Black Forest Labs has released FLUX.2 [klein], a family of image models designed for real-time generation and editing. The flagship 9B model delivers state-of-the-art quality with inference under half a second, while the 4B version runs on consumer GPUs with just 13GB of VRAM. Both unify text-to-image and multi-reference editing in a single architecture.

This release targets developers building interactive AI tools where speed is critical. The 'klein' name, German for 'small,' reflects the models' compact size and minimal latency. By open-sourcing the 4B variant under Apache 2.0 and providing accessible licenses for the 9B version, the company aims to lower barriers for local development and edge deployment.

The move responds to growing demand for AI agents that can generate and modify visuals in real time. FLUX.2 [klein] benchmarks claim it matches or exceeds larger competitors on quality while being significantly faster. The company also released quantized versions in collaboration with NVIDIA, further optimizing performance for RTX GPUs. What comes next is wider integration into creative and agentic workflows.