I’ve been experimenting with AI music tools recently, and ended up making something I didn’t expect: an 80s glam-metal parody song about software engineering chaos.
I wrote all the lyrics myself (“One Hot Feature”), fed them into Suno, and the model generated a surprisingly polished track — full vocals, instrumentation, arrangement, and mix. The whole thing took maybe 20 minutes end-to-end.
What surprised me most wasn’t the audio quality (which is good), but how well the model adapted to the tone, humor, and style. I didn’t prompt it with any melody — just the words — and it built a full composition around them.
Curious to hear what others think about the state of AI music generation, especially:
Are we heading toward “everyone can make a polished song”?
What does authorship look like here?
How do you all feel about AI generating new music in a distinct style?
Happy to answer questions about the process.
Rochus•2mo ago
Cool, convincing production. The audio quality of Suno is not there yet, but the model is very, very musical, and the resulting arrangements, melodies and instruments/voices are just great (and I say this as a musician, who was also a professional producer twenty years ago). There was recently a model posted on HN (assumingly from China) which has a significantly better audio quality, but is less musical so far (see https://news.ycombinator.com/item?id=45853478).
Your text is funny and surprisingly well suited for the song and the singer. Personally I very much like the possibility to generate great music with algorithms. I followed the developments for more than thirty years and did even a master thesis about an expert system which improvises music in the nineties, but only since this year we have AI good enough to generate convincing music. There were even recent scientific studies which found that essentially no one can distinguish AI from human music with certainty anymore (people had to select one AI out of three songs and only 3% got it right!). Nevertheless, there are still many people who dislike AI music; however, this is largely emotional and not objectively justifiable, even more that they very likely wouldn't notice the difference.
ztp123•2mo ago
Thanks for the thoughtful breakdown.
I completely agree with you on Suno’s “musicality over fidelity” profile. What surprised me most was how well it adapted to lyrics I wrote without any melody guidance. I didn’t prompt it with line breaks, rhythm, or phrasing; it just grabbed the vibe and built a whole arrangement around it.
I’ll check out the model you linked, I haven’t experimented much outside Suno yet. I’m curious how far fidelity can be pushed while still keeping that kind of intention-aware musical structure.
I can definitely imagine this becoming part of a normal songwriting workflow: not AI music as the finished product, but as a way to sketch ideas or generate demo material much faster.
Rochus•2mo ago
> it just grabbed the vibe and built a whole arrangement around it.
I would also be very interested to know why and how this works. I usually upload the audio of my songs, and it immediately gets everything including harmonies, structure, melodies and lyrics, just amazing.
> not AI music as the finished product
I'm even amazed by music fully generated by AI (even when based on uploaded audio); with a bit of tweaking amazingly interesing music can be generated (see e.g. https://rochus-keller.ch/?p=1428).
esafak•2mo ago
Don't you need some Jolt Cola to wash down the pizza?
ztp123•2mo ago
I would love a Jolt Cola! That is a memory unlock for sure.
ztp123•2mo ago
I wrote all the lyrics myself (“One Hot Feature”), fed them into Suno, and the model generated a surprisingly polished track — full vocals, instrumentation, arrangement, and mix. The whole thing took maybe 20 minutes end-to-end. What surprised me most wasn’t the audio quality (which is good), but how well the model adapted to the tone, humor, and style. I didn’t prompt it with any melody — just the words — and it built a full composition around them.
Curious to hear what others think about the state of AI music generation, especially:
Are we heading toward “everyone can make a polished song”?
What does authorship look like here?
How do you all feel about AI generating new music in a distinct style?
Happy to answer questions about the process.
Rochus•2mo ago
Your text is funny and surprisingly well suited for the song and the singer. Personally I very much like the possibility to generate great music with algorithms. I followed the developments for more than thirty years and did even a master thesis about an expert system which improvises music in the nineties, but only since this year we have AI good enough to generate convincing music. There were even recent scientific studies which found that essentially no one can distinguish AI from human music with certainty anymore (people had to select one AI out of three songs and only 3% got it right!). Nevertheless, there are still many people who dislike AI music; however, this is largely emotional and not objectively justifiable, even more that they very likely wouldn't notice the difference.
ztp123•2mo ago
I completely agree with you on Suno’s “musicality over fidelity” profile. What surprised me most was how well it adapted to lyrics I wrote without any melody guidance. I didn’t prompt it with line breaks, rhythm, or phrasing; it just grabbed the vibe and built a whole arrangement around it.
I’ll check out the model you linked, I haven’t experimented much outside Suno yet. I’m curious how far fidelity can be pushed while still keeping that kind of intention-aware musical structure.
I can definitely imagine this becoming part of a normal songwriting workflow: not AI music as the finished product, but as a way to sketch ideas or generate demo material much faster.
Rochus•2mo ago
I would also be very interested to know why and how this works. I usually upload the audio of my songs, and it immediately gets everything including harmonies, structure, melodies and lyrics, just amazing.
> not AI music as the finished product
I'm even amazed by music fully generated by AI (even when based on uploaded audio); with a bit of tweaking amazingly interesing music can be generated (see e.g. https://rochus-keller.ch/?p=1428).