I think we’ve all tried a bit to use artificial intelligence to make music. Initially amazed, then slowly you find the limits, and above all the costs.
My personal view is that machine learning can be used to enable and enhance the creative potential of all people, and I’d like it to be like that for everyone.
That said, there are many platforms on Web, even complex ones, that offer the possibility of creating a song from a text prompt. The “trial” generation is free, but if you need more, you have to switch to a payment plan based on the amount of rendering you need.
However, there is also the possibility of generating music with AI on your computer, downloading several different models, and thus avoiding the costs of online platforms.
I would like to talk here about two solutions that work locally, on your PC: Pinokio and Magenta Studio, two completely different approaches to AI-generated music.
Pinokio
Pinokio It is really a possible solution: its scripts take care of downloading everything you need and configuring the working environment without disturbing your file system in any way. At the installation you will be asked to indicate a Pinokio Home, and everything you download will go inside this directory, no mess around the PC.
Installation procedure and setup are very simple and easy, and it’s available for Windows, Mac, and Linux.
The scripts available obviously do not concern only music, but a myriad of applications in all the areas concerned: text, images, videos, and so on and so forth.
Warning: each application requires disk space, and the download is quite heavy. Make sure you have space on the disk where you created your Pinokio Home.
I have installed several libraries on my PC, currently the ones you see in the image below. Well, that’s 140 GB of disk space, and unfortunately appetite comes with eating.
Anyway, interesting. Worth a try.
Magenta studio
Magenta Studio follow a complete different path and is based on recurrent neural networks (RNN). A recurrent neural network has looped, or recurrent, connections which allow the network to hold information across inputs and these connections can be thought of as similar to memory. RNNs are particularly useful for learning sequential data like music. Magenta currently consists of several tools: Continue, Drumify, Generate, Groove and Interpolate.
These tools are available as standalone programs, but version 2 has become a integrated plugin for Ableton Live, with the same functionality as version 1. They use cutting-edge machine learning techniques for music generation, really interesting.
At the Magenta site you can also become familiar with the so-called DDSP-VST.
Okay, talking about Neural Synthesis may seem like science fiction, but it’s actually simpler than it seems. At the end of the day, it’s just a matter of installing a VST3, which is complex.
If you like to experiment, I find very interesting the part dedicated to the creation of your own instruments, where artificial intelligence can be trained with your samples.
If you use Linux or Mac, take a look at the Magenta Midi Interface and say wow.
In short, as they say a lot of stuff to play and many acronyms to learn.