May 27, 2026 · 5 min read
Offline, private face swap: what "100% on-device" actually means
"Runs on-device" gets used loosely in AI-tool marketing. Sometimes it means the interface is local but a cloud API still does the actual work. Sometimes it means the first pass is local and the "quality" pass happens on a server. It's worth being precise about what it actually means for a face-swap and voice-change tool, because the answer changes who else might see your source photos and voice clips.
What "100% on-device" means for NZCAM
NZCAM's entire pipeline (face detection, the swap itself, expression restoration, voice changing, every realism and style feature) runs on your own GPU, in the app, on your machine. Your source photo, your target video or webcam feed, and your voice reference clip never get sent anywhere. There's no cloud inference call happening in the background, no "processing on our servers" step.
The only two times NZCAM talks to a server at all:
- One-time activation, when you first install and license the app.
- Automatic updates, checking for and downloading a new version.
Neither of those touches your media. Your photos, video, and audio simply aren't part of what gets transmitted.
Why this is a real architectural choice, not just a privacy statement
Running everything locally is a harder engineering problem than routing frames to a server. It means NZCAM has to ship its own Python runtime, its own GPU acceleration stack (CUDA/TensorRT on NVIDIA, DirectML on AMD/Intel, CoreML on Apple Silicon), and every AI model it uses, all bundled into the install, all running against your own hardware in real time.
The upside of doing it the harder way: nothing about your face swap or voice change depends on a network connection, a server being up, or a company's data-retention policy for "improving the product." Once NZCAM is installed and activated, it runs fully offline.
What we don't do
We don't upload your photos, video, or audio. We don't store them. We don't use them to train models, ours or anyone else's. There's no "opt out of training data" setting, because there's no training pipeline your data could feed into in the first place; it never leaves your device to begin with.
What this doesn't mean
Being private by design isn't the same as being a tool without responsibility attached. NZCAM is built for lawful, consensual, 18+ creative use. See our Responsible Use page for what that means in practice. Local processing protects your data; it doesn't change the rules around whose face or voice you're allowed to use.
Curious about the technical side of how NZCAM stays real-time without a cloud round-trip? See how performance works, or check system requirements for your GPU.