Black Forest Labs

Black Forest Labs is an independent AI research company headquartered in Freiburg im Breisgau, Germany, with a second office in San Francisco, that develops foundation models for visual generation and editing. Its FLUX model families span text-to-image generation, in-context image editing (FLUX.1 Kontext), the multi-tier FLUX.2 line, and FLUX 3, a multimodal model announced in July 2026 that jointly learns from images, video and audio in a unified architecture and adds action prediction for robotics.

Models are delivered through three channels: a hosted API for production workloads, downloadable open weights for self-hosted deployment and fine-tuning (distributed via Hugging Face and GitHub), and a browser playground for experimentation. The company also offers enterprise licensing and cites SOC 2 and ISO 27001 on its site. Distribution partners and integration surfaces include Microsoft Azure AI Foundry and Adobe's creative tools, and its models are reported to be used by Adobe, Canva, Meta, Microsoft, Picsart, fal.ai, ElevenLabs, VSCO, Vercel and Snap.

The FLUX.2 family ranges from 4B and 9B [klein] variants optimized for sub-second inference on consumer hardware to a 32B open-weight [dev] model and a [max] tier supporting up to 10 reference images and 4-megapixel output. FLUX 3, built on the company's Self-Flow approach, generates video with native multilingual audio up to 20 seconds in a single generation and supports text-to-video, image-to-video, video-to-video, keyframe-to-video and agentic multi-shot chaining.

Founding story

Founded in 2024 by researchers who had previously developed latent diffusion and Stable Diffusion. Per Startup Intros, the founding group comprised ten co-founders including Robin Rombach (CEO), Andreas Blattmann, Patrick Esser and Dominik Lorenz, who had worked in university research labs at LMU Munich and Heidelberg University before joining Stability AI; they left to form an independent organization dedicated solely to visual intelligence. Their collective papers are cited as having more than 120,000 citations.

Business model

Dual-license model combining open-weight distribution with commercial monetization. Revenue is described as coming from per-generation API usage fees, annual commercial licenses for enterprises self-hosting the models, and large multi-year platform contracts. A self-serve commercial licensing portal offers standardized terms alongside negotiated enterprise agreements. Open [dev] weights serve as an adoption and prototyping funnel into paid licenses and hosted inference, with community contributions described as reducing R&D cost.

API usage priced per image or per megapixel (for example, $0.014-$0.015 per image for FLUX.2 [klein] variants and from $0.03 per megapixel for [pro] and [max]; $40 per 1,000 images for FLUX.1 Kontext [pro] and FLUX1.1 [pro] on Azure AI Foundry), plus enterprise open-weight licenses and multi-year platform contracts such as a reported $140M Meta agreement.

Traction

Sacra estimates roughly $96M in annualized revenue as of August 2025. Reported commercial contracts total about $300M across Meta, Adobe, Canva and Snap, including a $140M Meta agreement. Models are used by Adobe, Canva, Meta, Microsoft, Picsart, fal.ai, ElevenLabs, VSCO, Vercel and, per press reports, xAI's Grok chatbot. Early internal evaluations of FLUX 3 video report preference rates of 93% versus Luma Ray 3.2, 77% versus Runway Gen-4.5, 69% versus Grok Imagine Video and 60% versus Kling v3 Pro, though the company describes these results as preliminary.

Latest developments

In July 2026 the company introduced FLUX 3 in early access, a unified multimodal model for video with native audio (up to 20 seconds per generation), image synthesis and editing, and action prediction, with a staged launch plan covering API and private-weight access and an eventual open-weight FLUX 3 Dev backbone. It co-developed FLUX-mimic with mimic robotics for dexterous manipulation, referencing production testing at Audi. Company announcements also note that Martin Scorsese is an advisor and that the co-founder and CEO publicly urged G7 leaders to support open innovation.