Google MusicLM: How to Make Generative AI Music

Along with the new update to Google Bard and the upcoming Generative AI experience in Search, Google also announced MusicLM, an AI music generation model at Google I/O 2023. Luckily, we already have access to MusicLM, and in this article, we have tested and explained its potential and capabilities. Google says it’s an experimental AI tool that can create music based on your text input, but it’s only limited to 20 seconds. Nevertheless, to learn more about Google MusicLM, head to our explainer below.

What is Google’s MusicLM AI Tool?

MusicLM is a text-to-music generation model developed by Google as part of its AI Test Kitchen program. It creates high-fidelity music from simple text descriptions similar to a natural-language prompt. MusicLM generates music at 24kHz with a higher sampling rate, which means the generated audio is high quality. Also, the AI-powered music generation by MusicLM is very fast, almost instant.

The Google MusicLM AI model has been trained on a dataset of 5,500 uniquely created music by sound artists working at Google Arts and Culture Lab. The music is further captioned with rich text descriptions by human experts in the music field. Google has publicly released the dataset — called MusicCaps to support future research.

You can basically create your own music with simple text prompts. If you have got an idea about a particular type of music, but are unable to put your finger on it, MusicLM can drive your imagination and deliver what you want. Currently, MusicLM generates two tracks for your input for up to 20 seconds. But according to its research paper, it’s capable of creating coherent music for up to 5 minutes.

How to Use Google MusicLM to Generate AI Music

With that said, let’s go ahead and learn how to use Google MusicLM to generate AI soundtracks.

1. Head over to the MusicLM website by Google and click on “Get Started.”

2. Now, click on “Register your interest” and join the waitlist. Users are being approved instantly so no need to wait further.

3. Next, sign in with the Google account, and you will have access to MusicLM right away.

4. Now, go ahead and enter text descriptions to create the music of your music. For example, I started with the below prompt and I got a nice relaxing music.

Soothing instrumental music to help me focus

5. Next, I entered the below prompt to generate music for the workout. It did a decent job at that.

music to workout and get maximum energy

6. You can also enter longer prompts with more descriptions. And the output was quite good.

The main soundtrack of an arcade game. It is fast-paced and upbeat, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds, like cymbal crashes or drum rolls.

7. Finally, you can enter abstract prompts as well. This description is from Salvador Dalí’s The Persistence of Memory painting.

His melting-clock imagery mocks the rigidity of chronometric time. The watches themselves look like soft cheese—indeed, by Dali's own account they were inspired by hallucinations after eating Camembert cheese. In the center of the picture, under one of the watches, is a distorted human face in profile. The ants on the plate represent decay.

8. And for those wondering how we got the samples embedded above, you can download the AI music with ease. Click the “three dots” icon next to the generated tracks and select the “Download” option from the pop-up menu.

Google MusicLM Hands-on: First Impressions

While exploring Google MusicLM, it became clear that MusicLM is still an early demo of what AI can do in the music industry. Sometimes, it surprised me with its melodious output, and often, the generated music turned out to be decent to average. In its official video, Google demonstrated that MusicLM can create good music in different variations modeled on popular songs. However, in our testing, MusicLM completely failed to output similar results.

It may be because Google has not fully released the model as it’s still part of the experimental AI Test Kitchen initiative. In the future, you may get an option to upload your own humming audio and ask MusicLM to add a saxophone and render the music in jazz. I also asked it to create a fusion of Western rock and Indian classical music and the output was bad.

To sum up, for an early demo, MusicLM is indeed an exciting AI project for musicians, but to make it truly outstanding, it will require a larger quality dataset, rich text descriptions, and added capabilities like the option to upload your own music samples. We await the release of new models and capabilities from both Google and OpenAI for AI music generation.