FLUX.1

FLUX.1 was the debut release of Black Forest Labs on 1 August 2024, weeks after the lab was founded by researchers who had previously built Stable Diffusion. It was the model that ended the argument about whether open weights could match closed image generators: BFL published comparisons showing FLUX.1 [pro] and [dev] ahead of Midjourney v6.0, DALL·E 3 (HD) and SD3-Ultra on visual quality, prompt following, aspect-ratio variability, typography and output diversity. Technically it is a 12-billion-parameter hybrid of multimodal and parallel diffusion transformer blocks, trained with flow matching and using rotary positional embeddings and parallel attention layers — the rectified-flow recipe that later FLUX generations scaled up rather than replaced. The more consequential decision was commercial, not technical. FLUX.1 shipped in three tiers with three different licences, and the differences are legal, not just practical. FLUX.1 [pro] is closed and available only through the BFL API and partners such as Replicate and fal.ai. FLUX.1 [dev] has open weights on Hugging Face but is licensed for non-commercial use only — commercial deployment requires permission from BFL, which is the single most frequently misread fact about this family. FLUX.1 [schnell] is a distilled, faster variant released under Apache 2.0 with day-one ComfyUI integration, and it is the only one of the three that is unambiguously free for commercial work. That three-way split became the house pattern: FLUX.2 (November 2025) repeated it across four tiers, with only [klein] fully commercial. FLUX.1 itself has been superseded — first by FLUX.2, then by the closed multimodal FLUX 3 in July 2026 — but [schnell] remains in wide use precisely because of its licence.

▸Videos

No videos yet.

▸Reviews