Originally posted 2025-02-26, updated 2025-05-26
Mystery “Celtic” New Age from udio
I got the original version of this out of udio.com last Octber from a simple prompt like “new age/celtic/harp/female vocal/lush arrangement”
I thought it had a good sound and interesting vocal line but the weird thing was that the vocal was in an unknown language. I never had udio do this before even after a few hundred different song generations of all types. But the weirdest thing was that I specified “instrumental”…
Nothing special about the arrangement; just a short intro and outro with uillean pipe and three verses (that may be identical).
“Come back home to me” from udio
Experimenting with its lyric writing option
“Come back home to me”
In reading the Reddit group for users of Udio, one contributor pointed out he got exceptionally good results with a very detailed Prompt he had rewritten via ChatGPT. The Prompt was:
“A deeply emotional soft rock ballad with elements of country and folk, set in the early 1970s. The instrumentation includes warm acoustic guitars, gentle electric guitar leads with a touch of reverb, and a steady yet delicate drumbeat. The song is sung by a female vocalist with a rich, expressive, and slightly husky voice—think Linda Ronstadt, Stevie Nicks, or Emmylou Harris. Her delivery is raw and heartfelt, filled with longing and desperation, capturing the heartbreak of a young woman waiting for her lover to return from the Vietnam War. The arrangement builds from an intimate, acoustic intro into a powerful, soaring chorus with harmonies, evoking artists like Fleetwood Mac and Crosby, Stills, Nash & Young. The mood is wistful, dramatic, and deeply moving, with lush instrumentation that swells in the choruses before fading into a quiet, haunting outro.”
I thought this was intriguing. Even though this wasn’t a style of song I was that interested in, I put the prompt just as he wrote it into udio and generated a 2:11 song, with lyric generation turned on as well. udio creates two songs based on your “prompt”, both of which I include above (“Version 1” and “Version 2”) I didn’t do any tweaks, but just downloaded them and created some basic players for them.
If you compare the songs that udio generated, you can see that it wasn’t able to incorporate all the concepts in the Prompt: it’s more of an intimate folky style and doesn’t have a “powerful soaring chorus” or “swelling lush instrumentation”. I think Version 2 also uses piano in the arrangement that doesn’t show up in the Prompt.
The intelligibility of udio singers is sometimes not so good, and I got the feeling it was generating somewhat too many words, and some other minor issues.. But this is the kind of thing you tweak in the refinement phase.
To generate the illustration, I used Imagen with this prompt:
“photorealistic illustration described as follows: a young woman aged 18 – 20 sits at a desk facing away from the viewer. It’s a small room, probably on the second floor. Her hairstyle and clothing indicates the date and location is late 1960s or early 1970s America. In front of her is a window showing a winter landscape. She’s reading a letter. On the desk is a small framed photo of a smiling young man in a US Army uniform. There are a couple posters on the wall of the room protesting the Vietnam war.”
Lyrics for version 1 (also generated by udio — unedited)
These can use plenty of work in my opinion, but that’s the beauty of udio: you can use your own lyrics, modify their auto generated ones, or even create instrumental tracks. (I used the current version 1.5 of udio in January or February of 2025.)
[Verse 1] The letters lay untouched on the floor Whispers in these walls speak of war She's held onto hope through storm and rain Dreams held aloft, but the pain remains Silhouettes dance in her midnight skies Fogging the truth, she waits, she cries [Chorus] Oh, come back home to me The wind carries your name, you see With every tear shed down the lane In my heart, you still remain [Verse 2] The clock hands move, slow grains of sand She traces his words with trembling hands In the mirror, her eyes tell the tale Love and longing, a veil so frail Promises echo in the empty room Silent screams held in the waiting gloom [Chorus] Oh, come back home to me The wind carries your name, you see With every tear shed down the lane In my heart, you still remain [Bridge] The moonlight whispers secrets untold In her heart, she's forever bold Hope unwavering through the night’s song Knowing her love will keep her strong [Chorus] Oh, come back home to me The wind carries your name, you see With every tear shed down the lane In my heart, you still remain
The piece has possibilities: the first thing I’d do is pare down the lyrics, but the ideas the AI came up with are good and in any case different than I would have ever come up with. I have the udio first paid tier, which isn’t very expensive ($10 month at present).
I strongly recommend musicians check this out with an open mind. I’ve been around long enough to have lived through all the things that were going to kill music: synthesizers/drum machines, recordable cassettes, the ability to make unlimited digital copies of LPs, then CDs, MIDI, RIAA wars on MP3 sites, computer applications like Fruity Loops, Reason, Ableton Live. Then the genre wars ( jazz?! that’s not Music! rock?? that’s not Music! Rap?? That’s not Music!)
Somehow, creativity always wins.
Is AI what’s finally going to wipe out creativity in music? From a brief review of history, I don’t think so. But it’s going to be a hell of a ride.
(Last updated 2025-05-26)
crow caw music works 