Wanted to quickly flag a security risk. Since Faceswap loads PyTorch/TF models, users downloading untrusted model files (.pth, .h5, .pkl) from random forums could get hit with arbitrary code execution via insecure deserialization (like malicious pickle payloads). This is especially risky if users run Faceswap as admin/root to bypass the GPU permission issues mentioned in INSTALL.md. Quick mitigations we could add: UI/CLI Warning: Add a quick note warning users to only load models from trusted sources. Add a SECURITY.md: You list deefakesrepo@gmail.com in the Code of Conduct, but adding a standard SECURITY.md file gives researchers a proper, expected way to report things privately. Doc Tweaks: Reiterate in the docs to stick to Conda/venvs and never run as root/admin. I know this is fundamentally a PyTorch/TF limitation, but adding some guardrails would really help protect the community...