Black Forest Labs announced the release of FLUX 3 Image Klein, a 5.6 billion parameter open-weights model for text-to-image generation, editing and multi-reference generation. The model offers native 4K generation, pixel-perfect editing that preserves unedited regions and bounding box control for precise element placement.
What happened?
According to the company's official announcement, FLUX 3 Image Klein shares capabilities with the larger FLUX 3 Image model and was trained leveraging learnings from images, video and audio. It comes in two versions: base (for fine-tuning and research) and distilled (for fast generation in 16 steps). There are FP8, INT8 and NVFP4 builds developed with NVIDIA support.
In internal evaluations, the model leads in text-to-image, multi-reference and editing among open weights, outperforming even closed models like Ideogram V4.5 and FLUX.2 [pro] in some tests. Compared to the 9B FLUX.2 [klein], it gains 91 Elo in text-to-image and 70 in editing with 38% fewer parameters.
Why does this matter?
The big differentiator is accessibility on consumer hardware. Weights in BF16 take about 11.2 GB and in FP8 5.6 GB. With the INT8 build, it runs on NVIDIA GeForce RTX GPUs with just 8 GB of VRAM via ComfyUI or Diffusers. On an RTX 5090 it generates a 1 MP image in about 4 seconds; on an RTX 4070 in under 9 seconds (distilled version). On GB300 with NVFP4 it is under 0.6 seconds.
The FLUX Kommunity license allows free commercial use for individuals and companies with annual revenue under $5 million. Larger companies need licensing. There is day-one support in ComfyUI, Diffusers, vLLM-Omni and other tools, plus LoRA recipes.
What changes in practice?
Creators, developers and small businesses can generate and edit high-quality images locally, without relying exclusively on expensive APIs or cloud. This reduces costs, improves privacy and allows customization via fine-tuning. API pricing starts at $0.009 per 1K image.
Official source: Black Forest Labs blog.
Image credit: Black Forest Labs — Source: Official announcement
By GeekikiBot
