Black Forest Labs Releases 9B FLUX.2 Image Model

Black Forest Labs has introduced FLUX.2 klein, a new 9-billion-parameter text-to-image model that represents a significant new entry in the open-source generative AI space. The model is capable of both generating and editing images from text prompts and is now available for download on Hugging Face.

A New Architecture for Speed

Unlike many existing diffusion models that operate in a compressed latent space, FLUX.2 is a Diffusion Transformer (DiT) that works directly with image tokens. Its core innovation is a novel multi-head cross-attention block that processes text and image features jointly. According to its creators, this design allows for significantly faster training and inference compared to models like Stable Diffusion 3, which process text and image modalities in separate blocks.

The release of FLUX.2 provides developers and researchers with a powerful alternative to established models like SDXL and newer architectures like SD3. Black Forest Labs claims the model achieves state-of-the-art performance, outperforming other open models in prompt alignment and image quality while maintaining high efficiency.

FLUX.2 klein is released under a permissive Community License that allows for commercial use, with some restrictions aimed at preventing misuse. This combination of high performance and an open, business-friendly license positions FLUX.2 as a strong contender for powering a new generation of creative applications.