A while ago, I wrote an article about this topic and went over generating music using MusicLM. Today, and just a over a month after publishing that post, the folks at Stability AI released Stable Audio, and I think it's going to blow your mind! 🤯
At this pace, we're going to get to personalized exclusive content before we know it! (Interesting thoughts to read later)
- Music Generation: Using Stable Audio, you can generate music that is unique and custom to your needs. It doesn't matter the genre you're looking for whether it's trance, hip-hop or jazz. Just describe the mood, style, and elements you want, and it'll work its magic.
- High-Quality Audio: I've covered MusicLM before, but this model here is the first that can output 44.1 kHz stereo audio. Meaning professionals can work with the generated music files.
- Commercial Usage: With the paid plans, you'll be able to use the generated music files in your commercial projects.
Model & Dataset
To train the Stable Audio model, the engineers at Stability AI utilized a dataset of more than 800,000 audio files which included music, effects, and single-instrument stems. The total dataset consisted of over 19,500 hours of audio.
If you're interesting in the technical details you can check them out on the official site.
So there are three tiers available: Free, Professional, and Enterprise.
The Free tier allows for monthly track generations of up to 20 max, with a track duration of up to 45 seconds, and is for non-commercial use.
The Professional tier is priced at $11.99 per month (excluding VAT/sales taxes) and offers monthly track generations of 500 max, with a track duration of up to 90 seconds. You can use the audio in your commercial work.
The Enterprise tier is custom to your needs. So you'll need to get in touch with them directly.
- Free tier:
- Professional tier:
$11.99/month (excluding VAT/sales taxes)
- Enterprise tier:
Get in touch for custom pricing
Testing it Out
For our example today, I am going to use the same query that we've used to generate the audio using MusicLM:
Electro Disco Instrumental With Keyboard
Then I'm going to download the generated music and compare the outputs from MusicLM and Stable Audio.
Let's Get Started
Head over to Stable Audio's official site (right click, open in new tab) and sign up for a free account. As I've mentioned before, you can generate 20 tracks limited to 45 seconds each for free.
After creating your account, you'll simple need to follow the steps below to convert your text to music or audio:
You can use the little icon on the right side of the player to download and save a copy of your generated output. (I recommend you save every generated audio since each re-run will deduct from your 20 songs limit)
Here's the generated audio using MusicLM (from the previous post):
Here's the generated audio using Stable Audio:
Nice! I can vibe to that. I find it funny now that just a month ago I thought MusicLM's output was impressive. I take that back, Stable Audio really brings AI-powered text to music to the next level!
As an ex-wannabe music producer, I've composed a few beats a decade ago using Fruity Loops and some other tools and VSTs and I really understand the effort this takes for music producers. It's not an easy job to work with each instrument, and to arrange everything together in harmony so that it all comes together nicely.
But this right here is a true game-changer, and it will revolutionize music production. While it don't believe it will replace producers anytime soon, it'll act more like a trusty co-pilot if integrated properly into software like FL Studio and others.
If you've read my previous posts, you know that I am a firm believe that AI is the future of every industry, even I am shocked how fast things are moving. The sooner we embrace these technologies, models, services, and tools the faster we enhance our creative processes and productivity.
Subscribe to this blog and follow me on Twitter for more exciting updates, it would mean a lot to me.
Thanks for reading!