Black Forest Labs releases FLUX
The 12-billion-parameter FLUX.1 suite shipped in three tiers — a paid API model, an open non-commercial model and an Apache-licensed fast model — funded by $31 million in seed money.
- Open weights & ecosystem
- Models & capabilities
- Minor
A group of researchers who had previously built VQGAN, Latent Diffusion and the original Stable Diffusion models launched Black Forest Labs and released FLUX.1, a new text-to-image model family, with $31 million in seed funding led by Andreessen Horowitz. The suite used a 12-billion-parameter hybrid transformer architecture based on flow matching, and shipped in three tiers: FLUX.1 [pro], a paid API-only model the company described as state of the art on prompt-following and image quality; FLUX.1 [dev], an open-weight model available on Hugging Face for non-commercial use; and FLUX.1 [schnell], a faster, smaller model released under the fully permissive Apache 2.0 licence.
Within weeks, xAI began using FLUX to power the image generator built into Grok, positioning it as an alternative to the more heavily filtered image tools from OpenAI’s DALL-E and Google’s Imagen. That integration drew criticism after users found the combination produced images — including fabricated depictions of public figures — with fewer guardrails than competing products, and xAI later moved to its own in-house image models rather than continuing the FLUX partnership.
FLUX’s launch continued a recurring pattern in generative AI: researchers leaving an earlier open-model company to found a new one on similar terms, competing on a mix of open licensing and paid commercial access rather than choosing one model exclusively.