A new generative AI startup called Black Forest Labs has hit the scene, debuting with a suite of text-to-image models branded FLUX.1. Based in Germany, Black Forest was founded by some of the researchers involved in developing Stable Diffusion and has raised $31 million in funding from principal investor Andreessen Horowitz and angels including CAA founder and former talent agent Michael Ovitz. The FLUX.1 suite focuses on “image detail, prompt adherence, style diversity and scene complexity,” the company says of its three initial variants: FLUX.1 [pro], FLUX.1 [dev] and FLUX.1 [schnell].

“FLUX.1 introduces several technical innovations,” explains VentureBeat, citing “‘flow matching,’ a method that generalizes diffusion models, and incorporates rotary positional embeddings and parallel attention layers for enhanced performance and hardware efficiency.” VentureBeat calls the results “impressive,” while Ars Technica writes that Flux.1 “is eerily good at creating human hands.”

“The impact of FLUX.1 could extend far beyond the AI research community,” with “graphic designers, digital artists and creative professionals “ among those VentureBeat lists who “may discover new possibilities in the model’s ability to generate high-quality images across a wide range of styles and aspect ratios.”

“FLUX is an advanced, open-source text-to-image model with 12 billion parameters,” writes Decrypt, which ran tests comparing it to Midjourney, SD3 Medium and AuraFlow and says FLUX.1 came out on top.

The Black Forest announcement says FLUX.1 [pro] is the state-of-the-art offering, inviting users to sign-up via the API.

FLUX.1 [dev] is an open-weight, guidance-distilled model for non-commercial applications, available on Hugging Face.

FLUX.1 [schnell], the fastest model, “is tailored for local development and personal use,” and openly available under an Apache 2.0 license. Weights are downloadable on Hugging Face and inference code can be found on GitHub and in Hugging Face.

Decrypt explains that “Black Forest has partnered with Fal AI — developers of open-source model AuraFlow — to support cloud generations, noting that “the models are also available for testing free on Replicate.com.” Once users hit a daily quota, additional processing costs $1 for 33 images with Flux Pro or 333 with Flux Schell.

Ars Technica says the output of the two higher-end FLUX.1 models “represent a significant improvement over Stable Diffusion XL, the team’s last major release under Stability (if you don’t count SDXL Turbo),” and “are generally comparable with OpenAI’s DALL-E 3 in prompt fidelity, with photorealism that seems close to Midjourney 6.”