Comment by TheAceOfHearts | Hacker News Reader

Comment by TheAceOfHearts | Hacker News Reader

One of Suno's biggest weakness is their lyrics generation, and that you can't generate lyrics without also generating a song. I think it's better to use a different LLM to generate and iterate on lyrics, which you can then pass to Suno in order to generate a final song.

If anyone here has a subscription and they can spare the tokens, I think it would be fun if someone shared a song about Hacker News.

I'm hoping that in the future tools like Suno will allow you to produce / generate songs as projects which you can tweak in greater detail; basically a way of making music by "vibe coding". With 4.0 the annotation capabilities were still a bit limited, and the singer could end up mispronouncing things without any way to fix or specify the correct pronunciation. This blog post mentions that with 4.5 they enhanced prompt interpretations, but it doesn't actually go into any technical details nor does it provide clear examples to get a real sense of the changes.

Check out custom mode -- we've added a lyrics writing flow/editor to help create and edit lyrics, as well as Remi, a more unhinged lyrics model.

We can do better on user instruction for sure, duly noted. In my experience a lot of different stuff works (emotions, some musical direction sometimes, describing parts/layers of the track you want to exist, music-production-ish terminology, genres, stuff like intro/outro/chorus), but I think of it more as steering the space of the generated output rather than working 100% of the time. This can go in the style tags or in [brackets] in the lyrics. Definitely makes a difference in the outputs to be more descriptive with 4.5.

This song was produced a long time ago from a verbatim hacker news comment, and got released on Spotify and Apple and became a favorite at home.

Your comment inspired me to upgrade it to 4.5 because it did have that AI tinny quality. https://suno.com/s/tbZlkBL7XeLVuuN0

It sounds better but has lost some magic.

Here is the original comment - https://news.ycombinator.com/item?id=39997706

In that spirit, from the same “artist” here is your comment - https://suno.com/s/AumsIqrIovVhT0c9

And

https://suno.com/s/YGlpHptX6yXJVpHq

Not sure which I like more.

Fascinating. I find that song to be unlistenable and something about the thought of anyone enjoying it seems really wrong.

Well, that’s about as low effort as Suno gets, so I’m not sure you should expect to like it more than a laugh. But when you put effort into the songwriting and shaping you can produce legitimately good music. This is my third highest ranking streaming song and it actually got some airplay on US radio stations - https://music.apple.com/au/album/too-much-winning/1788045206...

> In that spirit, from the same “artist” here is your comment - https://suno.com/s/AumsIqrIovVhT0c9

I am feeling a deja vu with that vocal part at 0:40. I definitely heard something similar somewhere.

you can do this... just toggle to "custom" mode on song generation

Their biggest weakness is that every voice has a persistent synthy quality, like it's a vocaloid it's being sung into one of those tinny microphone toys for kids. I find Udio has much more natural-sounding vocals.

Udio has these random "halfway" singer voices that I find really interesting to listen to - a little uncanny-valleyish sometimes but interesting nonetheless. Depending on genre it can be a Nat-Sinatra, a Whitney-Dion, etc., it's really cool. It would be great if it could be saved from a song then loaded back for a new one on demand.

Another thing I did with LLM which I found very useful, is to give the LLM an existing song lyrics and ask him to do a similar one with different subject I give him

Reply on news.ycombinator.com