This is the source code for two projects I made with Epstein files. The files are heavily redacted and the diverse blobs of redactions can be used to make a lot of things. I hope this repo raises awareness of the opression and tyranny of the ruling class via internet culture.
Uses KNN to match feature vectors of the each image (extracted from pages of pdf) and maps them accordingly to video input to mimic the outlines of the input video.
Example Usage :
Doom but the video output is Epstein files
Bad Apple on Real Epstein files
Warning
Tested on linux w/ x11 only.
- Install the dependencies from requirements.txt
pip install -r requirements.txt- Setup the folder structure in following fashion:
❯ tree --gitignore
.
├── doom.py
├── input_pdfs
│ ├── EFTA00005705.pdf
│ ├── EFTA00005707.pdf
│ ├── EFTA00005711.pdf
│ └── EFTA0000XXXX.pdf
├── main.py
└── requirements.txt
- Download Bad Apple and save it as
bad_apple.mp4(use yt-dlp or something) - run main.py :
python main.py - finally use ffmpeg to stich the audio to it:
ffmpeg -i redacted_apple.mp4 -i bad_apple.mp4 -map 0:v -map 1:a -c:v copy -shortest final_with_audio.mp4It took me ~45min to render on my Ryzen 5 5500U, quite an CPU intensive work. I am sure it can be optimized but I haven't looked into it.
Warning
Doesn't work with wayland.
- Download Doom and DOSBOX
- Change config to make sure input capture is correct
- run doom.py :
python doom.py
Licensed under MIT. Check out my website : theindiandev.in My Discord server : discord.gg/qEYbuWu5NE