Flux

Black Forest Labs (BFL) was founded in 2024 by Robin Rombach, Andreas Blattmann, and Patrick Esser, the researchers behind the original Stable Diffusion. Their Flux model family uses a Rectified Flow Transformer architecture rather than the traditional diffusion approach, which means it learns more direct paths from noise to image. The practical effect is faster generation and sharper control over output.

The current FLUX.2 family (released November 2025, with BFL raising a $300M Series B at a $3.25B valuation in December 2025) includes FLUX.2 Max for maximum quality, FLUX.2 Pro for production-grade work, and FLUX.2 Dev as the open-weights option available on Hugging Face. FLUX.2 Max supports up to 4-megapixel resolution, 10 reference images simultaneously for character consistency, and real-time web context for grounded generation. The FLUX.2 klein family added faster, interactive-speed variants in January 2026.

For builders, Flux sits at an interesting middle ground: the Dev variant is genuinely open-weights and fine-tunable, while the commercial tiers are accessible via API through BFL directly and through aggregators like Replicate, fal.ai, and Hugging Face Inference. Adobe Firefly integrated FLUX.2 as one of its partner models, which is a signal of how seriously the professional creative tool market takes it as an alternative to both Midjourney and DALL-E.