avatars4all
Live real-time avatars from your webcam in the browser. No dedicated hardware or software installation needed. A pure Google Colab wrapper for live First-order-motion-model, aka Avatarify in the browser. And other Colabs providing an accessible interface for using FOMM, Wav2Lip and Liquid-warping-GAN with your own media and a rich GUI.
Based on the works:
- First Order Motion Model for Image Animation, https://aliaksandrsiarohin.github.io/first-order-model-website
- Avatarify, https://github.com/alievk/avatarify
- Webcam for Google Colab over Websocket, https://github.com/a2kiti/webCamGoogleColab
- Wav2Lip, Accurately Lip-sync Videos to Any Speech, http://bhaasha.iiit.ac.in/lipsync
- Liquid Warping GAN (Impersonator), https://svip-lab.github.io/project/impersonator
- Liquid Warping GAN (Impersonator++), https://www.impersonator.org/work/impersonator-plus-plus.html
- pyAudioAnalysis, https://github.com/tyiannak/pyAudioAnalysis
- pyannote-audio, https://github.com/pyannote/pyannote-audio
- U^2-Net, https://github.com/NathanUA/U-2-Net
- MODNet, https://github.com/ZHKKKe/MODNet
In this repository you will find:
- Colab for live real-time talking head deep-fakes from your webcam. (j.mp/cam2head)
- Colab for creating talking head deep-fakes (VoxCeleb model) from YouTube or other videos. (j.mp/vid2head)
- Colab for creating full body deep-fakes (Tai chi and fashion models) from YouTube or other videos. (j.mp/vid2body)
- Colab for creating full body deep-fakes (impersonator model) from YouTube or other videos. (j.mp/vid2act)
- Colab for creating full body deep-fakes (impersonator++ model) from YouTube or other videos. (j.mp/vid2warp)
- Colab for creating lip sync deep-fakes based on audio. (j.mp/wav2lip)
- Colab Green screen effect for video with optional background video, and sketch, bokeh and more effects. (j.mp/vid2green))
Features:
- The fastest purely online solution I am aware of for live real-time first-order-motion-model avatars from your webcam.
- A new auto-calibration mode that works in real-time!
- A new exaggeration factor to get those damn muppets to open their mouths!
- Drag and drop local/web images on the GUI to upload new avatars!
- Options to switch between avatars, including newly generated StyleGAN faces, as inspired by Avatarify, of:
- Smart auto-pad/crop/resize to the head or body, for images and for offline videos, tuned for best results.
- Full control of model parameters as well as zoom and buffering options in the GUI.
- Upload your own images and videos or pull them from the web including from YouTube, etc., and optionally trim videos.
- Visualization of facial landmarks and their alignment between source and target.
- Download videos with original audio and framerate, and optimized for compatibility.
- One click operation with Runtime -> Run all.
- Optional Wav2Lip post processing following head animation.
- Combining Wav2Lip with speaker diarization for automatic animated skit creation from audio ("Wav2Skit").
Reference implementations:
- https://colab.research.google.com/github/AliaksandrSiarohin/first-order-model/blob/master/demo.ipynb
- https://colab.research.google.com/github/tg-bomze/Face-Image-Motion-Model/blob/master/Face_Image_Motion_Model_(Photo_2_Video)_Eng.ipynb
- https://colab.research.google.com/github/alievk/avatarify/blob/master/avatarify.ipynb
- https://colab.research.google.com/github/a2kiti/webCamGoogleColab/blob/master/webCamGoogleColab_websocketVersion.ipynb
- https://colab.research.google.com/github/thefonseca/colabrtc/blob/master/examples/colabrtc.ipynb
- https://github.com/l4rz/first-order-model/tree/master/webrtc
- https://gist.github.com/myagues/aac0c597f8ad0fa7ebe7d017b0c5603b
- https://colab.research.google.com/drive/1tZpDWXz49W6wDcTprANRGLo2D_EbD5J8
- https://colab.research.google.com/github/svip-lab/impersonator/blob/master/impersonator.ipynb
- https://colab.research.google.com/drive/1bwUnj-9NnJA2EMr7eWO4I45UuBtKudg_
- https://terryky.github.io/tfjs_webgl_app/face_landmark
- https://eyaler.github.io/tfjs_webgl_app/face_landmark
Workshops, tutorials and talks