Battle of the Bands III: Riffusion vs. Suno vs. Udio

AI music models can create entire songs now, but which one does it best?

Housekeeping: I’m thinking of launching monthly video Q&A calls on AI topics, where you submit questions in advance or ask them live.

Paid subscribers would be the first to get their questions answered and also receive the session recording + transcript afterward.

Would that be useful to you? Let me know by taking this super quick survey:

---

Man, it’s been a hot minute since the second Battle of the Bands. Almost exactly one year, if you’re the kind of person who knows how to read a calendar.

Back then, AI music still sounded like someone threw random MIDI effects into a blender.

But things have changed.

Udio came out, Suno kept improving, and Riffusion re-entered the game with its brand-new FUZZ model.

I already had a Sunday Showdown between earlier versions of Suno and Udio, but it’s clearly time for a rematch.

So let’s prompt us some AI songs and see what happens!

The contestants

Ladies and gentlemen…

For the third round of our music battle, I bring you three song-writing hopefuls:

• Riffusion, reinventing itself for the third time. It started with a unique “spectrogram” approach, moved on to 12-second jingles, and finally launched the FUZZ model that writes entire songs in one sitting. But is it enough to unseat the two veterans of the AI music game?

• Suno, the original AI song writer. From its humble Discord beginnings as Suno Chirp Bot to its Microsoft Copilot integration to its impressive standalone career, can this old-timer prove it still has what it takes?

• Udio, rising to prominence in spring 2024 and taking the AI music world by storm. Can it maintain this positive momentum and come out on top?

But enough about our contestants, let’s look at the competition itself!

The contest

This time around, our challengers will face off in the following four categories:

• Creative interpretation: Can the participants handle a vague prompt that doesn’t specify the genre, style, or instruments?

• Instrumentals: How good are our contestants at creating tracks that closely follow instructions on genre and instruments?

• Songs (from scratch): Who can write the best song given a topic and musical direction?

• Song (from lyrics): Which model is best at turning the same starting lyrics into a full song?

By default, all three models generate two songs for each prompt. I’ll be picking my favorites to compare.

Let’s roll!

The results

We’ve got 12 songs to get through, so without further ado…

1. “Morning in the meadow” (creative interpretation)

Let’s kick things off with something calm and soothing.

I gave each model the following short prompt:

Morning in the meadow

I was curious how they’d try to interpret the setting and match the intended mood.

Here’s what they came up with.

Riffusion

Riffusion went for an acoustic, soft piano track that wouldn’t be out of place in a meadow during morning hours.

It’s quite lovely, isn’t it? If soft piano meadow music is your thing, that is.

Suno

Suno had roughly the same idea but chose to accompany the otherwise lonely piano with some strings.

I’ll take it!

Udio

Udio gave us an atmospheric folk track with a woman vocalizing and what appear to be high-pitched harp sounds.

It’s definitely the most unconventional track of the three. But is that a bad thing? You be the judge.

Verdict

It’s hard to pick a winner here. I like Riffusion’s and Suno’s tracks more, but there’s nothing strictly wrong with Udio’s take at all.

Winner: Tie

2. “Dystopian harmonica” (instrumentals)

For this challenge, I wanted to see how each model would incorporate specific instruments into a track. Here’s the prompt:

Dystopian cyberpunk soundtrack played on harmonica and bongos.

Let’s hear ‘em!

Riffusion

Riffusion went for an industrial electronic track reminiscent of The Crystal Method:

I dig it.

Riffusion claims to have picked tribal bongos, but I feel the percussion sounds are a bit too deep for them. The harmonica reprise is quite clear, though, and the entire thing works as a dystopian soundtrack as far as I’m concerned.

Suno

Suno’s track gives off a similar vibe but is a bit all over the place, with sporadic dubstep-like moments:

Notably, I don’t pick up any harmonica or bongos, either. But it certainly passes the “dystopian cyberpunk” vibe check, especially during the slower sections like the intro.

Udio

Udio went its own way again:

No bongos. No harmonica. We do have a dystopian-ish track with a synthesizer lead, so that’s something. Still, I didn’t give the model such impossible constraints as to justify it ignoring both of the requested instruments.

It also doesn’t help Udio’s case that the second track it generated was this. I’ll never know how it went from “dystopian cyberpunk” to a mediocre Buena Vista Social Club knockoff, but here we are.

Verdict

Riffusion takes this one, both for the consistent track and better prompt adherence.

Winner: Riffusion

3. “Alt-rock missing sock” (song from scratch)

I thought it’d be funny to have an over-the-top angry track about a common frustration we can all relate to, so I asked for:

Angry alt-rock song about a sock lost in the laundry, including vocal screams, distortion effects, and banjos.

How’d our models do?

Riffusion

Honestly? Riffusion just about nailed it. I was picturing Linkin Park in my head when writing the prompt, and, well, check this out (especially the transition from pre-chorus to chorus):

Alt-rock vibes? Check.

Angry screams? Check.

Distortion? Check.

Banjo? Well…uh…Riffusion keeps referring to the banjo in its text descriptions, but I honestly couldn’t hear it. Could you?

Suno

Folks, we’ve got ourselves some banjos!

This is the most prompt-adherent result of the three and a cool angsty track in its own right. Well done, Suno.

Udio

Is it just me, or does it feel like Udio somehow hears the other two tracks and always picks a completely different direction just to be contrarian?

I mean, it’s actually a pretty cool experimental punk take with a female lead and some clever turns of phrase. But why did Udio decide to scream the literal word “scream” instead of just, you know, screaming?

Also, where are those banjos, Udio? WHERE ARE THE BANJOS?!

Verdict

While I prefer the Riffusion track, Suno should take this one if we go by prompt adherence.

Winner: Suno (or Riffusion? Help!)

4. “Bubblegum zombies” (song from lyrics)

For this one, I partnered with ChatGPT to brainstorm ideas that juxtaposed innocent, naive-sounding songs with dark subject matter.

We settled on a bubblegum pop about the zombie apocalypse.

ChatGPT suggested the following prompt:

Bubblegum pop song featuring bouncy synths, handclaps, funky bass, retro arcade sound effects, and ultra-cute vocals.

It then provided the following lyrics for each model to use:

[Verse 1]
Woke up late, what a perfect day
Sun is shining, time to play
Oh look, my friends are running fast
Guess I better catch up—wait, why so mad?

[Pre-Chorus]
Knock, knock, who’s at my door?
Why’s your skin all pale and torn?
Arms stretched wide, you look so sweet
Is this a game? Are you playing with me?

[Chorus]
Run, run, hug! It’s a zombie chase
They just want love, don’t be afraid
Hands up high, spin around
Oops, another city down!

Run, run, hug! Don’t you slow
They just wanna say hello
One by one, we’ll all be friends
This party never ends!

[Verse 2]
Downtown streets are kinda weird
All my neighbors disappeared
Oh well, more room to twirl and dance
Broken windows? Let’s take a chance!

[Bridge]
Look, they’re everywhere
Hands in the air
I hear them sing
Just like me
We’re all one team
No more screams
Just one big dream
Come with me

[Chorus]
Run, run, hug! It’s a zombie chase
They just want love, don’t be afraid
Hands up high, spin around
Oops, another city down!

Run, run, hug! Don’t you slow
They just wanna say hello
One by one, we’ll all be friends
This party never ends!

[Outro]
No more running, time to stay
Everyone’s here, we’re all okay
Take my hand, close your eyes
Welcome to the other side

Let’s throw these at our contestants and see what they come up with!

Riffusion

Seriously, Riffusion just keeps on nailing it:

Go ahead and tell me that’s not catchy as Hell, I dare you!

“Oops, another city dooooown!”

Uh…where was I?

Oh, yeah. This track is great. I love what Riffusion did with the prompt and the lyrics.

Suno

Yeah! That works.

It’s the perfect juxtaposition of an upbeat and happy track with the concept of zombies chewing your face off. One might even argue that it’s bubblier and gummier than Riffusion’s bubblegum pop.

Still, something is preventing this track from feeling as viscerally catchy as the one from Riffusion. I can’t quite put my finger on it. Is it too generic? Too unvarying?

Tell me what you think.

Udio

Yeah, no.

Why is Udio so hellbent on being different and disregarding directions?

Are you going through a rebellious teen phase, Udio? Is that what this is?

It took me a few tries just to stop Udio from coming up with its own lyrics¹. And even then it spat out two tracks that didn’t have all that much to do with my instructions.

(Here’s the discarded track, in case you think I cherrypicked the crappy one.)

Verdict

Suno did good, but it’s hard to compete with Riffusion’s sheer catchiness.

“Oops, another city doooooooown!”

See? Now you’re singing it again!

It’s quite an earworm.

Winner: Riffusion

Final thoughts

After listening to several dozen tracks, here’s what I think.

🥇Riffusion

If you ask me, Riffusion is the overall winner of today’s contest.

It didn’t make any egregious errors, mostly stuck to instructions, and consistently managed to read my mind by coming up with songs that matched my implicit expectations.

This took me a bit by surprise if I’m honest.

I expected Suno and Udio to have the more polished products.

Well played, Riffusion. Well played.

”Oops, another city doooooown!”

Damnit!

🥈Suno

Suno is a very close second.

You might even think it deserves to win. I don’t know your life. I’m not you. (Don’t believe the rumors.)

Suno produces great-sounding, well-realized songs and might be the best at following instructions.

This old dog’s still got it!

🥉Udio

Udio is the undisputed disappointment of today.

It seems needlessly difficult to steer Udio in the right direction.

The results themselves are hit-and-miss, with some nice-sounding tracks, some mediocre ones, and a few WTF ones, too.

I definitely thought Udio would do much better than this.

Oh well, there’s always Battle of the Bands IV, right, Udio?

🫵 Over to you…

Do you agree with my judgment here? Am I being unreasonably critical of Udio? Maybe Udio needs a different prompting approach?

If you’ve worked with Udio and know how to consistently get solid results, I’d love to hear from you.

Leave a comment or drop me a line at whytryai@substack.com.

1. Hint: You want to select Advanced Controls, then Manual Mode, then slide the Lyrics Strength slider to 100% to force Udio to use the lyrics “as is.”