FLUX.2 [klein] is a new model family that combines image generation and editing in a compact architecture. It's designed for speed, capable of generating or editing images in under half a second on consumer-grade GPUs with as little as 13GB of VRAM. This model family emphasizes real-time performance without sacrificing quality, making it suitable for applications that demand quick visual outputs. The name "klein," meaning "small" in German, reflects both the model's compact size and its minimal latency. The FLUX.2 [klein] lineup includes two main models: the 9B flagship model and the more accessible 4B variant. The 9B model is said to outperform models five times its size in quality while maintaining sub-second inference times. It supports text-to-image generation, image editing, and multi-reference generation within a single model. The 4B model is fully open under the Apache 2.0 license, making it easier for developers to use it locally or on edge devices. There's also a focus on output diversity and customization, especially in the base models, which allow for more extensive fine-tuning and research applications. Quantized versions of the models, FP8 and NVFP4, offer faster performance and reduced VRAM requirements, making them compatible with a wider range of hardware. The FP8 version is up to 1.6 times faster and uses 40% less VRAM, while NVFP4 can be up to 2.7 times faster with 55% less VRAM. Performance benchmarks indicate that FLUX.2 [klein] matches or exceeds competitors like Qwen in quality while providing significant gains in speed and efficiency. This model family represents a significant advancement toward interactive visual intelligence, enabling real-time design tools and interactive content creation.