Xialei FLUX local setup toolkit — unofficial beta

Uses Black Forest Labs FLUX.1 schnell (Apache 2.0).
Official model/access terms: https://huggingface.co/black-forest-labs/FLUX.1-schnell

1. Install Python 3.10–3.12 with pip from https://www.python.org/downloads/ . On Windows include the Python launcher.
2. Sign in to Hugging Face and accept the official model access conditions. Create a read-only token if needed.
3. Extract this whole folder into a writable location. Keep all files together.
4. Windows: open Start FLUX.bat. macOS: open Start FLUX.command. Linux: run start-flux.sh with your shell. If your OS blocks the unsigned launcher, use Python to open setup.py instead; do not disable security protections.
5. Type YES to allow package installation. Enter your Hugging Face token when prompted (hidden input), or leave blank if already logged in. The official Hugging Face library stores the token on your computer, not on Xialei.
6. A local browser interface opens. Describe an image, approve the model download, then select Generate image. Save the result with its download button.

Packages are installed in .venv inside this folder. Models are cached by Hugging Face in your home folder; deleting this toolkit alone does not remove the model cache or saved login. Subsequent launches reuse installed packages and model files. Internet is needed for setup and the first model download; cached generation runs locally. No API payment required.

Allow at least 50 GB free disk space. 32 GB RAM and a high-memory NVIDIA GPU (roughly 16 GB VRAM or more with CPU offload) are a starting recommendation, not guaranteed compatibility. Apple Silicon uses Metal when available, sharing system memory. CPU-only mode is much slower and can require over 64 GB RAM. This is an unquantized model, not a lightweight Chromebook installer. Intel Macs use CPU mode. NVIDIA acceleration requires compatible drivers and a CUDA-enabled PyTorch installation; default packages may use CPU depending on platform. GPU drivers are not installed by this toolkit.

The app binds only to 127.0.0.1; no public sharing link is enabled. It does not deliberately send prompts to an image-generation API. Third-party package/model downloads contact PyPI and Hugging Face. Model loading may take a long time. To stop, close the launcher. Failed downloads, insufficient memory, missing approval or unsupported hardware can prevent generation. Reduce other memory use, check access, and reopen the launcher.

Independent wrapper, not affiliated with or endorsed by Black Forest Labs. No model weights or Python runtime bundled. Wrapper and app have not been tested on physical Mac/Windows/Linux hardware. Review upstream licensing and acceptable-use conditions before use.
