FLUX.2 [Klein]: Towards Interactive Visual Intelligence (bfl.ai)

The release of FLUX.2 [klein], a new family of image models, marks a significant advancement in real-time visual intelligence, combining state-of-the-art image generation and editing within a compact architecture. With sub-second inference times, it allows users to generate or edit images in under 0.5 seconds using consumer hardware with as little as 13GB VRAM. This model family stands out due to its exceptional performance in text-to-image generation and multi-reference editing, often exceeding the output of much larger models while being accessible for developers with open weights for customization. FLUX.2 [klein] includes both a flagship 9B model, which balances quality and latency across various tasks, and a more accessible 4B variant, running efficiently on widely available GPUs. Notably, the collaboration with NVIDIA has resulted in quantized FP8 and NVFP4 versions, providing even faster performance with lower VRAM requirements. This release not only enhances usability and accessibility in diverse applications such as real-time design tools and content creation but also signifies a move towards evolving AI systems that can interactively generate and iterate visual content, appealing to creators and developers alike.