Former Stable Diffusion developers launch Flux, a new image model challenging Midjourney

Black Forest Labs, a new AI startup founded by former Stability.ai developers, aims to create cutting-edge generative AI models for images and video. The startup claims that its first model family, Flux.1, sets new standards in text-to-image generation.

Based in Freiburg, Germany, Black Forest Labs was formed by AI researchers and developers who previously worked on Stable Diffusion and other generative AI models at Stability.ai.

The team left Stability.ai in spring and has since raised $31 million in seed funding. Andreessen Horowitz led the round, with participation from angel investors including Brendan Iribe, Michael Ovitz, Garry Tan, Timo Aila, and Vladlen Koltun. General Catalyst and MätchVC also invested.

The startup's founders have an impressive track record, having contributed to innovations like VQGAN, Latent Diffusion, Stable Diffusion XL, Stable Video Diffusion, Rectified Flow Transformers, and Adversarial Diffusion Distillation for real-time image synthesis.

Flux "Pro" aims to set new standards for generative image AI

To mark the launch, Black Forest Labs has released the Flux.1 suite of text-to-image models. The company claims that these models "define a new state of the art in image detail, prompt adherence, style diversity, and scene complexity for text-to-image synthesis". The startup seems confident that it's way ahead of the game.

All Flux.1 models support multiple aspect ratios and resolutions ranging from 0.1 to 2.0 megapixels.

According to Black Forest Labs, Flux.1 [pro] and [dev] outperform popular models like Midjourney v6.0, DALL-E 3 (HD), and SD3-Ultra in visual quality, prompt tracking, size/aspect variability, typography, and output versatility. Flux.1 [schnell] (meaning "fast") is touted as the most advanced fast model, outperforming competitors in its class and even some non-distilled models like Midjourney.

Flux.1 is available in three variants: Flux.1 [pro] for commercial use via Replicate and fal.ai, Flux.1 [dev] for non-commercial applications, and Flux.1 [schnell] for local development and personal use. The latter is the fastest model and is freely available under an Apache 2.0 license.

Black Forest Labs plans to release text-to-video models next, promising "precise creation and editing at high definition and unprecedented speed." The company is looking for AI and backend engineers to expand its team.