Lucid Sonic Dreams syncs GAN-generated visuals to music!
By default, it uses NVLabs StyleGAN2-ada-pytorch. It can also use pre-trained models lifted from Justin Pinkney's consolidated repository, but these are untested.
"save_frames" is no longer used as this version works in RAM and doesn't save frames to disk.
Sample output can be found on YouTube.
This implementation has been tested on Python 3.9. It now uses the PyTorch implementation of StyleGAN2-ada, and works with Ampere cards.
To install: git clone this repo and change directory into your newly created directory:
git clone https://github.com/nerdyrodent/lucid-sonic-dreams.git
cd lucid-sonic-dreams
It is suggested that Anaconda or Miniconda be used to create a new, virtual Python environment with a name of your choice. For example:
conda create --name sonicstylegan-pt python=3.9
conda activate sonicstylegan-pt
Install the packages required for both stylegan2-ada-pytorch and this repo:
pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install click requests ninja imageio imageio-ffmpeg tqdm psutil scipy pyspng
(Optional) If you already have stylegan2-ada-pytorch (recommended for training your own networks), create a symbolic link to it:
ln -s ../stylegan2-ada-pytorch stylegan2
Refer to the Lucid Sonic Dreams Tutorial Notebook for full parameter descriptions and sample code templates. A basic visualization snippet is also found below.
from lucidsonicdreams import LucidSonicDream
L = LucidSonicDream(song = 'song.mp3',
style = 'abstract photos')
L.hallucinate(file_name = 'song.mp4')
sg2-ada-pt-song-spleeter.py is an example with a variety of configuration options as defaults, based on the audio being split into 4 stems using spleeter - https://github.com/deezer/spleeter