra generation time: ≈ 10 seconds per 4MP sample (reported by Black Forest Labs / model README) (Source: https://replicate.com/black-forest-labs/flux-1.1-pro-ultra)

Model scale: ≈ 12 billion parameters (public documentation / model README) (Source: https://replicate.com/black-forest-labs/flux-1.1-pro)

Replicate usage (popularity indicator): Flagship variants show millions of runs (e.g., flux-1.1-pro: 66M runs; flux-1.1-pro-ultra: 19.7M runs on Replicate) (Source: https://replicate.com/collections/flux)

Published per-image price (Ultra): $0.06 per image (FLUX1.1 [pro] Ultra, documentation) (Source: https://docs.bfl.ml/quick_start/pricing)

Key Information

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