AI Generated / Assisted Music

Yeah I get where you’re coming from.

I’ve already gone the distributor route to Spotify and the other platforms, so I can say from experience it’s a slow burn. It’s not like you upload and suddenly things move, it takes time to even get a trickle going. I’m sitting on a few hundred listeners a month now, which is cool, but it didn’t happen quickly.

YouTube is honestly the bigger grind for me. People think this stuff is low effort because “AI did it”, but they don’t see the actual work behind it. Making the track is one thing - which in itself is a considerable bit of work, then you’ve still got to do visuals, edit videos, upload properly, try market it, and somehow connect with people. It adds up quickly. Especially if you then need to push it to various other platforms as well.

For me it’s not really about chasing money or even big numbers. It’s more about putting stuff out there and sharing ideas, stories, whatever comes to mind with like-minded people, or people who enjoy the same style - or enjoy the humour, depending on the content. That part matters more than the stats.

But yeah… I do agree with you on one thing. Every now and then it would just be nice if someone actually stopped, listened properly, and said something real. Not hype, not noise, just genuine feedback. That’s the part that’s weirdly hard to get.
Oh definitely. I’ve spent huge amount of time recording the bass, getting a overdub sorted in Fl Studio >> then to prompts/ Suno, and thank f I managed to
at least finish some of my old songs.

Because now that they’ve cut the training data, everything just sounds generic and super AI ( the main reason I never bothered with it in the first place until suno came along)

Super miffed we now have to be happy with a compact disc worth of training data.
The hatred for the “music industry” has doubled haha

IMG_0959.jpeg
 
Last edited:
Oh definitely. I’ve spent huge amount of time recording the bass, getting a overdub sorted in Fl Studio >> then to prompts/ Suno, and thank f I managed to
at least finish some of my old songs.

Because now that they’ve cut the training data, everything just sounds generic and super AI ( the main reason I never bothered with it in the first place until suno came along)

Super miffed we now have to be happy with a compact disc worth of training data.
The hatred for the “music industry” has doubled haha View attachment 1899622
I don't think any of the stuff you've published sounds generic - quite the opposite. But, I get that it takes some doing to make it exactly the way you want it.
 
Will publish this song later. gave it more upbeat melody and a bit more style changes, the record scratches is a nice touch.

Holy crap that's actually fking amazing.

If The Smashing Pumpkins released this it would be a chart topper.

You should DM them. 😅
 
I don't think any of the stuff you've published sounds generic - quite the opposite. But, I get that it takes some doing to make it exactly the way you want it.
Yeah, I'm happy with the stuff I've posted so far. But since about last week I noticed something off. So I spoke to dude who works there and all training data now has to be Warner approved. So the ocean of creativity has now become a suburban swimming pool. Suno tried to up the quality hoping nobody would notice. And I don't blame them , they are doing what they can, being hamstrung and all. It's just sad really, kinda like a censorship of sorts. Hope Time Warner crashes and burns haha

For example , this came out perfect


It's not possible do make this anymore. Just happy I got to finish at least 80% old song ideas and made some new ones.
Maybe they will tweak it. Will wait and see.
 
Yeah, I'm happy with the stuff I've posted so far. But since about last week I noticed something off. So I spoke to dude who works there and all training data now has to be Warner approved. So the ocean of creativity has now become a suburban swimming pool. Suno tried to up the quality hoping nobody would notice. And I don't blame them , they are doing what they can, being hamstrung and all. It's just sad really, kinda like a censorship of sorts. Hope Time Warner crashes and burns haha

For example , this came out perfect


It's not possible do make this anymore. Just happy I got to finish at least 80% old song ideas and made some new ones.
Maybe they will tweak it. Will wait and see.
Maybe make a custom model with your non-Suno material, and /or the Suno songs you've done that you are happy with. The workflow is a bit dumb if you do want to use your Suno stuff, I had to download my songs and then re-upload them to create a custom model, but now I can do music in that particular style and It works well for my purposes. It is a bit different from the stock 5.5 model, so might work for your requirements too.
 
Maybe make a custom model with your non-Suno material, and /or the Suno songs you've done that you are happy with. The workflow is a bit dumb if you do want to use your Suno stuff, I had to download my songs and then re-upload them to create a custom model, but now I can do music in that particular style and It works well for my purposes. It is a bit different from the stock 5.5 model, so might work for your requirements too.
Yeah , it's definitely better. Created one and it's better than standard, But it's essentially still being hamstrung.
This might sound a bit too metaphysical but imo you have to let the AI dream as much as possible.
 
If you need help give me a shout, I haven't figured out the nuances, but definitely helped me a lot.

My setup is basically using the lyrics as additional descriptions. 1000 characters really isn't enough. But basically I use chapgpt to describe the musical tone I am looking for give it the lyrics and let it add the specific tags, per lyrical line.

It has been really useful for EDM songs, I can control exactly where I get vocoder results pitch, reverb, delays.

I needed to learn basically like a sound engineer, what all the effects does, it is a hella of fun playing around, once you know what they do you can do really cool effects, honestly, adding delays, reverb and pitch shifts and micro manage every little detail.

But yeah you need to delve deep with the new model on suno it has a lot more nuances and way more complex.

But the tracks is doing well for less than a hour, but usually dies off dramatically after that

View attachment 1898435

Will publish this song later. gave it more upbeat melody and a bit more style changes, the record scratches is a nice touch.
Yeah, this is great. I can almost imagine it running on MTV in the late 90s on a rock stream.
 
Yeah, I'm happy with the stuff I've posted so far. But since about last week I noticed something off. So I spoke to dude who works there and all training data now has to be Warner approved. So the ocean of creativity has now become a suburban swimming pool. Suno tried to up the quality hoping nobody would notice. And I don't blame them , they are doing what they can, being hamstrung and all. It's just sad really, kinda like a censorship of sorts. Hope Time Warner crashes and burns haha

For example , this came out perfect


It's not possible do make this anymore. Just happy I got to finish at least 80% old song ideas and made some new ones.
Maybe they will tweak it. Will wait and see.

I disagree, you can pretty much make anything you you want. There is somethings you can't, but that isn't due to the training model, more you need know the prompts and tags.

This was a test song used a 8 seond monophonic clip from the old nokia ringtone area and pushed a monophonic arrangement.

This was a pitch experiment each word had a pitch difference +2 for each word for high pitch and -2 word each pitch for low pitch

Another experiment that intentionally breaks melody structure


Your song has a mono quality to it, just something I threw together I left the tags, so you can see how I approach out there concepts.

Suno has a bit of an issue with mono sometimes, it didn't follow it directly and only partial. But to get that effect is just to download the stems and force whatever I need into the mono channel


5.5v mostly followed the mono

5.5v more abstract prompt
The second one introduces uneven bass

Also discovered that if you mention bass synth with uneven double bass effect it selects a generic bass synth, but if you mention a particular bass synth as well as the actual bass range you can control way much more.

If you give it a generic prompt and no lyrical tags it will always default to standard "safe" melody sounds. It isn't that you can't do less quite the opposite you can do way way more, to the point you actually need to control

I resorted to version 4.5v it entirely ignored everything here.


So yeah you need to be HIGHLY specific with the prompts and tag the **** out of the lyrics with approved tags

Just some of the tags I discovered for 5.5 and these are all lyrical tags

[verse]Marks verse section✅ YesStructures vocal and instrumental arrangement
[chorus]Marks chorus section✅ YesUsually fuller instrumentation & stronger vocals
[bridge]Marks bridge/transition✅ YesGuides dynamic shift or texture change
[break]Indicates a break✅ MostlyCreates pause or energy drop
[intro]Beginning section✅ YesSets up track
[outro]Ending section✅ YesFade out or simplified section
[looped]Repeat lyric/motif✅ MostlyCreates rhythmic or melodic repetition
[stutter]Repeated/broken syllable✅ MostlyChopped vocal effect
[short burst]Clipped phrasing✅ MostlySharp, rhythmic delivery
[breathy]Airy vocal style✅ YesSoft, intimate vocal delivery
[whisper]Whispered vocal line✅ YesProduces soft, hushed delivery
[whistle]Whistling line✅ MostlyMicro-melodic effect
[whistling letters]1–3 letter whistle sounds✅ PartiallyShort melodic vocalizations
[pitch +1/+2/+3/-1/-2/-3]Pitch shift instructions✅ YesChanges line pitch
[micro pitch shift]Tiny per-word pitch changes✅ MostlySubtle melodic variation
[echoed]Reverb/delay effect✅ YesProduces repeated echo effect
[echo]Same as echoed✅ YesSynonym, works identically
[delayed]Delay effect✅ YesRhythmic echo on vocals or chops
[reverb]Adds spatial effect✅ YesCreates room/space on vocals/instruments
[processed]General processing cue✅ MostlyFilters, distortion, reverb implied
[filtered drums]Drum EQ/filtering✅ YesAlters drum texture
[double bass]Heavy bass line✅ YesEmphasized low end
[chopped synth]Staccato/glitch synth✅ YesCut-up synth patterns
[sparse]Minimal arrangement✅ MostlySparse textures, reduced instruments
[full]Dense arrangement✅ MostlyLayers instruments fully
[accent]Emphasize beat/syllable✅ MostlyProduces rhythmic emphasis
[repeat]Repeat a word/phrase✅ MostlyRhythmic repetition
[up]Rising pitch/energy✅ MostlySequential rising melodic effect
[down]Falling pitch/energy✅ MostlySequential falling melodic effect
[hold]Sustain note/word✅ MostlyLengthens syllables
[nep]Experimental short vocal sound✅ PartiallyMicro-vocal rhythm
[8/8 melody]Melody time signature cue✅ MostlyRhythmic guidance, less precise than BPM
[breakbeat]Drum pattern cue✅ YesGenerates breakbeat rhythm
[ambient pad]Background sustained sound✅ YesPads or atmospheric layers
[festival acid vibe]EDM style cue✅ MostlyGuides overall style/energy
[mono]Mono mix✅ MostlyCenters audio; affects panning
[stereo spread]Widen stereo field✅ PartiallyWorks if not conflicting with [mono]
 
Holy crap that's actually fking amazing.

If The Smashing Pumpkins released this it would be a chart topper.

You should DM them. 😅
I like that song but people don't seem to like it too much.


Freaks is my favorite from that session. It is super catchy and unsettling

Gutter glory is the other song in that session, less weird, steady beat. Nice vocal range. Weirdly enough I asked asked chatgpt to consider the songs that will be the lead single and picked this one, it actually predicted which one would get the most airplay

Did a second session and it picked this one as the lead single

Rugged pull under

But this is the better one, Shattered hearts club

Musical taste is varied to what people like, for some or other reason this has been getting steady airplay the last few days, I don't really consider it that great of song. the lyrics is pretty weak, the vocals are great, just when you think you have a decent song, nope you get stabbed in the back, it says nope we going to pick this one. lol

 
Something forgot to post about this, afrikaans has gotten a lot better with v5.5 so I have been going and checking my library and been converting some songs to afrikaans


Same melody just shifted it more to rock EDM which works very well with afrikaans, it is quite a common style for south african music.


The original which sound the same but lacks that EDM rock vibe
 
Last edited:
I disagree, you can pretty much make anything you you want. There is somethings you can't, but that isn't due to the training model, more you need know the prompts and tags.

This was a test song used a 8 seond monophonic clip from the old nokia ringtone area and pushed a monophonic arrangement.

This was a pitch experiment each word had a pitch difference +2 for each word for high pitch and -2 word each pitch for low pitch

Another experiment that intentionally breaks melody structure


Your song has a mono quality to it, just something I threw together I left the tags, so you can see how I approach out there concepts.

Suno has a bit of an issue with mono sometimes, it didn't follow it directly and only partial. But to get that effect is just to download the stems and force whatever I need into the mono channel


5.5v mostly followed the mono

5.5v more abstract prompt
The second one introduces uneven bass

Also discovered that if you mention bass synth with uneven double bass effect it selects a generic bass synth, but if you mention a particular bass synth as well as the actual bass range you can control way much more.

If you give it a generic prompt and no lyrical tags it will always default to standard "safe" melody sounds. It isn't that you can't do less quite the opposite you can do way way more, to the point you actually need to control

I resorted to version 4.5v it entirely ignored everything here.


So yeah you need to be HIGHLY specific with the prompts and tag the **** out of the lyrics with approved tags

Just some of the tags I discovered for 5.5 and these are all lyrical tags

[verse]Marks verse section✅ YesStructures vocal and instrumental arrangement
[chorus]Marks chorus section✅ YesUsually fuller instrumentation & stronger vocals
[bridge]Marks bridge/transition✅ YesGuides dynamic shift or texture change
[break]Indicates a break✅ MostlyCreates pause or energy drop
[intro]Beginning section✅ YesSets up track
[outro]Ending section✅ YesFade out or simplified section
[looped]Repeat lyric/motif✅ MostlyCreates rhythmic or melodic repetition
[stutter]Repeated/broken syllable✅ MostlyChopped vocal effect
[short burst]Clipped phrasing✅ MostlySharp, rhythmic delivery
[breathy]Airy vocal style✅ YesSoft, intimate vocal delivery
[whisper]Whispered vocal line✅ YesProduces soft, hushed delivery
[whistle]Whistling line✅ MostlyMicro-melodic effect
[whistling letters]1–3 letter whistle sounds✅ PartiallyShort melodic vocalizations
[pitch +1/+2/+3/-1/-2/-3]Pitch shift instructions✅ YesChanges line pitch
[micro pitch shift]Tiny per-word pitch changes✅ MostlySubtle melodic variation
[echoed]Reverb/delay effect✅ YesProduces repeated echo effect
[echo]Same as echoed✅ YesSynonym, works identically
[delayed]Delay effect✅ YesRhythmic echo on vocals or chops
[reverb]Adds spatial effect✅ YesCreates room/space on vocals/instruments
[processed]General processing cue✅ MostlyFilters, distortion, reverb implied
[filtered drums]Drum EQ/filtering✅ YesAlters drum texture
[double bass]Heavy bass line✅ YesEmphasized low end
[chopped synth]Staccato/glitch synth✅ YesCut-up synth patterns
[sparse]Minimal arrangement✅ MostlySparse textures, reduced instruments
[full]Dense arrangement✅ MostlyLayers instruments fully
[accent]Emphasize beat/syllable✅ MostlyProduces rhythmic emphasis
[repeat]Repeat a word/phrase✅ MostlyRhythmic repetition
[up]Rising pitch/energy✅ MostlySequential rising melodic effect
[down]Falling pitch/energy✅ MostlySequential falling melodic effect
[hold]Sustain note/word✅ MostlyLengthens syllables
[nep]Experimental short vocal sound✅ PartiallyMicro-vocal rhythm
[8/8 melody]Melody time signature cue✅ MostlyRhythmic guidance, less precise than BPM
[breakbeat]Drum pattern cue✅ YesGenerates breakbeat rhythm
[ambient pad]Background sustained sound✅ YesPads or atmospheric layers
[festival acid vibe]EDM style cue✅ MostlyGuides overall style/energy
[mono]Mono mix✅ MostlyCenters audio; affects panning
[stereo spread]Widen stereo field✅ PartiallyWorks if not conflicting with [mono]
Super useful info and definitely going to use, since that is the best way to get the most out of 5.5.

However , I'm f telling you, this is not a good development.
04890.jpg

the training data being Time Warnered. Like Disney destroyed Star Wars, Time Warner will destroy Suno.
However, even as it stands, still an amazing tool. My gripe is just the general direction.
Just everything lately, would've been cool to just get to keep one thing that doesn't become contaminated by
you know who sigh
 
Super useful info and definitely going to use, since that is the best way to get the most out of 5.5.

However , I'm f telling you, this is not a good development.
View attachment 1899663

the training data being Time Warnered. Like Disney destroyed Star Wars, Time Warner will destroy Suno.
However, even as it stands, still an amazing tool. My gripe is just the general direction.
Just everything lately, would've been cool to just get to keep one thing that doesn't become contaminated by
you know who sigh

Lol I hear you. Yeah you have to put in a lot more effort now. But it is a good thing it will hopefully push out the slop.

yeah I am not posting my tags, fck em. I had to spend lots of credits and lots of experimenting also setting up chatgpt to my tastes and tags and descriptions, considering removing lyrics as well. There is nothing stopping users taking lyrics and twisting them and reusing them.

For chapgpt I had to setup the afrikaans as will, and setup specific rules like not over using the double negative and pass tense, use modern afrikaans and not to stray into arcane words or dutch or avoid afrikaans/english words that has the same spelling and or different meaning, as it doesn't understand context often, even fed it the language rules from wiki.

Ended up getting better results because of it, so yeah it take some effort.

Don't mind sharing stuff I discovered here, it is my secret sauce and identity, suno users will copy that without issue. I wish there was a way to hide descriptions as well.
 
Something forgot to post about this, afrikaans has gotten a lot better with v5.5 so I have been going checking my library and been converting some songs to afrikaans


Same melody just shifted it more to rock EDM which works very well with afrikaans, it is quite a common style for south african music.


The original which sound the same but lacks that EDM rock vibe
Yeah had a listen to some of them earlier. Afrikaans sounds Afrikaans. Baie cool )


I tried to cover Kobus Van Der Vyver a while back but got too lazy to fix the lyrics so just made it instrumental, but definitely needs the vocals.

such a catchy song

 
Yeah had a listen to some of them earlier. Afrikaans sounds Afrikaans. Baie cool )


I tried to cover Kobus Van Der Vyver a while back but got too lazy to fix the lyrics so just made it instrumental, but definitely needs the vocals.

such a catchy song

I am happy I can now finally push more afrikaans, ironically suno doesn't know what South african boere musiek is, nearly fell of my chair when it tried kept on giving you folk polka....

It is hilarious

But it is doing a much better job now AI klipwerf boere orkes lol



I definitely want to experiment with a modern take on and introduce trekklavier into a custom afrikaans EDM rock song, still thinking about how it will go and thinking about the lyrics that makes sense.


Also going to use this track now I used example it has a nice beat, I like the ethereal style might work well for a welsh language song, even old english will work well here.
 
Another thing that has happened is straight up copyright infringement, even intentionally using snippets of songs.

Please don't listen to it, don't give it any more airtime. But it is exactly like the original, just like the original just remixed without vocals. I stay well and clear of copyrights, I don't even try and mimic it, I don't even touch lyrics or mimic it. I might like a particular synth and would use the synth rather , it is an instrument after all

been reported, seems their report system is bugged as well



 
I am happy I can now finally push more afrikaans, ironically suno doesn't know what South african boere musiek is, nearly fell of my chair when it tried kept on giving you folk polka....

It is hilarious

But it is doing a much better job now AI klipwerf boere orkes lol



I definitely want to experiment with a modern take on and introduce trekklavier into a custom afrikaans EDM rock song, still thinking about how it will go and thinking about the lyrics that makes sense.


Also going to use this track now I used example it has a nice beat, I like the ethereal style might work well for a welsh language song, even old english will work well here.
Dude, if you can hack the right combination, can actually make some $. Afrikaans massive loyal fanbase. Some of those artists get millions of listens.

Oh , and again, thanks for the tag list. Already having better results. Even if it means I have to add some other parts manually later on.
 
Dude, if you can hack the right combination, can actually make some $. Afrikaans massive loyal fanbase. Some of those artists get millions of listens.

Oh , and again, thanks for the tag list. Already having better results. Even if it means I have to add some other parts manually later on.
I will be happy as long as it doesn't sound like kurt tappit darren, but likely will need to delve deep into my kempton park zef side and make some kurt darren bangers it seems.

yeah just don't go sharing it with the entire world our dirty little secret. I did tell you to ask me if you are stuck lol. Didn't listen ne......as jy dom is moet suffer.....lol joke of course. Seriously if you need something or just want to bounce ideas give me a holler, a second opinion is always good, pretty lonely thread with basically just the 3 of us actively making and posting stuff

I going to keep on knocking out afrikaans tracks and curate the best of them and take it from there. already roped in some people to listen and pick their favorites.
 
So I made a Suno Music style descriptor custom GPT a while back, and recently updated it for Suno 5.5, using both ChatGPT and Gemini's Deep Research capabilities to look up and compile all the known documentation into the GPT knowledge base. It can be used either by telling the GPT, I would like to create a track that sounds like "song x" by "artist z", in which case the GPT will give you the simple and verbose descriptions for Suno 5.5 to replicate that style, and if you give it the lyrics to your song it will rewrite it, slightly tweaking it for flow as well as tagging it with the new 5.5 "structure" tags used in the lyrics box for finer control.

Edit : I also forgot to mention - it doesn't only replicate the style of a particular song or band, but you can also tell it, as an example - "I want a reggae track with orchestral backing and a choir chanting", and it will give you the description and tags that tends to give you what you're looking for. You can also tell it, to reimagine certain songs, or mash up genres and so on and so forth too.

I've made the GPT accessible via the GPT store or direct link : https://chatgpt.com/g/g-DRYjAV3Xg-music-style-descriptor
 
Last edited:
Top
Sign up to the MyBroadband newsletter
X