AIWiki
Malaysia
Back to all articles
Tools & Platformsimage generationdiffusion modelsflux

Black Forest Labs

3 min readUpdated October 2026
Black Forest Labs
Type
Private AI research company
Founded
2024, Freiburg im Breisgau, Germany
Founders
Robin Rombach and fellow latent-diffusion researchers
Key products
FLUX.1, FLUX.2, FLUX 3
Valuation
$3.25 billion (Series B, December 2025)
Related
Diffusion Models, Stable Diffusion, Text-to-Image Model
Black Forest Labs (BFL) is a German artificial intelligence laboratory that develops the FLUX family of generative visual models, covering image generation, video and — in its newest releases — multimodal systems that also predict physical actions. Founded in 2024 by the researchers behind latent diffusion architectures, it has become one of Europe's most valuable AI companies.[1]

History

Black Forest Labs was established in 2024 in Freiburg im Breisgau by researchers including Robin Rombach, Patrick Esser and Andreas Blattmann, whose earlier academic work on latent diffusion underpinned Stable Diffusion and similar text-to-image systems. The company launched with an initial investment of about US$31 million.[2]

FLUX.1 arrived in August 2024 in three variants — a commercial pro tier, a non-commercial dev tier and an Apache-licensed schnell model — and quickly became the default open alternative in image generation. Commercial traction followed: in September 2025 Bloomberg reported that Meta had signed a contract worth over US$100 million (reported at $140 million) for a multi-year licence of BFL's models, and in December 2025 the company closed a $300 million Series B at a $3.25 billion post-money valuation co-led by AMP and Salesforce Ventures, lifting total funding above $450 million with NVIDIA, Andreessen Horowitz and Meta among its backers.[3][4]

FLUX.2 was released in November 2025 as the next generation of the image models, and in July 2026 BFL released FLUX 3, a multimodal model combining video generation with native audio (up to 20 seconds), image generation and action prediction for robotic control.[5]

Technology

FLUX models are built on diffusion architectures refined with a transformer backbone rather than the U-Net design of earlier systems, and the company applies mixture-of-experts conditioning and flow-based training to improve prompt adherence and detail. The product line mixes open and closed tiers: schnell weights are openly downloadable and can run on local GPUs, while pro tiers are served through BFL's API — a dual licensing strategy that funds research while keeping the models visible in the open-source ecosystem.[2] The FLUX 3 release extends the same family from pixels to embodied outputs, generating video with synchronized audio while also emitting robot control actions from visual observations and text instructions.[5]

Why It Matters for AI

Black Forest Labs is the strongest European counter-example to the concentration of frontier generative AI in the United States and China. Its licensing deals show that model quality alone can build a large B2B business without a consumer app, and its open-weights tier set the de facto standard that open image models are measured against. The company's pivot from images to video, audio and action prediction mirrors the wider convergence of generative media models into multimodal systems.

>See Also

🇲🇾Malaysian Context

🇲🇾 Relevance to Malaysia: FLUX's open weights let Malaysian studios, agencies and e-commerce sellers generate advertising and product imagery on local hardware instead of paying per-image foreign APIs — significant for SMEs where AI content costs must stay in ringgit. Malaysia's creative and digital sectors aligned with MDEC frequently use FLUX-family models for campaign assets, while the open schnell tier suits on-premises deployment for firms handling unreleased product visuals. As generated media spreads, PDPA compliance and the watermarked-provenance expectations of Malaysia's AI content moderation rules make provenance-capable generation increasingly important for local publishers.

References

  1. ↑Black Forest Labs. Announcing Black Forest Labs. https://bfl.ai/announcing-black-forest-labs/
  2. ↑Wikipedia. Flux (text-to-image model). https://en.wikipedia.org/wiki/Flux_(text-to-image_model)
  3. ↑Bloomberg. (2025). Meta to Pay $140 Million to Use Black Forest Labs AI for Images. https://www.bloomberg.com/news/articles/2025-09-09/meta-to-pay-140-million-to-use-black-forest-labs-ai-for-images
  4. ↑Black Forest Labs. Laying the Foundations for Visual Intelligence — Our $300M Series B. https://bfl.ai/blog/our-300m-series-b
  5. ↑Black Forest Labs. FLUX 3: Multimodal Video, Image & Audio. https://bfl.ai/blog/flux-3