Real-time face swaps from a single photo, ethics filter included

What it does Deep-Live-Cam is a local, real-time face-swapping tool. You feed it one source image of a face, point it at a webcam or video file, and it overlays that face onto the target stream. It can also map different source faces to multiple people in the same frame and optionally preserve the original subject’s mouth movements while swapping everything else.

The interesting bit The project tries to thread a needle between viral accessibility and self-policing: it ships with a built-in guard that attempts to refuse nudity, graphic content, and sensitive material. At the same time, the open-source repository feels like a teaser for a separate, exclusive pre-built release that promises thirty-plus extra features and priority support.

Key highlights

• Real-time processing on webcam or pre-recorded video using a single source image.
• Mouth-mask mode retains original lip movement while replacing the rest of the face.
• Face-mapping can target multiple subjects in the same scene with different sources.
• Runs offline with support for NVIDIA CUDA, Apple CoreML, Intel OpenVINO, DirectML, and CPU fallback.
• A command-line interface exists but is explicitly marked unmaintained.

Caveats

• The README warns that manual installation is “not for beginners” and requires exact dependency versions; the project steers casual users toward a separate, exclusive pre-built release instead.
• Live webcam mode needs a noticeable 10–30 second warmup before the preview appears.
• The built-in ethics checks are mentioned as a safeguard, but the README offers no detail on how they are enforced or how reliable they are.

Verdict Experimenters and streamers curious about real-time generative media will get the most out of it. If you need a polished, beginner-friendly tool or a stable scripting API, the open-source build will likely frustrate you.