These AI music generators delight - and make you feel like a virtuoso (2023)

Generative AItook the internet by storm with its ability to drop images and blobs of text based on user queries. Despite the widespread hysteria thatartificial intelligence will make most of our jobs irrelevant, it could actually streamline workflows for a wide range of professions – including music production.

We picked a few of our favorite tools for simplifying music production and put them to the test, demonstrating a real-world scenario where this technology could be used for good, not evil. It's time to embrace this new technology as a help, not dismiss it as a doomsday device, and we hope this guide helps you relax.

What about AI music generators?

It's important to note that some of these programs can generate music - but we're still a long way from being able to press a button to produce a one-hit wonder. The AI ​​music generators we tested are legitimate tools, unlike the voice changers you may have seen that can make anyone sound like Drake. (New York TimesrecentlychronicleAI Drake's hit, which briefly went viral before it was pulled from the streaming service, and spoiler alert, wasit ishappy.)

Mess withRefusal, it didn't take long for me to realize that no music is necessarily ready for Spotify right out of the box - I guess having a modicum of musical talent also helps. But this basic idea made me realize that these programs are useful for generating ideas in songwriting, not necessarily "cheating" and doing the whole job as a musician. Most of these tools are much easier to useChatGPT,In the middle of the trip, and some other models we have already written about.

✅How do AI music generators work?Similar to AI image generators and text AI tools, AI music generatorsthey rely on deep learning-a method of machine learning in which computers process data in a way that mimics the human mind - to produce results. More precisely, the software must be fed in huge quantitiestraining data(here, music samples) and process it. But if, for example, you only train the system with the greatest hits of the Beatles, any tune the AI ​​music generator produces after that training period will sound an awful lot like ... the Beatles. Therefore, a huge library of training data from various sources is essential for this process.

Refusal

This super lightweight web app was one of the first AI music models I started playing with. Using Stable Diffusion—yes, the same platform that creates images—Riffusion can create an image that is then synthesized to create "music." I use the quotes there sparingly because a lot of the ideas he throws out are not only low fidelity, but pretty basic. However, it can definitely give a musician a cool riff or beat that they could use in their work.

The white panel you see on the left of the screen is an image synthesized by Riffusion. The application analyzes peaks and valleys and proceeds to translate the physical form into music in the same way that a record player interprets sound from a ridge tovinyl. Along with the instructions (see the image above for an example), I also have control over melodic patterns, creativity, tempo and that's it.

Nothing special, but more than enough to get the creative juices flowing.

ChatGPT

Despite some of the dumber things I've askedChatGPTto do, it didn't take me long to realize that it could actually be another incredibly useful tool for musicians looking to find samples – and even lyrics. Of course, most of his training data is from 2021 and earlier, but that's not a big problem because there are tons of great tunes to choose from even before the 21st century. As you can see below, I searched for five samples with similar vibes to Tijuana Brass.

In addition to finding patterns, it is also a good tool for writing texts. They're not perfect, but with a little human intervention, the lyrics below are a good springboard to write a country song aboutFormula 1.

Samplette.io

Although ChatGPT is a good starting point for finding new music,Example IOgiving you more control over the results. Instead of aimlessly searching YouTube, this app gives you eight filters to adjust your search results—genre, style, country, key, tempo, voice, views, and year—and findIn factwhat do you want. It also includes a keyword section to further narrow your results.

Samplette is also a super useful tool for finding new music. I also never had to skip ads while listening to music – make of that information what you will. Don't tell Spotify, but I think Samplette's algorithm is much better at finding fresh music. I tend to leave it haphazard while I work.

Isn't this a copyright infringement?

AIanythingbecomes significantly more complex when it comes to intellectual property, especially when it comes to music. (This article fromEdges provides deeper context to the seemingly endless gray areas of music copyright and how we should proceed in an era where the generation of AI-assisted music is likely to remain.)

Generative AI models – including music generators – must be trained on existing material. This is often associated with content theft, but there really is no other way for these programs to generate music. It's no different than how people couldn't make music without hearing it first. You're not necessarily stealing it, but instead riffing from other works.

This can be compared to the way samples - notes and melodies taken from other songs to embellish someone else's work - are played in the music industry. Copyright law regarding samples sounds incredibly simple, considering the artistsrequires express permissionuse someone else's work.

But this is where things actually get incredibly complicated. Can you copyright a chord progression? How similar does a song have to be to raise a red flag? Who decides how similar two works areactuallyis?Earlier this year there was, for example, Ed Sheerancleared of all wrongdoingfollowing claims that his original song, "Thinking Out Loud", strayed too far from Marvin Gaye's "Let's Get It On".

In the case of AI music, the situation is a work in progress. Additionally, the use of copyrighted music to train these models is currently the biggest problem for most musicians.

Other applications

While AI is mostly a gimmick right now, we believe it could soon revolutionize the production process for small film producers—among many other uses. These collectives often have very little money available to create music to complement their films.

In my few weeks with these programs, it's clear that AI music creation has a long way to go before Hollywood music composers start worrying about the security of their jobs. AI tools can give you a pretty good product, but most programs I've used have given me limited control over the finished product; you can't change individual notes and chord progressions, but you can adjust mood, dynamics and lyrics.

It's moving forward

Like the other AI tools we've written about, music generators are nothing to panic about. They're good at messing around and can get you great results, but they still can't compete with the creativity and intricacy of a human composing a song or playing an instrument. I wonder if artists and composers can start using these tools for inspiration in the not too distant future. Let us know your thoughts below.