Artificial intelligence is changing the way people create and listen to music. Anyone can boot up an AI song generator like Suno, Riffusion, VoiceMod and Udio to create original tracks in minutes.
The broader category of text-to-music apps like our MIDI app AudioCipher have aimed to solve sample, chord and melody generation for music producers. They were built as creative inspiration tools that fit into your DAW workflow.
AI text to song apps, on the other hand, seem to be driving toward the complete elimination of human agency, in favor of instant gratification. It's like playing a game with all the cheat codes turned on. That may sound dystopian, but like many AI tools, there could be a silver lining.
Users describe a genre and enter lyrics to hear their vocals come to life over an original instrumental track. People are sharing these creations on social media and experiencing the same kind of feedback and validation loops that other professional musicians have historically worked hard to earn.
On a more positive note, text to song democratizes the music creation experience so that anyone can have a go at it. Music producers can sample their favorite moments and pull them into the DAW, without paying exorbitant licensing fees.
Where genres like hip hop flourished under sample-based production, Sony hit a small rapper TreFuego with an $800,000 penalty and won the case in April 2024, signaling to others that sampling copyright music is totally off limits. The unintended consequence of this behavior is that beatmakers are more likely to turn to sampling AI generated music instead.
Drake released a diss track in April featuring AI clones of Tupac and Snoop Dogg's voices. Tupac's estate fired back with a lawsuit for the unauthorized cover song, but as Billboard magazine pointed out, the consequence of Drake's actions have had a measurable impact on the public's attitude toward this activity.
A new category of "AI music fan fiction" has emerged, where fans use AI drake and AI Kendrick voices to imagine follow up diss battles between the two artists. It's important to note that fake Drake AI voices were already very popular in 2023, but this is the first time we're seeing rapper beefs explored casually at scale.
With every month that passes, the psychology of fans seems to be transforming from passive consumption to active co-creation with artificial intelligence.
Before the end of year, the solo experience of "instant-song creation" will evolve into a more robust peer-to-peer ecosystem of music collaboration apps. For now, here are the four biggest and easiest AI song generators available to the public.
Suno V3: AI Song Generator turns lyrics into vocals
As of April 2024, Suno is one of the two most powerful AI song generators available. Their biggest competitor is Udio, but we'll get to them in a moment.
Suno's v3 model is a major step up from their previous version, with higher audio quality and better songwriting. It's able to niche down into the instrumental style of almost any obscure sub genre and even fuse unrelated genres together.
Their interface is incredibly simple and easy to use, which may explain the app's viral popularity amongst musicians and non-musicians alike. The actual AI text to song mechanism is twofold:
Lyric to vocals: Type in a song concept and use AI to transform it into lyrics, or write the lyrics yourself. An AI singer will perform the vocals over an instrumental track of your style. You can't control the vocalist who sings it or the melody, yet.
Genre description into AI instrumentals: Describe the kind of music you want for the instrumental layer of the song. You won't be allowed to use the names of famous artists, but you can describe genres and moods. It will boil your description down to 3-4 attributes and build music from there.
Once a song's been created, you can extend it forward from any point in the song to continue making more music in that style. If you're a music producer, it's easy to download the track and import it to a DAW for further editing.
Pricing: Suno currently offers a competitive rate of $10/mo to generate 500 songs. You can top off if you run out of credits before the month is over.
Udio Music: AI text to song (Google Deepmind Devs)
Udio is the first serious competitor to Suno and when compared side by side, they're almost identical in their feature set. Both offer lyric-to-voice, genre prompting, instrumental options, and a community space to explore other people's creations.
According to some publications, like the Rolling Stone, Udio's music quality sounds better. In my opinion they're on par with each other. However, having used both apps at length, I found Udio's model took more risks musically and was more likely to veer off in unexpected directions when encouraged to do so.
The company raised a big round of venture capital in late 2023 and moved to hire heavyweight engineers from Google's AI team, Deepmind. They also buy in from record producer Will.I.Am and, to many people's surprise, the conscious rapper Common. It's yet to be seen how this kind of software fits into the music industry, but it's interesting to see major figures from rap as early adopters.
Pricing: Udio launched in April and during this first phase, they're allowing users to create 1200 songs per month, based on 600 prompts. This is a staggeringly large budget for a free plan, but it makes more sense when you consider the 30 second duration of each clip. If you do the math, Suno still gives you more music overall.
Riffusion: Popular at launch but dead on arrival?
Riffusion was the first major text to sample generator, beating companies like Google and Meta to the bunch by close to 6 months. They published their first version as a free web app in December 2022. About 10 months later, they secured $4M in venture capital and released version two as a more mature website.
As the screenshot above shows, they follow the same basic mechanism as Suno and Udio. Users can provide lyrics and describe the sound they want to hear. However, the audio quality is noticeably worse than their competitors and after a warm reception by the public during their initial launch in late 2023, they have gone silent across all social media platforms.
Tracks are only 12 seconds long and lack long short term memory. Even in such a short time period, they lose the plot and veer off into noise. It's unclear whether Riffusion is rebuilding in stealth or has suffered some critical blow that's setting them back. We've contacted their team to learn more and will report back as soon as we have an update.
VoiceMod's Text to Song app for making memes
The Voicemod Text to Song generator, nicknamed the Musical Meme Machine, was a precursor to Riffusion V2, Suno, and Udio. They nailed the lyric-to-vocals component with the purpose of serving their gamer niche on Discord. The biggest limitation of their service is that it doesn't generate original music. It's also worth noting that there hasn't been a major update to the app for close to a year.
You may have noticed the music memes popping off on social networks this year, thanks to the growing interest in artificial intelligence. Bedroom producers started using AI text generators like GPTChat to imitate song lyrics in the style of a specific rapper. They pass those lyrics into a text-to-speech tool like UberDuck and add instrumental beats to parody the artist.
There's a problem though — your average music producer doesn’t know how to make the text sing. It’s tough to get the voice of a robot rapper to sound decent. The vocals need to be treated and chopped up to match the beat. Free autotune plugins like GVST’s Gsnap make the process accessible, but most people don’t want to invest the time.
That’s where Voicemod Text To Song comes in. People want a free tool that lets them participate in the fun without the advanced skillset. It needs to be something easy that lets you type in some lyrics and spit out a video clip for social media. Voicemod’s music generator solves that problem by rendering your lyrics and melodic vocals, in close to real-time.
How to use VoiceMod Text To Song
Voicemod’s musical meme machine is free to use and it takes less than a minute to generate a song. You’ll go through four simple steps, beginning with selecting your song style. You can take a listen to the MIDI tracks in advance by hitting play. Pay attention to the lead melody because that’s where your lyrics will be added.
Once you’ve chosen a song style, it’s time to choose your singer. There are currently seven options to pick from, complete with names and illustrations. Voicemod already controls a powerful AI Voice Changer tool that predates the meme machine. So they’ve been able to leverage that tone transfer technology to make this song generator produce higher quality vocals than anything you would get from UberDuck.
Now that you selected your singer, it’s time to add vocals. Write down your lyrics and see if you can match the rhythm of the initial melody. If that’s too difficult, don’t worry about it. The app is good at holding vowels while the melody changes. Just make sure you write enough content to get through the full ~30 second song.
Feeling spicy? It’s not advised to use explicit language. You’ll get a warning as soon as you start typing in curse words and repeat attempts could get you banned from the service.
If you’re ready to commit, go ahead and generate the song. You’ll have the option to listen back to the music with karaoke style lyrics ala Spotify. Download the song as a music video for your TikTok reels and give your subscribers something to laugh about!
So anyway, that's a wrap. Without a doubt, Suno and Udio are currently ahead in this race. We'll see if Riffusion comes back strong in mid-to-late 2024. If they wait too long, they're likely to lose most of the mindshare they built as new players enter the market.