Black Forest Labs— Generative image and video AI models.

Black Forest Labs is a generative AI lab building state-of-the-art text-to-image and video models, including the open FLUX family, from the team behind Stable Diffusion.

Visit website
Founded
2024
Headquarters
Freiburg im Breisgau, Germany
Legal name
Black Forest Labs Inc.
Legal domicile
Delaware, United States

Highlights

Notable milestones from Black Forest Labs’s journey — funding rounds, launches, and other moments worth knowing.

recognition

A 90-person frontier lab gets a seat at the G7 AI table

Co-founder and CEO Robin Rombach joined President Macron, President Trump, Chancellor Merz, President von der Leyen and other G7 leaders to argue for open, responsible AI development. BFL framed itself as a 90-person Germany-and-US lab whose team had co-developed three of the five most popular open generative AI models of all time.

Robin Rombach seated at a G7 AI roundtable with national and European leaders.
Rombach used the G7 forum to make the case for open visual AI as critical infrastructure.
Source
partnership

Envato makes FLUX core creative infrastructure

Envato disclosed that FLUX models had generated more than 51 million images on its creative platform and now account for about a quarter of total image generation volume. The partnership gives Envato early access to new BFL releases and private endpoints, letting it put FLUX.2 into production on launch day.

Source
partnership

FLUX.2 ships on consumer laptops through ASUS and NVIDIA

Black Forest Labs made FLUX.2 [klein] 4B the first FLUX model to ship on consumer hardware, preloaded in ASUS's MuseTree app on next-generation ProArt RTX laptops. The on-device build targets sub-5-second generation on 8GB of VRAM without an API call or internet connection.

ASUS MuseTree running an on-device FLUX.2 generation workflow with a tiger astronaut image and gallery.
FLUX.2 [klein] running locally inside ASUS MuseTree on ProArt RTX laptops.
Source
launch

FLUX.2 [klein] makes image generation interactive

BFL shrank FLUX.2 into fast 4B and 9B variants that unify generation and editing, run on as little as 13GB of VRAM, and can return images in under a second. The 4B model shipped with Apache 2.0 weights, pushing the open side of the FLUX line back toward consumer hardware.

Source
funding

$300M Series B at a $3.25B valuation

Salesforce Ventures and a16z's Anjney Midha co-led, with NVIDIA, General Catalyst, and Temasek joining. The round took total funding past $430M and made BFL the best-capitalised open image lab in Europe — all within 18 months of launch.

Round
Series B
Raised
$300M
Valuation
$3.25B
Led by
Salesforce Ventures, Andreessen Horowitz
Black Forest Labs' Series B announcement graphic reading "Our $300M Series B".
The December 2025 raise that valued the lab at $3.25B.
Source
launch

FLUX.2 reaches for production realism

The second generation closed much of the gap with the leading closed models — handling dense prompts, legible text, and consistency across multiple images well enough to lean on for real product work.

A range of photorealistic and illustrated scenes generated by FLUX.2, including readable signage.
Further FLUX.2 samples spanning photography, illustration, and product shots.
Early samples leaned into the hard cases — legible text, crowded scenes, consistent lighting.
launch

FLUX.1 Kontext puts editing inside the model

Rather than generate from scratch, Kontext edits an image from a plain text instruction — change the background, restyle a character, fix a detail — and keeps the subject recognisable across a chain of edits.

One subject edited through successive text prompts while keeping its identity intact.
Demos showed the same subject surviving a chain of text-only edits — exactly where earlier editors fell apart.
launch

FLUX1.1 [pro] doubles down on speed

A rebuilt flagship that generated up to six times faster than the original while sharpening detail. Tested anonymously as "blueberry", it quietly took the top spot on the Artificial Analysis image leaderboard before BFL revealed it was theirs.

partnership

Became the image engine inside Grok

Two weeks after launch, xAI wired FLUX.1 into Grok on X. Overnight it reached millions — and Grok's loose guardrails dropped it straight into the public fight over what AI image tools should refuse to make.

launch

FLUX.1 arrives, from the Stable Diffusion team

The researchers who created Stable Diffusion left Stability AI and reappeared months later with FLUX.1 — three models at once, from a fast Apache-licensed version to a closed flagship. Output quality drew instant comparisons to Midjourney and DALL·E.

A grid of varied images generated by FLUX.1 [pro] — portraits, landscapes, and rendered text.
A grid of varied images generated by FLUX.1 [dev].
A grid of varied images generated by FLUX.1 [schnell].
The closed [pro] tier turned out the cleanest results of the three at launch.
Source
funding

A $31M seed to fund the launch

Andreessen Horowitz led an unusually large seed for a months-old European lab, valuing it at $150M and bankrolling the compute behind the FLUX.1 release.

Round
Seed
Raised
$31M
Valuation
$150M
Led by
Andreessen Horowitz