Skip to content

Noorquacker/deepfake-webcam

 
 

Repository files navigation

Real-Time Deepfake Webcam

Based on the First Order Motion Model for Image Animation, this code is forked from their wonderful code repository.

Prerequisites

In order to run in real-time, you need a GTX 1060 or above in order to go above 20 FPS and look somewhat decent. This repo was tested on a GTX 1060 6gb and it worked at 20-25 FPS.

You also must be running on Linux if you want to use a live webcam. We do not have a cross-platform way of having virtual webcams, but you can help.

Installation

  1. As root (which is not advised, but do what you want), do what you do on every other Python repo:
    sudo pip3 install -r requirements.txt
  2. Make a new folder called REPO and put the vox model checkpoints in there (download vox-cpk.pth.tar and vox-adv-cpk.pth.tar and move them in there)
  3. Install and load the v4l2loopback kernel module for your distro
  4. Create a fake webcam device. Instructions can be found in v4l2loopback's documentation
  5. Edit webcam.py to point to the proper input and output webcams
  6. Run webcam.py as root, unless you have permissions set up nicely
  7. Show up to Zoom meetings

About

Uses TensorFlow and first-order-model AI to create a deepfaked virtual webcam

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 90.6%
  • Python 9.4%