Ethics & integrity
A powerful tool deserves clear guardrails
Face swapping is dual-use. It can power legitimate creative and accessibility work - and it can be pointed at people who never consented. This page is a straight, honest read on the built-in integrity features, why they aren't a hard wall, and how to use the tool responsibly.
What the project ships
The upstream Facefusion project includes features for content analysis and integrity checking. Like most checks that run inside a client-side, user-editable tool, they exist to encourage safe use, not to enforce it. The exact behavior of these checks also changes between releases, and third-party patches to the project exist that can alter or disable them.
Patches change the guarantees
Because the source is open and user-editable, a patched build can remove or change the built-in content-analysis and integrity checks entirely. That means you should never treat a check that ships in the default build as a reliable enforcement boundary. If a platform or workflow depends on those checks holding, verify the actual build you're running - and expect the guarantees to vary from the default.
Practice that keeps it human
- Get consent. The people in the source face and the target footage should know and agree to how their likeness is used.
- Label synthetic media. Clear the swap in a caption, watermark, or metadata wherever the output leaves your machine.
- Keep provenance. Retain the original source and target so edits are auditable and reversible, not a one-way alteration.
- Know the detection ecosystem. Deepfake detection is improving but is not perfect; don't assume a labeled or low-stakes context is what the output will be used for.
A note on version sensitivity
Everything on this site - providers, models, flags, checks - lives in a specific snapshot of the codebase. Verifying against your installed release and reading the linked source keeps both your workflow and your understanding accurate.