INDEX VERIFIED
Archive record Verified: Cluster 0x89A Updated 14 mins ago
State-of-the-Art DiT 12B Parameters Apache 2.0 / Open-Weights Available
neurology

FLUX.1

by Black Forest Labs

Autoregressive-caliber Flow Matching Diffusion Transformer (DiT). Engineered for unprecedented typography synthesis, strict prompt fidelity, and ultra-high dynamic range composition.

Architecture
Flow Matching DiT
MMDiT Hybrid Block
Text Rendering
Dual T5-XXL + CLIP
Photorealism
96.7% Score
Skin, hands & micro-texture
Release Tier
Tri-Variant
schnell / dev / pro
AiRecMark Composite Metric Top 0.5% Global
95.4 / 100

Aggregate benchmark evaluated across 10,000 deterministic prompts including synthetic glyphs, optical anatomy, and multi-actor staging.

Typography & Glyph Rendering 98 / 100
Architectural Openness 97 / 100
Prompt Nuance Adherence 96 / 100
Anatomical & Hand Precision 95 / 100
Step Latency Efficiency 92 / 100
Architecture Segmentation

FLUX.1 Model Architecture Matrix

Three specialized variants addressing hyper-speed local processing, research exploration, and high-fidelity enterprise production.

Apache 2.0 Distilled 4-Step

FLUX.1 [schnell]

Optimized for ultra-fast localized inference and live generation pipelines with unprecedented 4-step fidelity.

Inference Steps 1 - 4 Steps
Minimum VRAM 12 GB (Quantized)
Latency Target < 0.8s on RTX 4090
Licensing Commercial Allowed
Community Standard
Non-Commercial 12B Parameters

FLUX.1 [dev]

Guidance-distilled base model directly derived from the Pro core. Ideal for research, LoRA fine-tuning, and offline rigs.

Inference Steps 20 - 28 Steps
Minimum VRAM 24 GB (Native FP16)
Prompt Guidance Guidance Distilled
Licensing BFL Non-Commercial
Closed API Frontier Tier

FLUX.1 [pro]

State-of-the-art closed enterprise model offering extreme realism, complex multi-object composition, and high dynamic range.

Inference Steps Dynamic Adaptive
Deployment BFL Cloud / Replicate API
Color Grading Full Precision HDR
SLA Enterprise Dedicated
Head-to-Head Empirical Benchmarks

Archive record vs Industry Flagships

Deterministic evaluations across synthetic datasets: Text Synthesis Benchmark v4, Anatomy & Occlusion Matrix (AOM), and Complex Scene Spatial Graph.

Metric Evaluation Comparison

N = 2,500 Trials per Model
Text Typography Precision (%) FLUX leads by +11.4%
FLUX.1 [dev]
98.2%
Ideogram v2
94.6%
Midjourney 6.1
86.8%
Fingers & Hand Anatomy Coherence AOM Accuracy
FLUX.1 [pro]
96.1%
Midjourney 6.1
93.4%
SD 3.5 Large
88.0%
Complex Multi-Subject Spatial Adherence Graph Matching
FLUX.1 [dev]
95.4%
Midjourney 6.1
91.2%
Ideogram v2
89.1%
bfl_node_run.sh
28 STEPS
$ python flux_infer.py \
--prompt "Retro terminal with legible UI" \
--variant=dev --guidance=3.5
> Model loaded: 12B Flow Matching DiT
> Hardware: 1x NVIDIA H100 80GB PCIe
> Allocated VRAM: 14.2 GB (FP16 weights + context)
> Dual Encoders: T5-XXL (4096 tokens) + CLIP-L
> Latency: 2.34s (28 steps)
> Text glyph accuracy: 100% [VERIFIED]
developer_board
Quantization Ready
GGUF (Q4_K_M, Q8_0) & NF4 formats supported
6.8 GB Min
Economics & Cloud Pricing

Inference Cost & Compute Matrix

Aggregated across official Black Forest Labs API, Fal.ai, and self-hosted compute instances.
FLUX.1 [schnell] API
~$0.003 / image
Standard 1024x1024 output. Lowest latency generation pipeline for real-time workflows.
FLUX.1 [dev] API
~$0.025 / image
Base guidance-distilled endpoint on BFL cloud and serverless platforms like Fal.ai and Together.
28-step high quality
FLUX.1 [pro] API
~$0.040 / image
Frontier enterprise quality. Full dynamic range and high-fidelity text typography.
Adaptive multi-stage
Self-Hosted Compute
$0.00 license cost
Run locally or in private VPC. HuggingFace weight repository for Schnell and Dev models.
RTX 3090 / 4090 / A10G
Ecosystem Comparison

Related SOTA Diffusion Models

View Matrix (14 Models) arrow_forward
Midjourney v6.1 94.8

Frontier aesthetic curation with stylized lighting defaults. Closed proprietary subscription ecosystem.

Closed Discord/Web Diff Specs →
Ideogram v2 93.6

Specialized in graphic design, typographic posters, and brand layout composition. Closed API available.

API Only Diff Specs →
SD 3.5 Large 91.2

Stability AI 8B MMDiT. Open weights with commercial community license for under $1M revenue organizations.

Open Weights Diff Specs →
Recraft v3 94.1

Native vector SVG and raster synthesis engine tailored specifically for product design and brand asset kits.

API & UI Diff Specs →
verified
AiRecMark Analyst Evaluation Dossier #BFL-2024-F1
“FLUX.1 represents the current frontier in diffusion image synthesis, setting the industry benchmark for complex typography rendering and anatomical correctness. Its hybrid flow-matching architecture delivers unprecedented speed-to-fidelity ratios.”
Technical Lead: Dr. J. Vance, AiRecMark Evaluation Cluster Verified on: CUDA 12.4 • PyTorch 2.4.0 • cuDNN 9.1