FLUX.2 Klein: A Compact 4B Open-Source Image Model

Black Forest Labs has released FLUX.2 klein, a new 4-billion-parameter open-source model for text-to-image generation and editing. Released under a permissive Apache 2.0 license, the model provides a new, highly efficient option for developers and researchers working with generative AI.

An Alternative to Latent Diffusion

Unlike many popular image models that rely on latent diffusion (the architecture behind Stable Diffusion), FLUX.2 is built on a newer architecture combining flow matching with multimodal transformers. This design aims to achieve high-quality image generation with significantly faster inference speeds, making it a compelling alternative for applications where performance is critical.

The "klein" version is a smaller, more accessible variant of the FLUX.2 architecture. Its relatively compact 4B parameter size makes it easier to run and fine-tune on consumer-grade hardware. As a base model, it's designed to be a strong foundation for a wide range of custom applications, from creative tools to specialized image editors.

The model weights and card for FLUX.2 klein base 4B are available on Hugging Face for immediate use, allowing the community to begin experimenting with this promising new architecture.