Can You Make Real Music With AI? The Questions I Keep Getting Asked

I am not the one to decide what counts as real music. What I can speak to is my own experience, and it is this: making good music with AI does not happen by typing a prompt and publishing whatever comes out. The composition still has to be yours: the lyrics, the melody, the structure, the taste to know when something is working. What these tools actually change is how fast you find out whether an idea was worth having.

I have been making music for about thirty years and released it under nine artist names, most actively as Cometfall, a cosmic alternative rock project from Helsinki. By day I build AI search analytics software. That combination means I get asked the same handful of questions constantly, in comments, in DMs, and by friends who found a Spotify link and were not sure what to make of it.

Nobody has interviewed me about the music. So I wrote the questions down and answered them myself. The questions are real. The interviewer is me.

In short

1. Imagination was never the bottleneck in songwriting. Verification was, and that is what got cheap.
2. Songs made purely by prompting are audible from a long way off, and the lyrics give them away first.
3. AI cannot supply the part that matters most: human energy, live performance, and the persona of an artist.
4. Volume is a real problem. Deezer received around 90,000 fully AI-generated tracks per day in June 2026, more than half of everything uploaded, and that content drew only 1% to 3% of streams.
5. No song was ever 100% one person's doing. Authorship lives in the seed, and how the last steps get finished is a question of method rather than ownership.
6. For an independent artist, the interesting shift is not cheaper production. It is being able to test an idea on a real audience before committing to it.

Do you need to play an instrument well to write good songs?

No, and I want to be precise about that, because I do play. I spent close to a decade playing bass in various lineups, I compose mainly on guitar, and the bass on the Cometfall records is me. What I am not is a virtuoso. I was never willing to spend years mastering a single instrument when I could play a few of them well enough to get an idea out, and I have always identified as a songwriter rather than a player.

My instruments are writing tools. Enough guitar to find the chord movement and hear where a song wants to go, enough bass to hold the low end of a real arrangement together. That is a different relationship with an instrument than a lead player has, and it is the one most songwriters actually have.

The part I trust is the imagination. I can write lyrics. I can hear a melody before it exists. I know what genre an idea wants to be, how the finished thing should sound, and how it should be presented. Those are separate skills from execution, and the music industry has always quietly known it. That is why producers, topliners, and songwriters get credits of their own.

What is new is that you no longer need to convince four other people to donate their evenings before you can find out whether your idea works.

Can you actually make real music with AI?

Again, I would rather talk about good than real, and yes. The reason people doubt it is that they are imagining a different process than the one I use. Most people who have never made a song assume the hard part is the sound. It is not. The hard part is having something worth turning into a sound.

Here is the honest shape of it. I bring the composition: lyrics, melody, structure, the genre it wants to be, the emotional target. The tools collapse the distance between that and actually hearing it. Twenty years ago, bringing a vision to life meant instruments, bandmates, rehearsal rooms, and recording time, and you had to spend all of that before you could tell whether the idea was even worth pursuing.

That speed does not subtract one gram of the imagination that went into the composition. The song was mine before any tool touched it. What changed is that I find out sooner whether I was right.

Do you still make music without AI?

Yes, constantly. There has never been a single route from an idea to a finished song, and AI did not replace the other ones.

Sometimes a song gets there by jamming it with other people until it becomes something none of us planned. Sometimes it gets there alone at a computer with MIDI, which is how most of my catalogue was built for twenty years. Sometimes it now goes through AI. Very often it is a mixture: a part played, a part programmed, a part generated, then rewritten.

None of those is the correct way and none of them is the wrong way. They are different roads to the same place, and which road a song takes usually depends on what the song needs and who happens to be around.

Is a song ever 100% yours?

No, and mine never have been. This is worth saying plainly, because a lot of the argument about AI and authorship assumes a kind of purity that does not exist in music and never did.

Think about what actually goes into a finished song. Whatever I was listening to when the melody arrived, and what that did to its shape. The people I first jammed the raw idea with, who pushed it somewhere I would not have taken it alone. Whoever recorded it, whoever mixed it, whatever decisions happened in production that I did not make. I have never finished a song that works and been able to say honestly that it was 100% my doing.

What I care about is the seed: the idea, the core melody, the words. That part I want to be mine, and it is. Beyond that, the particular combination that takes the last steps is not the interesting question. Humans, AI, or humans and AI together. It does not matter much. Good music matters.

The thing I do find ugly is people taking credit for work that is not theirs. That is not a new problem and it is not really an AI problem, but AI makes it easier and there is a lot of it about. It is also a different failure from using a tool. One is a question about how a song got made. The other is just lying.

Are these old songs or new ones?

Both. Part of the catalogue is archive, and part of it is written this month.

The archive part is real and specific. "Keeper of the Light," on the album coming in December, started as a MIDI file from 2007. "Paper Radio," the single released on 11 September 2026, came from a 2010 demo. Finishing something that old is a strange kind of collaboration, because you are working with a person who no longer exists. He had worse gear, more time, and considerably more confidence.

Not all of those demos survive contact. The ones that do are the ones where the melody was already right and only the production was missing. That is the filter: if the idea worked in 2007 on a General MIDI patch, it works now.

But most of what I am doing is new. New lyrics, new melodies, ideas that occur to me this month and get finished this month. The archive is a starting inventory, not the point. If excavation were the only thing I could do, this would be a nostalgia project, and it is not.

How do you know whether a song idea is worth pursuing?

You listen to a plausible version of it early, and then you decide. This is the part of the process the tools genuinely changed, and almost nobody talks about it.

The old process punished exploration. To find out whether an idea had legs, you booked time, gathered people, learned the parts, and recorded something. After all that you might discover the chorus did not land. Because verification was so expensive, you only tested ideas you were already confident about, which means the interesting, risky, strange ones mostly died in a notebook.

Now I can run a laboratory. I take an idea, hear a version of it, and then ask better questions. Does this want to be alternative rock or something heavier? Does it belong to Cometfall or to my black metal project? Does the chorus actually lift, or did it only lift in my head? Which of my projects is this song's home?

That last question matters more than it sounds. Working out which lineup an idea fits is genuinely creative work, and now I can hear the options instead of guessing at them. Calling that an extension of imagination is not a defence, it is just accurate. I explore more of my own ideas than I ever could before, not fewer.

Can you tell when a song was just prompted from zero?

Almost always, and the lyrics give it away first.

Prompt-only songs have a specific flatness. The structure resolves too neatly, the sections do not earn each other, and there is usually a moment where you can feel that nobody decided anything and the track simply continued. But the reliable tell is the writing. You cannot shortcut lyrics. Generated lyrics reach for the nearest available image, rhyme on the obvious word, and never say anything that cost the writer something. Anyone who reads carefully will catch it inside a verse and a half.

Getting something genuinely good out of an AI-assisted workflow takes what it always took: effort, taste, a clear vision of the target, a real seed melody, deliberate structure, and the specific craft of writing lyrics that fit a particular melody. Syllable counts. Where the stresses fall. Where a line has to breathe. That skill is invisible to people who have not tried it, and it is the one that separates the two piles.

The tool does not have taste. That is still the entire job.

What can AI not do in music?

It cannot supply human energy, and that is the part that matters most.

I would split music into two halves. The first is writing and producing: getting from an idea to a recording that stands on its own. With taste and effort you can genuinely finish that half alone, and publish real songs out of it. That is the half I have been working in.

The second half is energy. What musicians put into a song when they are in a room together. What happens live. The persona of an artist, the history behind them, the reason anyone cares what you have to say. That part is human or it is nothing, and it has not shifted an inch.

This is also where the volume problem sits, and it is worth separating from the tools question. In June 2026 Deezer reported receiving nearly 90,000 fully AI-generated tracks every day, more than half of all daily uploads, up from 60,000 in January of the same year. The revealing number is what happens next: that material accounts for only 1% to 3% of actual streams. An enormous amount of it is uploaded and essentially nobody listens to it. That is an argument about volume, not about whether a songwriter should use the tools.

My own lines are simple. No voice cloning. No impersonation. No training on a living artist's catalogue in order to imitate them. And no invented band: Cometfall is a solo project and says so everywhere.

Which is exactly where it is heading next. I have played in a lot of lineups over the years and I know good musicians. The honest truth is that the next phase is taking these songs to them and selling them the concept, because the material is finally ready for it and the energy is the thing I cannot make alone. The recordings are the pitch.

I am deliberately not precious about the form. A standing live band, a rotating cast, or session players on the next record would all be a win. What I am specific about is the intent: I want these songs in a room with people who can take them somewhere I cannot take them alone.

I wrote about this tension at more length in Notes on AI and Music, from a Lifetime of Making It.

Should you tell listeners you used AI?

I do not lead with it and I do not hide it. If someone asks, they get a straight answer.

"Made with AI" is not a genre and it is not a selling point. It is a detail about tooling, like which DAW or which compressor you used, and nobody opens a record by announcing their plugin chain. Leading with it also invites people to evaluate the process instead of the song, which is a fight the song always loses.

What I refuse to do is invent a band. No fake members, no fake history, no photographs of five people who do not exist. That is the line where disclosure stops being a stylistic choice and starts being honesty.

Is it cheating to test songs on social media before they are finished?

I do not think so, and this is probably my most unorthodox habit, so let me describe it plainly.

I publish at demo stage. I put a lot of versions and short clips of the same idea into social feeds, deliberately, before anything is finished. Then I read what happens: which hook people replay, which version holds attention, what gets saved rather than merely liked. Those signals feed the decision about which songs get finished, which get promoted, and which quietly stay in the folder.

It reverses the traditional order. Normally you finish, release, and then find out. I find out, and then finish.

People who think about music purely as art will find this distasteful, and I understand why, because it sounds like optimising a feeling. But the traditional order was not chosen for artistic reasons either. It existed because pressing and distribution were expensive and gatekept, so you got one shot and had to spend it on your best guess. As an independent artist in 2026 I am not bound by that constraint. I can test the sound first and build on what I learn.

The entrepreneurial framing is the honest one: I look for signal before committing resources. It does not change what a song is about and it does not write the lyrics for me. It only tells me which one to send out first.

Why release under nine different artist names?

Because streaming platforms build a listener profile per artist, not per person. Put black metal, reggaeton, phonk, and cosmic alternative rock under one name, and every algorithm trying to find your audience receives a contradictory signal and gives up. So does any listener who liked one thing and got served the other.

So the projects are separate. ihmissuti is the home base, indie rock and dreamy pop, where two decades of songwriting lives. Cometfall is cosmic alternative rock. rotwurm is black and pagan metal. Amzart is experimental rap. erizou is Spanish-language dance music. halogeist is industrial electronic. Each one gets a coherent audience and its own visual language.

It connects to how I work, too. When an idea arrives, part of deciding whether to pursue it is deciding whose it is. Having several distinct projects means far fewer ideas get thrown away for not fitting the single box I would otherwise have.

How does one person release an album a month?

By treating it as an engineering problem, which is the part of my day job that transfers.

The entire catalogue lives in a git repository. Every track has a folder holding its lyrics, its story, its artwork, its video assets, and its release metadata in YAML. On top of that I run AI agents that do the work a small label's junior staff would do: check what is blocking the next release, generate the Spotify Canvas files, draft the social calendar, find peer artists worth supporting, reconcile the repository against what is actually scheduled at the distributor, and collect weekly statistics to tell me what is working.

I approve, the agents execute. It is the only reason three albums and a run of singles in a single year is survivable alongside a full-time job.

What does Cometfall sound like?

Cometfall is cosmic alternative rock from Helsinki: psychedelic, spacious, and heavier than it first sounds. Progressive structures that do not resolve where you expect, real bass low in the mix, wide space around the guitars, and choruses built to be big.

The lyrics reach into the cosmos, the mind, and the afterworld. Old folktales, lighthouse keepers, the machinery of stars. They tend to land somewhere quite human and quite stubborn, usually near the joy of still being alive.

If you want reference points, the neighbourhood is somewhere between Soundgarden, Tool, Tame Impala, and Muse.

The debut album Buried Elsewhere arrived in August 2026. Paper and Hope follows on 16 October 2026, and Move Me, Great Power on 11 December 2026, with singles on most Fridays in between.

What have you learned about getting listeners from zero?

That followers matter more than monthly listeners, and almost nobody tells you that early enough.

In early September 2026 Cometfall had roughly 74 monthly listeners, several hundred streams, a healthy save rate, and a follower count in single digits. The people arriving were staying. They just were not subscribing. With six releases landing between November and December, every single follower would receive all six automatically. Spotify confirms that followers get songs from a new release in their Release Radar, provided you deliver at least seven days ahead. Listeners were the vanity number. Followers were the compounding one.

The second lesson is the demo-testing point in another form: one song packaged ten ways beats ten songs packaged once.

The third is that consistency beats intensity. A release every Friday teaches an algorithm what you are. A burst followed by six months of silence teaches it nothing.

Key takeaways

1. The scarce inputs in songwriting are imagination, taste, and lyrics. None of them are what AI supplies.
2. Cheap verification is the real change. You can now explore risky ideas because finding out costs almost nothing.
3. Prompt-only output is recognisable, and lyrics are where it falls apart first.
4. Human energy, live performance, and artist persona remain entirely human, which is why a solo catalogue is a pitch to musicians rather than a replacement for them.
5. There is no single correct route from an idea to a song. Playing it with people, building it alone in MIDI, and working with AI are all roads to the same place, and most songs use more than one.
6. Independent artists can now test material with a real audience before committing, which inverts the traditional finish-then-release order.

Frequently asked questions

Is Cometfall a band or one person? One person. Cometfall is my solo project: I write the lyrics and melodies, produce the records, and play bass on them. There are no other members and no invented ones.

Which tools are actually involved? Suno at the production stage, LANDR for mastering, and DistroKid for distribution, on top of MIDI demos and recordings I made myself. Not every song takes the same route, and plenty of the work involves no AI at all. The writing is always mine.

Does Cometfall play live? Not yet. Bringing the songs to musicians is the next phase, and the form is open: a live band, a rotating cast, or session players on the next record.

How much music is coming out in 2026? Three Cometfall albums (Buried Elsewhere, Paper and Hope, and Move Me, Great Power) plus singles on most Fridays between them.

Where can I listen? On Spotify and every other major platform, and at cometfallmusic.com.

A closing note

The debate about AI and music tends to collapse two separate arguments: whether there is too much generated audio being uploaded, and whether an individual songwriter should use these tools. The first is a real problem with a measurable shape. The second is a question about craft, and craft has not changed much. You still need something to say, and you still have to write the words yourself.

There is a version of this that is genuinely good for people like me, who spent decades with more ideas than means. There is also a version that produces 90,000 tracks a day nobody asked for. The difference between them is not the software. It is whether there is a real seed underneath, and whether the person putting their name on it is being honest about how it got there.