Black Forest Labs Releases FLUX 3 Image Klein

Black Forest Labs has officially released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model designed for text-to-image generation, image editing, and multi-reference workflows. Capable of handling up to ten reference images at once, the system offers native 4K resolution output, pixel-accurate editing features, and direct support for bounding boxes to precisely guide element placement.

Despite its expanded feature set, the model requires significantly fewer hardware resources than its predecessors. An INT8 quantized build enables the software to run on consumer-grade NVIDIA GPUs featuring as little as 8 GB of VRAM, rendering a full image in under 9 seconds on an RTX 4070 graphics card. The model weights are distributed under the FLUX Kommunity License, making them free for commercial deployment by enterprises earning less than $5 million in annual revenue. For cloud execution, Black Forest Labs provides access via API at a rate of $0.009 per 1,000 images, along with availability on Playground, Hugging Face Spaces, and local frameworks like ComfyUI and Diffusers.

Internal evaluation reveals that FLUX 3 Image Klein earned the top Elo rating among open models, outperforming closed competitors like Ideogram V4.5 as well as the developer's own FLUX.2 [pro]. Compared to the 9-billion-parameter FLUX.2 [klein], the new architecture achieves a 91-point Elo increase while utilizing 38 percent fewer parameters. Developers can customize the system immediately, as day-one LoRA training is supported through the Ostris AI Toolkit and NVIDIA NeMo AutoModel.

For machine learning practitioners and creative developers, this release dramatically lowers the hardware barrier for high-resolution visual AI. Advanced capabilities like precise bounding box layout control and multi-image conditioning no longer demand expensive enterprise cluster infrastructure, enabling local workstation setups and independent teams to deliver production-grade output efficiently.