I Tested 5 AI Song Makers to Create a Full EP in One Weekend — Here’s What Happened

Last Saturday morning, I set myself a challenge: produce a 5-track EP using nothing but AI song makers. No DAW. No instruments. No prior recordings. Just me, my laptop, and five different AI music platforms.

Each track would use a different tool. I wanted to see which ones actually deliver on the promise of “type a prompt, get a song,” and whether the output was good enough to publish — not just as a demo, but as something I’d genuinely put on a playlist.

Here’s what happened, track by track.

The Rules I Set

Before I started, I gave myself a few constraints to keep this fair:

I’d spend no more than 2 hours per track, including prompt writing, regenerating, and any editing the platform allowed. Every track had to be a different genre so I could test range, not just one sweet spot. I’d use each platform’s free tier or lowest paid plan — nothing behind a premium paywall. And I’d judge each track on three things: did it sound like a real song, would I actually listen to it, and could I use it for something (a video, a podcast, social media)?

Alright. Let’s get into it.

Track 1: Indie Folk Ballad — Suno

The prompt: “A quiet indie folk ballad about leaving a small town. Acoustic guitar, soft female vocals, wistful and bittersweet. Think Phoebe Bridgers meets Gregory Alan Isakov.”

What happened: Suno delivered the first usable result on my second try. The first generation was good but the lyrics felt too generic — lots of “open roads” and “chasing dreams” clichés. So I wrote my own lyrics and fed them back in with the same style description.

The second generation was genuinely surprising. The vocal performance had emotion in it — slight breathiness in the verses, a bit more push in the chorus. The acoustic guitar sounded real enough that I had to remind myself this was AI. The song structure was solid too: intro, verse, chorus, verse, chorus, bridge, final chorus. Nothing revolutionary, but exactly what you’d expect from a good indie folk track.

Time spent: About 45 minutes, mostly tweaking lyrics and regenerating twice.

Verdict: This is where Suno earns its reputation. Vocals are its superpower. If your track needs singing — especially in pop, folk, or rock — Suno is the tool to beat right now. It has over 2 million paid subscribers and honestly, I get why.

Would I publish this? Yes. With maybe one more lyric pass, this could go on a playlist and not sound out of place.

Track 2: Cinematic Orchestral Piece — AIVA

The prompt: I selected “cinematic” as the style profile, chose a minor key, set the tempo to 72 BPM, and asked for strings, piano, and a gradual build.

What happened: AIVA works differently from the other tools on this list. You’re not typing a freeform prompt — you’re selecting parameters. Style, key, tempo, instruments, duration. It feels more like filling out a form than having a conversation.

But the output? Genuinely impressive. The orchestral arrangement had layers — cellos underneath, violins carrying the melody, piano adding texture in the quieter sections. The build from a sparse intro to a full orchestral swell felt natural, not forced. I could picture this playing over a film trailer.

The big caveat: AIVA only gives you full copyright ownership on the Pro plan at €49/month. The cheaper Standard plan ($11/month) means AIVA retains copyright. So if you’re planning to use this commercially, read the fine print before you commit.

One thing I loved — AIVA lets you export MIDI and stems. So even if the AI-generated arrangement isn’t perfect, you can pull it into a DAW and tweak individual instruments. That’s something most other AI song makers don’t offer.

Time spent: About 30 minutes. Less time prompt-writing, more time exploring the parameter options.

Verdict: If you need orchestral, classical, or cinematic music, AIVA is in a league of its own. It’s not trying to be a pop song generator, and that focus pays off. Not great for modern genres though.

Would I publish this? As background music for a video or podcast, absolutely. As a standalone “song,” it’s more of a composition than something you’d add to a Spotify playlist.

Track 3: Lo-fi Hip Hop Beat — MusicWave.ai

The prompt: “Chill lo-fi hip hop beat, rainy day vibes, vinyl crackle, mellow piano chords, slow boom-bap drums. Something you’d study to.”

What happened: This is the track where I went off-script a little, because MusicWave.ai has this feature I wanted to try — image-to-music. Instead of just using a text prompt, I uploaded a photo I took of a rainy window with city lights blurred in the background.

And it worked. This AI song maker analyzed the mood of the image and generated a lo-fi track that genuinely matched the vibe. Muted piano, soft drum pattern, a bit of atmospheric texture in the background. Was it exactly what I had in my head? Not perfectly. But it was close enough to be spooky.

After the initial generation, I ran the track through MusicWave’s built-in stem splitter to isolate the piano and drums separately. Why? Because I wanted to adjust the balance — the drums were a touch too loud in the original mix. Having stems meant I could pull the drum layer into a free editor, lower the volume, and recombine. That workflow — generate, split, adjust — is something I haven’t found on most other platforms.

I also tried the voice swap feature on a whim, adding a soft vocal hum over the beat. It added a nice human texture to the track without actual lyrics.

Time spent: About 1 hour and 15 minutes. The image-to-music part was quick, but I spent extra time playing with stems and the voice swap.

Verdict: MusicWave.ai surprised me. It’s not just a song generator — it’s more like a mini production studio in a browser tab. The image-to-music feature is a fun creative trigger, and having stem separation built in means you can actually refine what the AI gives you instead of just accepting or rejecting the whole thing.

Would I publish this? The lo-fi beat? Yes — it’s solid study/chill music. The kind of thing that works perfectly as background audio for a YouTube video, a Twitch stream, or just a focus playlist.

Track 4: Electronic/Synthwave — Udio

The prompt: “Dark synthwave, 1985, neon-lit highway at midnight. Heavy analog synths, pulsing bassline, gated reverb drums. Think Kavinsky meets Perturbator.”

What happened: Udio generated something punchy on the first try. The synths had grit. The bassline was thick. The drums sounded like they were pulled straight out of a John Carpenter soundtrack. Instrumentally, this was probably the best single generation of the entire weekend.

But I wanted to push it further. This is where Udio’s inpainting tool comes in — you can highlight a specific section of the track and regenerate just that part. The intro felt a bit abrupt, so I selected the first 8 seconds and asked for “a slow synth pad build-up before the drums kick in.” Third try got me exactly what I wanted: a moody atmospheric opening that builds tension before the beat drops.

That level of control is what separates Udio from the crowd. Most AI music tools give you an all-or-nothing choice — you either keep the whole song or throw it away and start over. Udio lets you fix the parts that aren’t working without losing the parts that are.

One downside: the vocals. I tried adding a vocal layer and it sounded flat and robotic compared to what Suno produced. For instrumental tracks, Udio is excellent. For anything with singing, it’s a step behind.

Time spent: About 1 hour. The initial generation was fast, but I spent time using the inpainting tool to refine the intro and bridge.

Verdict: Udio is the producer’s choice. If you want hands-on control over your AI-generated music and you care about instrumental detail, this is the platform. The inpainting feature alone is worth trying. Just don’t expect great vocals.

Would I publish this? The synthwave instrumental? In a heartbeat. It’s got real energy. I’d use this in a short film or a gaming montage without hesitation.

Track 5: Upbeat Pop — Riffusion

The prompt: “Catchy upbeat pop song, summer vibes, hand claps, bright guitar riff, feel-good energy. Like a Carly Rae Jepsen deep cut.”

What happened: Riffusion is fast. Noticeably faster than everything else I used this weekend. Where Suno takes 60-90 seconds and Udio is similar, Riffusion had a track ready in under 40 seconds.

The approach is unique — Riffusion generates music through spectrograms (basically visual representations of sound). The result has a distinctive quality. The instrumental was bright and catchy — the guitar riff had a nice bounce to it, and the rhythm section was tight.

But the vocal attempt didn’t work. I asked for an upbeat female vocal and got something that sounded like a text-to-speech engine trying to sing karaoke. Not terrible, but not natural. Nothing close to what Suno delivered on Track 1.

So I ditched the vocal version and went with a pure instrumental. As a pop instrumental, it actually works well — the kind of background music you’d hear in a travel vlog or a product launch video.

Time spent: About 40 minutes. Generation was fast, so most of the time was spent trying (and failing) to get usable vocals.

Verdict: Riffusion is the speed demon of AI music. It’s currently free during its public beta, which makes it great for experimentation and quick ideas. The instrumental quality is solid, but if you need vocals, look elsewhere.

Would I publish this? The instrumental version, yes. Perfect for content creators who need quick background music. The vocal version? No chance.

The Final EP — What I Ended Up With

Here’s what my weekend produced:

Total time: roughly 5 hours and 10 minutes across the full weekend.

Five tracks. Five different tools. About five hours of work. A year ago, producing a 5-track EP — even a rough one — would have taken me weeks in a DAW.

What I Learned

Vocals are still the differentiator. Suno is miles ahead of everyone else when it comes to AI-generated singing. If your music needs vocals, start there. Everything else either doesn’t offer vocals or produces results that sound artificial.

Control varies wildly between platforms. Udio gives you surgical precision with its inpainting tool. MusicWave.ai gives you stems and voice swapping. AIVA gives you MIDI exports. Riffusion gives you speed. Suno gives you the best raw output. No single tool wins on everything.

The “all-in-one” approach is underrated. I didn’t expect MusicWave.ai to be as useful as it was. Having stem separation, image-to-music, and voice tools built into the same platform where you generate the song meant I spent less time switching between apps and more time actually creating. For the track that needed the most post-generation work (the lo-fi beat), that integrated workflow made a real difference.

Free tiers are genuinely useful. I did this entire experiment without spending a dollar. Suno’s free tier, Udio’s free tier, AIVA’s free tier, MusicWave.ai’s 10 free credits, and Riffusion’s open beta all gave me enough to produce at least one quality track each.

AI-generated music isn’t replacing producers — it’s creating new ones. The people most excited about these tools aren’t professional musicians (though many are using them as starting points). They’re content creators, podcasters, indie game developers, and hobbyists who never would have made music otherwise. As one industry analysis noted, AI song generators are lowering the barrier to music creation in a way that DAWs never quite managed.

Which AI Song Maker Should You Start With?

If you’re thinking about trying this yourself, here’s my honest take based on the weekend:

Start with Suno if you want complete songs with vocals — it’s the most polished overall experience. Go with Udio if you’re more hands-on and want to edit specific sections. Try MusicWave.ai if you want one tool that does everything — generation, stems, voice swapping, even video. Pick AIVA if your focus is cinematic, orchestral, or classical. And use Riffusion if you just want to fire off quick ideas and see what sticks.

Or do what I did — use all of them. They each have strengths, and a 5-track EP built across five platforms sounds like a challenge, but it was honestly the most fun I’ve had making music in a long time.

The tools aren’t perfect. Some generations missed badly. Some prompts needed three or four tries. But the stock music era is ending — and what’s replacing it is more creative, more accessible, and a lot more interesting.

Now if you’ll excuse me, I have a 10-track album to plan for next weekend.