SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.

Trying to cover original songs from a band 25 years ago with new arrangements using an AI band.

2026年5月4日 21:25

A shocking encounter with Suno AI

I am an old guy turning 65 this year, and while I had never even used ChatGPT and didn't know what this AI that everyone is talking about was, as someone who still somehow manages to keep up band activities (a working adult band), I started playing around with Suno AI, a famous music generation AI on the internet, at the end of March.
Apparently, if you input lyrics, it will create a song and sing it. I heard that AI can also generate lyrics, but I didn't really understand that, so I entered the lyrics of an old original song into the Lyrics field. I entered "boogie, blues, rock" into the Styles field. Then, wondering what to do next, I pressed the Create button. After a while, two songs were created, so I listened to them. The intro of the first song started with a riff like that famous rock and roll Chuck Berry guitar style. Well, that's how rock and roll starts, I suppose. When it entered the vocal part, a strange, light J-punk rock song started playing. Hmm, it was put together as a song in a decent way, and it even had breaks, so the composition wasn't monotonous and was arranged well. The singing was good (better than us), it didn't go off-key, and it even had some human-like inflections.
I listened to the second song as well. This time it was a slightly bluesy rock song. So, one input results in two different variations, huh? Compared to the original song, both felt far off to me. I changed the Styles and pressed Create again, and two more songs were generated. Still not quite right. I changed it again and pressed Create; after repeating the generation process several times with various styles, it started to get a little closer to my image. However, I ran out of credits for the free trial. I didn't think I could create anything satisfying with the 50 credits I'd get the next day. I guess I had no choice but to switch to a paid plan. The next day, I was stuck in a generation swamp all day long. I generated 34 songs and finally managed to be satisfied with one for now. The melody, singing style, and chorus were different, and the chord progression was quite different too, but since the original song was a simple boogie-woogie, the atmosphere was somewhat close. Well, as long as the soul of what we want to perform is similar, that's fine. Since we are a live performance band, I consider the sound source to be just a demo song; naturally, the band arrangement will be added during the actual performance.
Although it is unfinished in many ways, I have at least created a demo sound source that I can let people hear. It means I can share my original songs with others. Things that I used to find troublesome to produce (I used to do multi-track recording on a cassette MTR) can now be done by one person with a single computer, or rather, even a single smartphone, to produce a sound source with vocals.
Even though I hadn't touched DTM until now, it's such a fun era where you can easily create full-band sound sources if you have an internet connection, and you can share your own songs.

Reviving songs from a band that never reached the stage of sound production back then

With that in mind, I decided to try and have Suno AI generate covers of my old original songs, and I took on the challenge. First, generating with just lyrics input makes it difficult to get a melody close to the original song, so I tried generating by humming into Suno's Cover feature. I sing the song into my iPhone. I checked and corrected the lyrics read into the uploaded file (in V5 it was displayed in Roman alphabet reading; in V5.5 it was displayed in Japanese reading), and tried various genre settings in Styles to generate. Even with just vocal input, if you put meta-tags properly in the lyrics field, it creates a pretty good backing performance that matches the melody. However, it quite often does whatever it wants with the intro, interlude, chorus, and ending.
So next, I decided to let Suno listen to me playing guitar and singing.
Actually, I had a cerebral infarction two years ago, and I still have some paralysis in my left fingers, so I still can't press guitar chords well, but I've managed to press the bass to some extent. So, I tried inputting by singing while playing the bass. Then, Suno properly picked up the performance and added accompaniment. This allows for a certain flow of the song. I used prompts to specify the actual band instruments like guitar, piano, saxophone, and drums to generate the song.
I researched various meta-tags and prompt instructions, and after trying them on several songs, I was able to create songs that were close to the original or had new arrangements. But it feels a bit too clean and too polished. So, I tried having Suno "Cover" the audio from a live video. The sound generated from the initial count-in has a live, raw feel to it. It even copied the riffs and ad-libs from the intro and ending. The one I covered with V5 was nice. I feel like V5.5 always goes too far, screams too much, or does too much on its own even when I use Cover. It feels like it's working too hard.
I decided to create a Voice (persona) using the style and voice of the version created by covering this live song, and have it operate as an AI band called "AI-Boogie Lunch," the only cover band for the real band "Boogie-Woogie C-Lunch."

AI-BOOGIE LUNCH #1 (C-Lunch Cover Demo Collection)

This is the world's only cover song collection by an AI band of the Okinawan band "Boogie-Woogie C-Lunch" that I (mino) belong to. (Not that there could be any others!) This is a demo collection that has not yet undergone fine adjustments or mastering.