Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/hacksider/Deep-Live-Cam/llms.txt

Use this file to discover all available pages before exploring further.

Deep-Live-Cam is an open-source tool that lets you swap any face in real time using a single source image. Built on InsightFace for face detection and ONNX Runtime for inference, it runs on CPU or GPU and works with your webcam feed, video files, and still images. You select one photo of a face, click a button, and the swap appears live — no per-frame manual editing required.

Key features

Deep-Live-Cam version 2.1.5 (GitHub Edition) ships with the following capabilities, all operating in real time:
  • Real-time webcam face swap — swap your face on a live camera feed with sub-second latency. Pipe the output into OBS or any virtual camera for streaming.
  • Image and video processing — apply a face swap to a static image or an entire video file and save the result to disk.
  • Face mapping — assign different source faces to different subjects in the same frame simultaneously, useful for multi-person content.
  • Mouth mask — retain your own mouth and lip movements on top of the swapped face so speech looks natural.
  • Face enhancer — optional post-processing with GFPGANv1.4 (or GPEN-256/GPEN-512) to sharpen and restore swapped faces.
  • Multi-face processing — the --many-faces flag applies the default source face to every detected face in the frame at once.
  • GPU acceleration — supports CUDA (NVIDIA), DirectML (Windows AMD/Intel), CoreML (Apple Silicon and legacy), OpenVINO (Intel), and ROCM (AMD Linux). Falls back to CPU automatically.
  • NSFW filtering — a built-in content filter powered by opennsfw2 blocks processing of nudity, graphic content, and other sensitive material.
  • Headless / CLI mode — passing -s, -t, and -o arguments skips the GUI entirely so Deep-Live-Cam can run in scripts and pipelines.
  • Frame interpolation — temporal smoothing between frames reduces flickering during webcam or video processing.
  • Live mirror and resizable preview — flip the webcam feed horizontally or resize the preview window to fit your workflow.

Use cases

Live streaming

Stream with a swapped face over OBS or any virtual camera. Select a source image, click Live, and capture the preview window.

Watch movies with any face

Point the webcam at your screen while a movie plays to replace every character’s face in real time.

Memes and short clips

Process a video clip or still image offline. Output is saved automatically next to the target file.

Live shows and performances

Use Deep-Live-Cam as a live visual effect during performances, presentations, or interactive shows.

Hardware requirements overview

Deep-Live-Cam runs on any machine with Python 3.11 and ffmpeg. CPU-only mode works out of the box but is slower. For smooth real-time performance a discrete GPU is strongly recommended.
Execution providerHardwareNotes
cpuAny x86-64 or ARM CPUDefault fallback. Slower but always available.
cudaNVIDIA GPURequires CUDA Toolkit 12.8 and cuDNN v8.9.7. Best performance on most systems.
directmlAMD or Intel GPU on WindowsNo separate driver install beyond onnxruntime-directml.
coremlApple M1/M2/M3 (Silicon)Requires Python 3.11 and onnxruntime-silicon.
coremlApple legacy (Intel Mac)Uses onnxruntime-coreml.
openvinoIntel CPU/GPUUses onnxruntime-openvino.
The insightface models used by Deep-Live-Cam are licensed for non-commercial research purposes only. If you use a real person’s face, you must obtain their consent and clearly label any output as a deepfake when sharing online.

Next steps

Installation and requirements

Step-by-step instructions for setting up Python, ffmpeg, models, and GPU acceleration on Windows, Linux, and macOS.

Quickstart

Get to a working face swap in five steps: clone, download models, create a venv, install dependencies, and run.

Webcam mode

How to use the Live button for real-time face swapping and how to connect the output to OBS or a virtual camera.

Image and video mode

How to select a source face and target file, start processing, and find the output on disk.

Build docs developers (and LLMs) love