FLUX.1 [dev]

FLUX.1 [dev] is Black Forest Labs' 12-billion-parameter open-weights text-to-image model,

released August 1, 2024 alongside the company's launch (founded by the core Stable Diffusion researchers).

It is a guidance-distilled rectified-flow transformer with a hybrid multimodal/parallel diffusion-transformer

architecture, noted at launch for strong prompt adherence, accurate human anatomy (hands), and unusually good

text rendering inside images — quality widely compared to Midjourney v6 and DALL-E 3.

Provider-reported

human-preference tests at launch (Black Forest Labs, ~800 human judges): visual quality 1044 Elo and prompt

following 1035 Elo, placing it above Stable Diffusion 3 Ultra (1031) and Midjourney v6.0 (1026) in BFL's own

evaluation.

The weights are free for non-commercial use under the FLUX.1 [dev] Non-Commercial License

(see license note); generated outputs may be used commercially except to train competing models. It powers

features inside Adobe Photoshop and Picsart and remains one of the most-downloaded text-to-image checkpoints

on Hugging Face.

Provider-reported (Black Forest Labs, internal human-preference tests): visual quality 1044 Elo (1st), prompt following 1035 Elo (1st) — ahead of SD3 Ultra (1031) and Midjourney v6.0 (1026) in BFL's own evaluation with ~800 judges.

Independent: widely benchmarked by third parties at launch; Artificial Analysis covered FLUX.1 pricing/quality; remains a top-downloaded T2I model on Hugging Face.