Black Forest Labs Releases FLUX 3 Image Klein: 5.6B Open-Weight Model Runs on 8GB VRAM

German AI company Black Forest Labs released its new open-weight image generation model, "FLUX 3 Image Klein," on October 15. The model features 5.6 billion parameters and supports text-to-image generation, image editing, and multi-reference generation using up to 10 reference images.

According to Black Forest Labs' internal evaluation, FLUX 3 Image Klein outperformed larger open-weight models, including Qwen-Image 2.1 (7B) and Krea 2 (12.9B), in three areas: text-to-image, multi-reference generation, and image editing. Compared to the previous generation FLUX.2 [klein] 9B, the new model reduces parameter count by 38% while improving Elo ratings by 91 points for text-to-image and 70 points for image editing.

The model is available as a base version for fine-tuning and a step-distilled version for low-latency generation, with FP8/INT8/NVFP4 quantized versions co-developed with NVIDIA. The INT8 version's weights are approximately 5.6GB, allowing users to run the model via ComfyUI or Diffusers on GeForce RTX graphics cards with 8GB of VRAM. Official data indicates that combining the INT8 version with the distilled model can generate a one-megapixel image in about 4 seconds on an RTX 5090 and under 9 seconds on an RTX 4070.

API pricing starts at $0.009 per 1K image. The license is the "FLUX Kommunity License v.1.0," which allows individuals and companies with annual revenue under $5 million to use generated images commercially, but prohibits commercial fine-tuning or redistribution of the model itself.