piguet / deeptelegram

This telegram bot uses motion supervised co-part segmentation to produce deepfake video notes directly in telegram.

deeptelegram

This telegram bot uses motion supervised co-part segmentation to produce deepfake video notes directly in telegram.

Based on this repository: https://github.com/albertoxamin/telegram-deep-fakes-bot that utilizes first-order motion model.

This version is instead based on https://github.com/AliaksandrSiarohin/motion-cosegmentation

How to use it

Open the python notebook with colab and follow the instructions on the notebook

Enabled users

You can change the enabled users by simply adding your Telegram ID(not username or name, but ID) to the enabled array

For example:

enabled =[12345678,12344321]

here we have enabled two users, 12345678 and 12344321

Input type

You should send a square picture with a face (you can search directly in telegram and resize the picture with the in-app editor), and a circular video-note or a square video

Output type

Output can be in circular videonote format or square format. Toggle with /output

Facial parts used

You can use only face or face with some hair. Toggle with /hair