FLUX 3 Image Klein: open weights, 5.6B parameters

Black Forest Labs has released FLUX 3 Image Klein, an open-weights text-to-image model carrying 5.6 billion parameters. The model is notable for combining what the company describes as frontier image quality with the practical advantages of open weights - namely, the ability to fine-tune, customize, and run the model independently of a hosted API. This puts Klein in a distinct position relative to proprietary alternatives that restrict access to model internals.

Klein is the first FLUX 3 model to support multi-reference generation, a feature that allows users to provide multiple reference images to guide a single output. This is useful for tasks like style blending, subject consistency across images, and character-driven generation. Alongside this, the model supports image editing, rounding out a feature set that goes beyond standard text-to-image inference.

At 5.6 billion parameters, Klein sits in a range that balances capability with practicality. Larger models often demand hardware that is out of reach for individual researchers or smaller teams, while very small models can struggle with complex compositions, accurate text rendering, and fine detail. A 5.6B open-weights model at this quality tier gives the community a meaningful baseline to build from - whether through direct use or as a starting point for fine-tuning on specific domains or aesthetics.

Black Forest Labs has been building out the FLUX model family since the release of FLUX.1 in mid-2024, which itself drew significant attention for its image quality and its open-weights variants. FLUX 3 Klein continues that trajectory, and the addition of multi-reference support in particular suggests the team is broadening the model's utility beyond single-prompt generation toward more compositionally complex workflows. Details on licensing and weight access are available through the BFL website.