Introducing 'FLUX.2[klein],' an AI that generates images at speeds of less than one second, capable of running on a home PC with 13GB of VRAM.



German AI startup Black Forest Labs has released a new high-speed, integrated AI model, ' FLUX.2[klein], ' from its image generation AI model family, '

FLUX.2 .' The model is relatively small in size and prioritizes low latency, completing image generation in less than one second.

FLUX.2 [klein]: Towards Interactive Visual Intelligence | Black Forest Labs
https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence






FLUX.2[klein] is the fastest image model in Black Forest Labs' image generation AI history, integrating generation and editing into a single, compact architecture, enabling end-to-end inference to run in less than one second while achieving cutting-edge quality. The 'klein' in the model name comes from the German word for 'small,' reflecting the model's compact model size and minimal latency.

FLUX.2[klein] is built for applications that require real-time image generation without sacrificing quality, and it runs on consumer hardware with as little as 13GB of VRAM. Black Forest Labs says that on modern consumer hardware, it can generate or edit an image in less than half a second.

FLUX.2[klein] is available in three main models: 'FLUX.2[klein] 9B,' which is flagship-level, fast, and high-quality; 'FLUX.2[klein] 4B,' the most practical model for consumer GPUs; and 'FLUX.2[klein] Base (9B/4B), a basic model for research and fine-tuning. In addition, 'FP8/NVFP4' quantization models, developed in collaboration with NVIDIA, are also available, which can further reduce VRAM consumption and inference time on RTX GPUs.




The graph below shows the Elo score (vertical axis) indicating image generation quality and the latency (horizontal axis) indicating processing time. It shows that FLUX.2[klein] achieves Elo scores equal to or better than Alibaba's image generation AI model ' Qwen-Image-2512 ' in the Text-to-Image, Image-to-Image, and Multi-Reference tasks, with much lower latency.



The graph below shows Elo score on the vertical axis and VRAM usage on the horizontal axis. FLUX.2[klein] achieves the same or higher Elo score compared to Qwen-Image-2512 and Qwen-Image-2509, while using significantly less VRAM. In particular, in the Multi-Reference task, which uses multiple reference images, FLUX.2[klein] significantly reduces VRAM usage compared to

Qwen-Image-Edit-2511 , while still achieving significantly higher image quality.



Until now, there has been a trade-off between speed and quality, but FLUX.2[klein] achieves both speed and quality by integrating generation and editing into a single model. Black Forest Labs states that 'visual intelligence is entering a new era,' and argues that as the capabilities of AI agents improve, models capable of generating visuals that can respond in real time and rapidly iterate will be needed to keep up.

Black Forest Labs said, 'FLUX.2[klein] is more than just a high-speed model; it's a step towards our vision of interactive visual intelligence. We believe the future belongs to creators and developers, armed with AI that can visualize, create, and iterate in real time. FLUX.2[klein] is a system that enables new categories of applications, including real-time design tools, agent-driven visual reasoning, and interactive content creation.'

You can use FLUX.2[klein] by selecting the model 'FLUX.2[klein] 9B' or 'FLUX.2[klein] 4B' from the following page.

FLUX Playground - Black Forest Labs
https://playground.bfl.ai/image/generate

You can also try out FLUX.2[klein] image generation in the demo below without registering.

FLUX.2 [klein] | Black Forest Labs
https://bfl.ai/models/flux-2-klein#try-demo



Additionally, ComfyUI already supports image generation with FLUX.2 [klein].




Related Posts:

in AI, Posted by log1e_dh