libskibidi: Two Minute Papers for Gen Alpha
libskibidi is ed tech for the next generation of ML engineers. It automatically generates minecraft parkour videos with AI narration of the latest arXiv ML papers' abstracts superimposed. More specifically, it:
- Fetches top N latest arXiv papers in a category
- Narrates their abstracts with randomly-chosen OpenAI TTS voices
- Chooses a random segment of a random "free to use" Minecraft parkour video long enough to fit the narration
- Muxes the abstract text onto the video with FFmpeg
- Uses a "novel" forced text alignment "implementation" based on Whisper to identify and highlight each word as it's spoken for maximum brainrot, while retaining obscure spellings and LaTeX present in the original abstract
Demo
demo.mp4
Usage
libskibidi requires Python 3.12 or later:
python3 --version # libskibidi requires >=3.12
The single non-pip-installable prerequisite is ffmpeg:
Clone the repo:
git clone https://github.com/milkey-mouse/libskibidi
cd libskibidi
Create a venv:
python3 -m venv venv
. venv/bin/activate
pip install -r requirements.txt
Delve into the future of education: