I made my first DeepFaceLab deepfake on Google Colab.
Most of the material about using DeepFaceLab on Colab assumed the user was already familiar with DeepFaceLab's workflow.
However, if you have a Mac or other computer without an Nvidia GPU, you might need to use colab to get started with DeepFaceLab. So I added explanation on how to use DeepFaceLab on Colab for newbies.
Fork from DFL-Colab, a DeepFaceLab fork for Google Colab made by @chervonij.
I made a (very undertrained) deepfake using DeepFaceLabs.
This is my first deepfake using an autoencoder model, the kind of model that DeepFaceLab uses. Other deepfakes I've made with Avatarify used First Order Motion Model.
I trained this model on a small amount of data and iterations: 38 source images for 500 iterations. That's not enough to produce a quality deepfake.
It's common to train for 100k or more iterations. And use hundreds or thousands of source images.
But my goal with this first deepfake was to get through the DeepFaceLab workflow, not make a great deepfake. Hopefully, this repo can help you get through it as well.
The preview image for my model at 400 iterations.
- Download the repo
- Upload DFL_Colab_Newb.ipynb to your Google Drive and open in Google Colab.
I've added extra explanations and code within the Jupyter Notebook file to help newbies get started.
Please consider donating to DeepFaceLab. This project is open source and a lot of hard work went into it.
This project provides you IPython Notebook to use DeepFaceLab with Google Colaboratory.
You can create your deepfake video even if you don’t have a powerful GPU on a desktop PC.
Keep in mind that working with a remote server imposes some restrictions.
Because of this, working with DFL-Colab is not so convenient as working with DeepFaceLab directly on your computer.
DFL-Colab uses DeepFaceLab and to understand the workflow, you should first understand the DeepFaceLab workflow.
Official Links
- DFL-Colab Notebook: Link -- Works with latest DFL version
- DFL-Colab Notebook 1.0: Link -- Works with DFL 1.0 version
- DeepFaceLab Repository: Link
- Youtube channel: Link
Community support