Model Releases
Black Forest Labs Releases FLUX.2 Klein Family of Fast Open Weights Image Models
Black Forest Labs introduces FLUX.2 Klein, a family of distilled image generation models optimized for sub-second inference on consumer-grade GPUs.
January 15, 2026
Model ReleaseFLUX.2 Klein
Introduction
On January 15, 2026, Black Forest Labs (BFL) disrupted the generative AI landscape with the release of the FLUX.2 Klein family. While the industry has long been chasing higher parameter counts to achieve visual fidelity, BFL has taken a different approach: extreme distillation for real-time utility. The Klein series represents a strategic pivot toward high-speed, local-first deployment without sacrificing the structural integrity that made the original FLUX models famous.
The 'Klein' designation—a name that will later resurface in the FLUX 3 era—signifies a specific architectural goal: minimizing the latency between prompt and pixel. By leveraging advanced distillation techniques from the massive FLUX.2 base models, BFL has delivered a suite of models capable of sub-second generation on standard consumer hardware, making high-quality AI imagery accessible to developers building latency-sensitive applications.
- Release Date: January 15, 2026
- Primary Focus: Speed and local deployment
- Architecture: Distilled Rectified Flow Transformers
- Availability: Open weights via Hugging Face
Key Features & Architecture
The FLUX.2 Klein family is bifurcated into two distinct model sizes, each optimized for specific developer workflows. Both models utilize a rectified flow transformer architecture, which has proven more efficient for high-resolution synthesis than traditional diffusion processes. The distillation process ensures that the models inherit the complex prompt adherence and spatial reasoning of the larger FLUX.2 base models while operating at a fraction of the computational cost.
The 4B variant is designed for agility, offering text-to-image capabilities alongside multi-reference image guidance. This makes it an ideal candidate for real-time creative tools. The 9B variant, conversely, is the quality-focused powerhouse of the family. It supports outputs up to 4 megapixels and can ingest up to four reference images, allowing for sophisticated style transfer and character consistency tasks that were previously reserved for much larger, slower models.
- FLUX.2 Klein 4B: Optimized for speed, multi-reference guidance, and text-to-image.
- FLUX.2 Klein 9B: High-fidelity output up to 4MP, supports up to 4 reference images.
- Quantization Support: Native FP8 and NVFP4 variants for maximum efficiency.
- Core Tech: Distilled Rectified Flow Transformers.
Performance & Benchmarks
In terms of raw performance, the Klein models represent a massive leap in throughput-per-watt. On a standard NVIDIA RTX 4090, the 4B model achieves sub-second generation, a feat that puts it in a different class than the original FLUX.2 base models. While the base models prioritize absolute peak fidelity, the Klein series optimizes the 'quality-to-latency' ratio, making it the preferred choice for interactive applications.