README.MD

Deepfakebot

A Discord bot that invokes A. Siraohin's First Order Motion Model for Image Animation with user-supplied input.

Quick disclaimer

I'm not a Python expert (I've learned most of this while coding the bot itself), this was made mostly for personal use but I'm sharing this because it might be useful for other people. If you have any questions or issues, contact goulart#3978 at Discord.

Also, if you host this on your own computer, be aware that the deepfake creation process (independent from the bot) is resource-intensive and will probably slow down your computer (or worse) while it is being done.

Requirements

This was made using Python 3.6.9.

This bot does not include the animation program itself, so you'll need to get it up and running — if you're able to run demo.py and crop-video.py, you're good to go.

You'll also need the face-alignment library, and the pre-trained checkpoint vox-cpk.pth.tar, which should be saved on the root of the animation program folder (both are also available on the link above).

Afterwards, just run pip install -r requirements.txt to install the requirements for the bot itself.

If the crop-video.py script of the animation program throws an error on some videos where it's expected to work, run pip install --upgrade imageio and pip install imageio-ffmpeg to fix that. (Replace pip for pip3 on the commands if needed)

A NVIDIA GPU is not required, but massively recommended; it's possible to run the animation program on the CPU, however that is massively slow and not recommended at all.

Deployment instructions

• Create a bot account on Discord's Developer Portal;
• Get the animation program up and running;
• On deepfakebot.py, edit path_to_exe = 'path/to/first-order-model-master/' (line 20) to the directory where the animation program is located in your system;
• Edit the last line (bot.run('token')), replacing token with your bot's token;
• Create a data folder on the folder where the bot is located, and a temp folder inside of the data folder;
  • Alternatively, edit the path_to_temp variable (line 21) for the location where you want the bot to store its temporary files;
• If needed, replace all python3 references in the code to your Python 3 executable location;
• Last but not least, run the bot with the python deepfakebot.py command.

Running on CPU (NOT RECOMMENDED!)

If you want to run the animation program on your CPU rather than on the GPU, edit both create_subprocess_exec calls to add the --cpu parameter to the scripts being called.

Usage instructions

• >deepfake link_to_image* link_to_video [--silent] [--dont_crop_image] [--dont_crop_video] [--smart_crop] [--find_best_frame] [--absolute]
  • *link_to_image can be a URL or an attachment;
  • Parameters enclosed in brackets are optional;
  • --silent hides details about the progress of the deepfake creation;
  • --dont_crop_image adds black bars to non-square images instead of cropping them;
  • --dont_crop_video adds black bars to non-square videos instead of cropping them;
  • --smart_crop enables the usage of crop-video.py (instead of just center-cropping it)(ignored if --dont_crop_video is set);
  • --find_best_frame makes the alignment start from the frame that is the most aligned with the source image;
  • --absolute aligns the face using absolute coordinates, instead of relative (it is useless to use --find_best_frame together with this option).
  • --smart_crop and --find_best_frame tend to increase the execution time a lot; use with caution.
• >deepfake (no arguments): Shows usage instructions for the bot;
• >help: Shows discord.py's default help message;
• >adv_help: Shows advanced instructions to help users yield better results.

TO DO

• Import the animation program as a submodule, instead of using create_subprocess_exec;
• Make the bot edit the message to show the current percentage / remaining time for the deepfake creation process;
• Same as above, but for the crop-video.py script;
• Cleaning, optimization, etc etc.