Black Forest Labs FLUX — The Former Stable Diffusion Team

When the Founders Left to Build What They Wanted to Create

Black Forest Labs was founded by the developers who created Stable Diffusion, the most famous open image generation model. After leaving Stability AI, they didn’t stop — they founded a new company in Germany and launched FLUX, a model said to have surpassed Stable Diffusion in everything that distinguished it. Within one and a half years, they raised $450 million at a $3.25 billion valuation, and signed a $140 million deal with Meta.

Founding Story: Experts Starting Fresh

Black Forest Labs was founded in April 2024 in Berlin by a team of the most prominent researchers who formed the backbone of the image generation field:

• Robin Rombach (CEO): The lead author of the Latent Diffusion Models paper that established Stable Diffusion.
• Andreas Blattmann (CTO): PhD from Ludwig Maximilian University.
• Patrick Esser: Another lead author in Stable Diffusion works.

The 20-person team (70% of whom hold PhDs) knew exactly what they wanted to build: a better, more controllable model with a license enabling true commercial use.

They raised $31 million in August 2024 in a seed round led by Andreessen Horowitz, and on nearly the same day launched FLUX.1 — a clear signal that the company came ready.

Main Products and Models

FLUX.1 (August 2024)

The launch that stunned the community. A 12 billion parameter model with a new Flow-matching Transformer architecture, with a text understanding component from Mistral-3. Included three versions:

• FLUX.1 [pro]: API only, highest quality.
• FLUX.1 [dev]: Open weights at 12 billion parameters for non-commercial use.
• FLUX.1 [schnell]: Apache 2.0 licensed for commercial use, faster but lower quality.

According to blind tests at launch, FLUX.1 surpassed Midjourney v6 in text accuracy and visual consistency.

FLUX.1 Kontext (2025)

A qualitative leap in contextual editing: the model can modify an existing image based on text instructions while maintaining visual identity consistency. Adobe added this model in October 2025 to Photoshop Beta within the Generative Fill feature — giving BFL widespread reach among professional designers.

FLUX.2 (November-December 2025)

The next generation with resolution up to 4K and ability to use 10 reference images to maintain visual style:

• FLUX.2 [max]: Highest quality.
• FLUX.2: Balance between quality and speed.
• FLUX.2 [klein] (January 2026): The fastest FLUX model — image generation in less than a second on supported devices, capable of local operation and fine-tuning.

Achievements and Numbers

Funding

• Seed (August 2024): $31 million led by a16z with participation from General Catalyst and Nvidia.
• Series A (previously unannounced): Revealed with Series B, led by a16z with BroadLight Capital, Creandum, Earlybird VC, and Northzone.
• Series B (December 2025): $300 million led by Salesforce Ventures and Anjney Midha (AMP), with participation from a16z, Nvidia, General Catalyst, Temasek, Bain Capital Ventures, Canva, and Figma Ventures — $3.25 billion valuation.
• Total: Over $450 million.

Commercial Partnerships

• Meta (September 2025): $140 million contract for using FLUX in Meta’s creative products.
• Adobe: Integration of FLUX.1 Kontext [pro] in Photoshop Beta, Project Concept, and Adobe Express.
• Canva and Figma Ventures: Both investors and strategic partners.
• Snap, VSCO, ElevenLabs, Vercel, Picsart: Among companies integrating FLUX into their products.
• Microsoft Azure AI Foundry: Official distribution via Azure platform.
• xAI/Grok: Adopted FLUX as a primary model for image generation.

Community

• Over 1 million inferences in the first week of FLUX.1 [pro] launch.
• FLUX.1 Kontext hackathon attracted 58 projects.
• FLUX models available via Hugging Face, Replicate, and fal.ai.

Competition and Challenges

FLUX competes with:

• OpenAI’s DALL-E 3: Integrated in ChatGPT with a massive consumer base.
• Midjourney: Enjoys high artistic quality and strong community loyalty.
• Stability AI’s Stable Diffusion 3: The direct open competitor.
• Adobe’s Firefly: Integrated in Creative Cloud with legal IP protection.

BFL’s essential advantage is the dual licensing: open source for the community, and a clear commercial path for enterprises. This distinguishes it from Midjourney (closed) and Stability AI (open but with a troubled commercial model).

The biggest challenge: maintaining performance leadership in a market that updates every six months. Google and OpenAI have computing and data resources that exceed what a company with $450 million in funding can bear.

Future Vision 2026–2027

SOTA (internal name for video model): BFL confirmed in October 2025 that it is developing a text-to-video model, the logical next expansion in its journey.

Expansion in enterprise uses: With a self-service commercial licensing portal and unified annual contracts, BFL is building an organized enterprise revenue layer.

Specialized models: FLUX.2 [klein] indicates a direction toward fast models suitable for edge devices and applications needing immediate response.

Analytical Conclusion

Rarely does a researcher leave a successful team to create something better than that same team in the same field. Black Forest Labs may have done so. FLUX.1 surpassed Stable Diffusion in image quality and text accuracy, and the $140 million Meta deal and Adobe integration prove enterprise acceptance.

The striking number: in less than 18 months from founding, they raised $450 million and signed commercial contracts worth hundreds of millions. The open question is whether they can maintain performance leadership with resources much smaller than Google and OpenAI, and whether their bet on video generation will succeed in time.