Black Forest Labs' open-weight image model FLUX 3 Image Klein can run on consumer GPUs with only 8 GB of video RAM.

Meta Description: Black Forest Labs releases FLUX 3 Image Klein, a 5.6-billion-parameter open-weights image model with native 4K, editing, ten references, and low VRAM.

Tags: FLUX 3 Image Klein, Black Forest Labs, open-weights AI, image generation, NVIDIA

Featured Image: Abstract technology style image showing a neural network grid, layered image frames, and a visual editing interface, with no real people, no company logos, and no text.

Background

Black Forest Labs has expanded its FLUX line with FLUX 3 Image Klein, an open-weights model aimed at text-to-image generation, image editing, and generation guided by reference images. The release places the company in a competitive field of AI image tools, where model access, parameter count, editing precision, and deployment cost shape what creators and developers can build. The new model is described as having 5.6 billion parameters and supports up to ten reference images, making it relevant for workflows that need multiple visual references.

Development Process

The model's reported capabilities cover three connected tasks: creating images from text, editing existing images, and using up to ten reference images to steer generation. It supports native 4K output, pixel-accurate editing, and bounding boxes for element placement. These features indicate a design focused on controllability rather than only broad scene creation.

Black Forest Labs says internal testing shows FLUX 3 Image Klein reaching the highest Elo rating among all open-weights models tested. Elo is a comparative score used to rank outputs against other models. The company also says the model beat closed competitors, including Ideogram V4.5, and its own earlier FLUX.2 [pro] system. Compared with FLUX.2 [klein], FLUX 3 Image Klein gained 91 Elo points while using 38 percent fewer parameters than that model's 9 billion.

Latest Progress

For deployment, the company points to an INT8 build that runs on NVIDIA GPUs with just 8 GB VRAM. In the reported setup, an image is generated in under 9 seconds on an RTX 4070. The release therefore highlights a low-VRAM path for running the model on consumer GPUs.

Access is available through multiple channels. The weights are released under the FLUX Kommunity License, which allows free commercial use by companies making under $5 million annually. Users can also call the model through the BFL API at $0.009 per 1,000 images, use it in Playground, access it through Hugging Face Spaces, or run it locally with ComfyUI and Diffusers.

LoRA training is supported from day one through Ostris AI Toolkit and NVIDIA NeMo AutoModel. That matters for developers and artists who want to train compact adaptations of the base model without replacing the full 5.6-billion-parameter release.

Current Status

FLUX 3 Image Klein is now presented as a released, open-weights option with editing, reference-image generation, native 4K support, and low-VRAM INT8 deployment. The current stage is distribution and adaptation: the model is available through hosted and local routes, while the FLUX Kommunity License defines who can use it commercially for free. The next meaningful signal will be how the reported internal Elo advantage, 8 GB VRAM build, and day-one LoRA support translate into real use across the BFL API, Playground, Hugging Face Spaces, ComfyUI, and Diffusers.