How I Made 67 Songs in 25 Days

By Evechon · Cyprus · 4 August 2026

Between 18 June and 13 July 2026 I published 67 finished songs, each with its own official lyric video. You can check the dates on the channel.

Here is the part that matters more: in the same period those videos earned a combined 4,746 views and 28 subscribers. So this is not a success story. It is a process story with an honest ending, which I think is more useful than the version where someone sells you a course.

What the workflow actually is

There is no secret tool. The pipeline is four stages, and only two of them are automated.

1. The brief. I decide what a song is about before anything is generated — a situation, not a genre. “The lift is broken, the coffee spills, and the day hasn’t decided to start yet.” That became Meet Me Downstairs. If I skip this step and ask for “an upbeat pop song,” I get something that sounds like a stock music library, every time.

2. Generation, then rejection. This is the AI part, and the ratio is the whole story: most of what comes back is discarded. Not tweaked — discarded. The bottleneck was never making music. It is deciding what is worth keeping, and no tool does that for you.

3. Lyric editing. Every published lyric has been rewritten by hand. Generated lyrics have a recognisable failure mode: they are fluent and say nothing. They reach for “forever” and “the stars” when the specific detail is what makes a line land. A second-hand bike and a father laughing by the driveway beats any amount of cosmic imagery.

4. The publishing pipeline. This is where the speed comes from, and it is ordinary software — Python against the YouTube Data API. Descriptions, chapters, tags, thumbnails, caption files, Shorts cuts and the scheduling all run from scripts. Rendering a lyric video is deterministic work; a computer should do it.

What it gets wrong

Speed hides problems until they compound. Three that cost me real time:

Everything sounds like a genre, not like a person. Generated music converges on the average of its training data. You can push it toward a reference, but the result is competent rather than distinctive — and competent is invisible.

Volume buries you. 67 songs in 25 days means no song ever gets a push. Each one gets a day, then the next one buries it. I would trade sixty of them for six that had a fortnight of attention each.

The clips were wrong for months. I was cutting vertical clips by putting the 16:9 lyric video in the middle of a 9:16 frame and filling the rest with a blurred copy of itself. Roughly 65% of the screen was blur. That reads instantly as a recycled horizontal video, and short-form feeds punish it in the first second. Median Short: six views. I only found it by looking at a single exported frame properly, months in.

On disclosure

Every release is marked as AI-assisted with the distributor, and it says so on this site. I get why that costs listeners. I would still rather say it, because the alternative is a claim I would have to keep defending, and because platforms are increasingly detecting it anyway — Deezer already excludes undisclosed AI tracks from algorithmic surfaces. Disclosure is protection, not liability.

The honest framing: I am not pretending a machine wrote nothing, and I am not pretending I played every instrument. The songwriting brief, the lyric edits, the arrangement calls and the decision about what is good enough to publish are mine. Most of what gets generated never leaves the folder.

The numbers, in full

Songs published: 67 in 25 days, 18 June – 13 July 2026
Total channel views: 4,746 across 224 videos
Median views per Short: 6
Subscribers: 28
Google clicks to this site, all time: 0
Songs on streaming platforms: 4 of 67

I publish these because every article about AI music production stops before this part. Making the music turned out to be the easy half. Getting one person to hear it is the entire problem, and no amount of throughput touches that.

What I would do differently

Release six songs, not sixty-seven. Spend the time saved on the first two seconds of every clip rather than on the next track. Build the audience before the catalogue, because a catalogue nobody knows about is a folder, not a body of work. And look at your own output on a phone, at arm’s length, the way an actual listener will — I would have caught the blurred-frame problem in five minutes.

If you want to hear the result

The whole catalogue is on the music page with full lyrics, or as lyric videos on YouTube. The four distributed singles are on Spotify.

Writing about AI and music? Everything you need — bios, assets, contact — is on the press page, free to use without asking.