Advertisement

Former Stability AI engineer and Latent Diffusion inventor releases FLUX.1 text-to-image model

Last week, AI startup Black Forest Labs announced its launch along with the release of its first text-to-image AI model suite, FLUX.1. The German company was founded by researchers who developed and invented the latent diffusion technology behind Stable Diffusion, aiming to create advanced generative AI technologies for images and videos.

It can be used on replicate: https://replicate.com/black-forest-labs/flux-pro

The release of FLUX.1 comes only about seven weeks after Stability AI launched the controversial Stable Diffusion 3 Medium in mid-June. Stability AI's product was widely criticized by image synthesis enthusiasts for its poor performance in generating human anatomy, with users sharing numerous examples of limb and body distortions on social media. Prior to this problematic release, three key engineers from Stability AI—Robin Rombach, Andreas Blattmann, and Dominik Lorenz—resigned. They then went on to co-found Black Forest Labs with Patrick Esser, a co-developer of latent diffusion technology, and others.

FLUX.1 offers three different versions:

FLUX.1 [pro]

The top-tier version of FLUX.1, providing cutting-edge image generation performance with state-of-the-art prompt-following capabilities, visual quality, image details, and output diversity. The inference compute for FLUX.1 [pro] in the API is being progressively scaled up. Additionally, FLUX.1 [pro] is accessible via Replicate and fal.ai. The team also offers tailored enterprise solutions (to make money lol).

FLUX.1 [dev]

FLUX.1 [dev] is an open-weight guided distillation model designed for non-commercial applications. Distilled directly from FLUX.1 [pro], FLUX.1 [dev] offers similar quality and prompt-following ability while being more efficient than standard models of the same size. The weights for FLUX.1 [dev] are available on Hugging Face and can be tried directly on Replicate or fal.ai. For commercial applications, the team can be contacted separately.

FLUX.1 [schnell]

The fastest model, specifically designed for local development and personal use. FLUX.1 [schnell] is publicly available under the Apache 2.0 license. Similar to FLUX.1 [dev], its weights can be found on Hugging Face, and inference code is available on GitHub. FLUX.1 [schnell] can also be accessed via Replicate and fal.ai. (Open-source to build community and attract users)