A zero-price configuration pack for a self-hosted SeFi image engine on a local GPU host. It is provided only for test and non-commercial evaluation.
This archive contains no checkpoint weights, Hugging Face token, cache, output, log or runtime database. The gated SeFi RL and Turbo checkpoints are governed by their upstream CC BY-NC 4.0 terms. Commercial mode is rejected by the installer.
Configure
Run bash install.sh. The installer asks for:
- Turbo or RL checkpoint;
- an absolute content-cache directory;
- an existing token file with mode
0600(the token value is never copied); - exact acceptance of the non-commercial checkpoint license;
- a localhost-only listener;
- an absolute outputs directory.
It writes a mode-0600 environment file. Checkpoints must be preloaded
separately by an administrator. Inference is offline after preload. The pack
refuses a lan configuration because it cannot itself verify the required host
firewall and bearer-authentication evidence.
For unattended test setup, pass the corresponding long options shown by
python3 configure.py --help. Passing --usage commercial always fails.
The included systemd unit is a reference and is not enabled automatically. Deploying a LAN listener is a separate reviewed host procedure; this pack never creates one. Human review is required before installing any generated asset.