Black Forest Labs releases FLUX.3 Image Klein generative models

Black Forest Labs has launched its FLUX.3 Image Klein artificial intelligence model collection, adding faster inference options and extended visual control capabilities to its generative portfolio ^[1].

The release provides text-to-image synthesis alongside multi-reference image-to-image generation and explicit layout control ^[1]. These additions allow operators to guide compositional structures and feed multiple reference images into a single generative workflow ^[1].

To accommodate varying speed and compute requirements, Black Forest Labs structured the release into two distinct offerings: a primary base model and a specialized time-step distilled checkpoint ^[1]. According to release figures, the distilled model can generate a 1024 x 1024 resolution image in approximately two seconds on an Nvidia H100 GPU ^[1].

The dual-checkpoint release reflects ongoing engineering efforts across the generative ecosystem to reduce compute latency without forfeiting complex structural control ^[1]. By pairing a full-capacity base model with a distilled checkpoint optimized for two-second generation cycles, the company offers developers a direct tradeoff between pure foundational generation capacity and production-oriented inference speed ^[1].