Black Forest Labs has introduced a new series of AI image generators called FLUX.2 [klein]. This model family focuses on high speed and low hardware requirements. The company states that these models unify image generation and editing within a single architecture. According to the announcement, the systems produce images in less than one second on modern hardware.
The name klein refers to the German word for small. It describes both the compact size of the models and the reduced waiting time for users. Black Forest Labs claims these models deliver performance typically seen in much larger systems. The technology targets applications that require real-time interaction, such as design tools or automated visual reasoning.
Technical capabilities and variants
The release includes two primary versions based on their parameter counts.
- FLUX.2 [klein] 9B: This is the flagship small model. It uses a 9 billion parameter flow model and an 8 billion parameter text embedder. Black Forest Labs claims it matches or exceeds models five times its size in speed and quality.
- FLUX.2 [klein] 4B: This version is designed for accessibility. It can run on consumer graphics cards with 13GB of video memory.
The models support several advanced functions. Users can create images from text or edit existing images using one or more reference pictures. VentureBeat reports that the models also accept specific hex color codes for precise color control. Additionally, they can process structured instructions in formats like JSON to help developers build automated pipelines.
Licensing and accessibility
The company uses different licenses for the two versions. The 4B model is available under the Apache 2.0 license. This allows developers and businesses to use the technology for commercial purposes without paying fees. VentureBeat notes that this positions the model as a competitor to other open systems.
The 9B model and its development versions use the FLUX Non-Commercial License. This means individuals and researchers can download and use the weights for free, but commercial use requires a separate agreement. Black Forest Labs has released the model weights on Hugging Face and the code on GitHub.
To improve performance further, the company collaborated with NVIDIA to create optimized versions. These versions require less memory and run faster on specific hardware. According to the developers, the most optimized variant is up to 2.7 times faster than the standard version.
Sources: Black Forest Labs, VentureBeat
