rootlocalghost/FLUX.2-klein-4B-FP8

High Realism

Editing Exmples

Extra Generation Examples

FLUX.2-Klein 4B - FP8

This repository is a derivative of the Black Forest Labs FLUX.2-Klein 4B image model, adapted for personal use and local experimentation. All original licensing terms and usage conditions from the upstream Black Forest Labs project are retained and honored.

Overview

FLUX.2-Klein 4B is a compact image generation and editing model (text-to-image / image-to-image) designed for lightweight inference while retaining strong image quality in the 4B parameter class. The repository contains all necessary artifacts for local use. This specific model is Quantized to FP8 precision for even more efficient use. Both the Transformer Model and the Text Encoder has been quantized to FP8 precision.

Inference

Inference scripts are available in the rootlocalghost/FLUX repository. These scripts are designed for Intel ARC GPUs using PyTorch XPU and have been validated on an Intel ARC A770 GPU.

Recommended environment

• OS: Windows 10/11 or Linux with oneAPI-compatible drivers and oneapi-level-zero installed
• GPU: Intel ARC
• Python: 3.12
• PyTorch: 2.10.0 with XPU support
• oneAPI/oneAPI DPC++ runtime installed

Validation

• This configuration is tested on a single Intel ARC A770 in the local environment.
• Report any compatibility issues in the original repo issue tracker and include environment details.

Notes

• This repository is intended for users familiar with local model hosting and inference pipelines.
• Verify and comply with the original FLUX.2 and Black Forest Labs license terms before redistribution.

License

Apache-2.0 (inherits terms from upstream model release)