Open Source FLUX 3 Image Klein Neural Network for Home

German company Black Forest Labs has released the open-source FLUX 3 Image Klein neural network for image generation, which runs on consumer graphics cards without using cloud computing.

Key Model Features

• The model contains 5.6 billion parameters and can run locally on a single GPU, such as an RTX 4070, with generation taking less than 9 seconds.

FLUX 3 Image Klein solves three main tasks: generating images from text descriptions, editing them with precise adjustments in a single pass, and assembling an image from multiple references.

Users can define the frame layout by specifying frames and descriptions for each object on a 1000×1000 pixel grid.

Training and Application Features

The neural network was trained on data from images, videos, and audio, allowing it to handle various styles: from photographs and engravings to pixel art, anime, and infographics.

1. The non-distilled version is intended for fine-tuning and scientific research.
2. The distilled version is optimized for fast image generation.

Model weights are available on Hugging Face, and operation is performed through the ComfyUI interface. Premium API access provides prompt enhancement and 4K resolution generation features.

Having open weights does not mean free commercial use. Integrating the model into a commercial product requires a separate license, the terms of which are negotiated directly with the developer's sales department.