Multichannel Looper/Feedback System for Riffusion (with Automatic1111) made for live performance
- Record or load audio
- Resize loaded files (great for drum loops) or randomly crop loaded files
- Any # of prompts, any # of channels
- Max MSP looper patch included for playing back generated samples
dj7.mp4
riffusionMixer.mp4
dj3.mp4
- Install the Riffusion Github if you haven't: https://github.com/riffusion/riffusion
pip install scipy PIL sounddevice- (IF NOT USING GOOGLE COLAB) Install Automatic1111: https://github.com/AUTOMATIC1111/stable-diffusion-webui
Some examples:
python dj.py -p "violin" "opera singer" "funky" in a loop, record the user, uses prompts in order, saves to 3 channels (1.wav through 3.wav)
python dj.py -p "violin solo" -c 5 in a loop, record the user, uses "violin solo" as prompt, saves to 5 channels (1.wav through 5.wav)
python dj.py -p "EDM techno house" -S 0.4 -c 2 -n 2 -l ~/Desktop/break.wav --resize uses loaded file, uses "EDM techno house" as prompt, sets denoising strength to 0.4, saves to 2 channels (1.wav through 2.wav), quits after 2 iterations, resize the spectrogram instead of cropping.
python dj.py -p "opera singer" "orchestra" -c 4 -l ~/Desktop/raga.wav in a loop, uses loaded file, uses prompts in order, saves to 4 channels (1.wav through 4.wav), randomly crops input spectrogram each iteration
All options:
-h, --help show this help message and exit
-p PROMPTS [PROMPTS ...], --prompts PROMPTS [PROMPTS ...]
specify prompts to use in quotes seperated by spaces
-s, --shuffle shuffle prompt order
-r, --reverse reverse samples pre-riffusion
-C, --clear delete previously generated files
-S STRENGTH, --strength STRENGTH
denoising strength (default: 0.5)
-cfg CFG, --cfg CFG CFG scale (default: 7.5)
-st STEPS, --steps STEPS
steps count (default: 50)
-H HOST, --host HOST specify Automatic1111 web host (default:
http://127.0.0.1:7860)
-a AUTH, --auth AUTH specify "username:password" if using basic http
authentication
-l LOAD, --load LOAD load an audio file instead of recording (specify path
in quotes if there's spaces in it)
-rs, --resize resize spectrogram instead of randomly cropping it
(only applies to loaded files)
-rl, --reload use previously generated spectrogram (ignores loaded
file/record mode)
-c CHANNELS, --channels CHANNELS
number of wavs to generate, ex: 3 would cycle from
1.wav to 3.wav (default: length of prompt list)
-i INDEX, --index INDEX
set starting index, ex: -i 3 -c 2 would cycle from
3.wav to 4.wav (default: 1)
-t THREAD, --thread THREAD
thread # identifier, meant for running multiple
instances of this script (default: 0)
-n NUM, --num NUM number of iterations before quitting (default:
infinite)