Music It Up
A few months ago, my brother introduced me to MusicLM. It was a text-to-music service that was able to create interesting music snippets. It’s not surprise with the current pace of development of AI services that only a few months later, we have have MusicGen, a new tool that offers significant improvements and is FREE.
For the time being, you can try it out here.
One of the biggest benefits over previous generations is that MusicGen remains coherent over longer periods of time. If you ask it for a particular style of music, even after two minutes, the style persists. Having tested it a few times, I saw it adapts to some extent the style of the music of a sample, if you choose to provide one. There is still a bit of a creepy, not-quite-right element to the output. Applying the rule of Two Minute Papers videos, “do not look at the current paper, but two papers down”, you can see that this technology is about to have a breakthrough.
Two elements will help push adoption of these breakthroughs:
First, interfaces that will make these technologies easy to use by non-developers, especially to those who are already creating music will allow people to ease in to use.
Second, making the workflow of generating and updating music much easer will increase usage. If I don’t like what the service produced, how do I update it? What about making tweaks? Why can’t I just enter a title of a song I want it to adopt the style from?
Even if there are shortcomings today, these will likely be solved within a year and we will be amazed by some very exciting music mashups.
