A roughly 70-person lab in Freiburg, Germany has spent two years quietly setting the pace in open image generation. Their founding team wrote the papers behind latent diffusion and Stable Diffusion. Their FLUX.1 [dev] became, by their own account, the most popular open image model in the world. Martin Scorsese is an advisor. In December 2025 they raised $300M at a $3.25B valuation.
If you use AI image tools at all, you are probably using their models — sometimes without the interface telling you.
Sourcing note: every figure, date, partner name, and model detail below comes from Black Forest Labs' own announcements, about page, model pages, documentation, and GitHub. Where a claim is BFL describing itself ("most popular open image model"), it is labelled as their claim rather than presented as independently measured. Last verified July 29, 2026.
What this article solves
The pain point: "black forest labs" is a high-volume search with mostly thin results — either a press-release rewrite of one funding round, or a model list with no sense of what the company is actually doing.
The differentiator: this puts the funding, the team, the full model lineage, and the open-core business model in one place, and is explicit about which claims are BFL's own marketing versus verifiable fact.
The basics
| Founded | 2024 |
| HQ | Freiburg, Germany, and San Francisco |
| Team size | ~70 (their own figure) |
| Funding | $300M Series B, December 1, 2025 |
| Valuation | $3.25B post-money |
| Positioning | "Frontier AI research lab for visual intelligence" |
| Certifications | SOC 2 Type II, ISO 27001:2022 |
BFL's about page describes the founding team as pioneers of "powerful (Latent Diffusion), accessible (Stable Diffusion), and controllable (FLUX.1) visual AI" — which is an unusually direct claim of lineage from the research that made open image generation viable in the first place.
The money and who is behind it
The Series B closed at $300M on a $3.25B post-money valuation, announced December 1, 2025. Per BFL's announcement, the round was co-led by Salesforce Ventures and Anjney Midha (AMP), with continued participation from a16z, NVIDIA, Northzone, Creandum, Earlybird VC, BroadLight Capital, and General Catalyst, and new money from Temasek, Bain Capital Ventures, Air Street Capital, Visionaries Club, Canva and Figma Ventures.
Two names on that list are worth noticing: Canva and Figma Ventures. Design platforms investing in the model vendor they may also license from tells you where BFL expects distribution to come from.
In the same post, BFL named Adobe, Canva, Meta and Microsoft as partners building on their models, and said their enterprise models are among the most widely adopted on Fal.ai, Replicate and TogetherAI.
The advisor nobody expected
In June 2026 BFL announced Martin Scorsese as an advisor. Their published framing is about storyboarding — the problem of communicating what a director sees in their head to a production designer, art director and cinematographer.
Read cynically, it is a credibility play with the film industry. Read at face value, it fits a company whose stated goal is models that support "everyone from storytellers to architects, designers, and engineers." Either way, it is a signal about which market BFL is courting next.
The model lineage
| Generation | Released | What it does |
|---|---|---|
| FLUX.1 | 2024 | Text-to-image, plus Fill / Canny / Depth / Redux / Kontext / Krea editing tools |
| FLUX.2 | November 25, 2025 | Production image generation and editing, up to 4MP, up to 10 reference images |
| FLUX Tools | Through 2026 | Outpainting (May 14), Erase (May 21), Virtual Try-On (May 28), Deblur |
| FLUX 3 | July 23, 2026 | Multimodal: image, video, audio jointly, plus action prediction |
FLUX.2 couples a Mistral-3 24B vision-language model with a rectified flow transformer — the VLM is where the real-world knowledge and prompt understanding come from.
FLUX 3 is the current frontier: one architecture trained jointly on images, video and audio, producing video with native audio up to 20 seconds per generation. It is in request-gated early access, video first. The same backbone runs FLUX-mimic, a video-action model built with mimic robotics and, per BFL, tested and deployed at Audi.
Using FLUX-family output for creative work? Flux 3 AI is an independent browser workspace for generation, editing, reference workflows and upscaling — no GPU, no licence negotiation. Open the image generator.
How "open core" actually works here
This is the part worth understanding, because it explains every product decision BFL makes.
The open side. BFL publishes downloadable weights, and some are genuinely permissive:
- FLUX.1 [schnell] — Apache 2.0
- FLUX.2 [klein] 4B and 4B Base — Apache 2.0
- FLUX.2 VAE — Apache 2.0
- FLUX.2 [dev] (32B), FLUX.1 [dev], and the klein 9B variants — downloadable under a non-commercial licence
The commercial side. Two revenue lines: a per-megapixel API ($0.014 to $0.07 for the first MP depending on tier), and weight licences sold in tiers — Builder (klein models, 10K images/month, one domain), Platform (klein Base 9B + FLUX.2 [dev], 100K/month), Professional (FLUX.2 [dev], 100K/month, up to three domains, aimed at agencies), and Enterprise. There is also a synthetic-data licence granting rights to train on FLUX outputs for a contracted use case.
BFL's own framing on their licensing page positions this between "closed APIs" (naming Google, OpenAI, Midjourney) and "open source" (naming Stable Diffusion, Qwen, Wan) — that is a vendor's map of its own market, and worth reading as positioning rather than neutral analysis.
Why it works: the Apache tier seeds an ecosystem of tools, tutorials and fine-tunes at zero acquisition cost, the non-commercial dev tier gets the highest-quality weights into researchers' hands while keeping commercial use billable, and the API captures teams who do not want GPUs. The klein models shipping on-device with ASUS ProArt laptops in June 2026 is the same strategy pushed one layer down the stack.
What to watch next
- FLUX 3 Image early access, described as opening "in the following weeks" from July 23, 2026
- FLUX 3 Dev — the promised open-weight multimodal backbone, stage four of the launch plan, no date
- Physical AI — BFL frames robotics as a natural extension of the same backbone rather than a new direction
- Whether the open tier stays generous as valuation pressure builds
FAQ
What is Black Forest Labs? A German AI research lab founded in 2024, based in Freiburg and San Francisco, that builds the FLUX image and multimodal models.
Who founded Black Forest Labs? Their about page describes the founding team as pioneers of latent diffusion, Stable Diffusion, and FLUX.1, without naming individuals on that page.
How much funding has BFL raised? A $300M Series B at a $3.25B post-money valuation, announced December 1, 2025, following earlier rounds with a16z, NVIDIA and others.
Is FLUX open source? Partly. Some models are Apache 2.0 (FLUX.1 [schnell], FLUX.2 [klein] 4B and 4B Base, the VAE); most others are downloadable under a non-commercial licence with commercial rights sold separately.
Is BFL related to Stability AI? Not corporately. BFL's founding team includes people behind Stable Diffusion, but the companies are separate.
What is BFL's newest model? FLUX 3, announced July 23, 2026 — multimodal, currently in gated early access.
Is Martin Scorsese involved? As an advisor, announced June 3, 2026, focused on filmmaking and storyboarding use.
Bottom line
Black Forest Labs is a small, well-funded lab executing a specific bet: give away enough model weight to own the ecosystem, sell API access and commercial licences to everyone who needs to ship, and use the resulting scale to fund a jump from image generation into multimodal and physical AI.
Two years in, the bet is working — their models sit under a lot of tools that never mention them. FLUX 3 is where they find out whether the same team can compete a tier up, against labs with far more compute.
If you are building on the FLUX family today, start with a workspace rather than a GPU and step up to weights or API only when volume justifies it.
Sources
- Laying the Foundations for Visual Intelligence — Our $300M Series B (BFL, December 1, 2025) — funding amount, valuation, investor list, named partners
- About Black Forest Labs — team size, locations, founding-team lineage claim
- Martin Scorsese × Black Forest Labs (BFL, June 3, 2026) — advisor announcement and stated use case
- FLUX 3 — Real World Models (BFL, July 23, 2026) — current frontier model and launch plan
- FLUX 3 x mimic (BFL, July 23, 2026) — robotics collaboration and Audi deployment
- FLUX.2: Frontier Visual Intelligence (BFL, November 25, 2025) — FLUX.2 architecture, open-core statement, adoption claims
- FLUX open weights licensing — commercial tiers, synthetic-data licence, market positioning
- BFL API pricing — per-megapixel commercial model
- black-forest-labs/flux on GitHub — open model list and licences
- BFL announcements index — dated release history
Scope note: figures current as of July 29, 2026. Flux 3 AI is an independent creator workspace, not affiliated with or endorsed by Black Forest Labs.
